feat(C219): paridade de busca e filtros nas gravações do acervo (Closes #1295)

Classificação automática com proveniência (job + CLI recordings:classify), facetas ano/tema/alcance/município/duração na URL canônica, busca por tema no corpus próprio, ordenação e a barra de filtros portada do design aprovado sobre os shells compartilhados. Deep-links existentes byte-idênticos; Câmara intocada.

docs/changelog/2026-09-24-c219.md

@@ -0,0 +1 @@
+**C219 — Paridade de busca e filtros entre as fontes do acervo (2026-09-24):** "Gravações enviadas" passa a ter os mesmos gestos de garimpo das "Falas da Câmara" — busca `termo|tema`, filtros de Ano, Tema, Alcance, Município citado e Duração (Pessoa segue exclusiva), chips ativos, contagem, ordenação e vazio honesto — com classificação automática e proveniência auditável sobre a transcrição já existente. Schema: `recording` ganha `year` derivado de `recordedAt` (hook espelho do `Speech`, que agora **limpa** o ano quando a data é limpa), `topics`/`scopes` (mesma taxonomia `lib/speechFacets`), `classifiedBy` (`gazetteer|llm|manual`, **nullable** — nulo = "não classificada", nunca proveniência inventada) e `mentionedMunicipalities` (relationship), com `preserveManualFacets` protegendo curadoria humana de escritas sem `req.user` (job/CLI); migration aditiva `20260924_040210_add_recording_facets` com backfill SQL determinístico do ano. Classificação: `classifyRecordingFacets` (novo, `server-only`) reusa o dono `classifySpeech` (gazetteer offline + refinamento LLM validado na DeepInfra, nunca lança) e resolve os slugs do gazetteer no catálogo de municípios; o seam é o fim do `runRecordingJob` (injetável, fora da transação de save — o LLM não entra no lock) e o acervo existente é coberto pelo CLI idempotente `pnpm recordings:classify` (`--dry-run`, `--limit`, `--all`, guard de DB local + `RECORDING_CLASSIFY_CONFIRM=1` para alvo remoto), com o shape de escrita em dono único (`recordingFacetWriteData`). URL/filtros: `recordingListUrl` ganha `mode|year|topic|scope|municipality|duration|sort` na ordem canônica da Câmara, com `sort=recentes`/`mode=termo` nunca serializados e os deep-links `?source=enviadas&q&person&page` byte-idênticos; `buildRecordingListWhere` ANDa as facetas (duração reusa `durationBucketWhere` exportado do dono da Câmara) e, nos sorts de duração, exige `durationSeconds` existir (senão o NULLS FIRST do Postgres abriria a lista com as gravações sem duração); o loader ganha opções de ano/municípios, injeção da expansão de tema e a degradação honesta (`themeUnavailable`/`themeApplied`); `expandSpeechSearchTheme` ganha o corpus `'recording'` (3º consumidor do mesmo mecanismo, prompt próprio) e o adaptador puro `recordingOmnibox` monta chips/seeds/toggles sobre `lib/campaignListOmnibox`. UI (Impeccable B, porte do design aprovado `docs/plans/acervo-paridade-busca-filtros-ui-design.html`): `RecordingAcervoFilters` novo sobre os shells compartilhados (omnibox com chips dentro do campo, seletor `Termo exato|Por tema`, seis facetas em popovers-chip, absorvendo o antigo `RecordingPersonFilter`), `RecordingSortSelect` ("Mais recentes" default · "Duração (maior)" · "Duração (menor)" com o hint das gravações sem duração), `RecordingThemeFallbackNotice` + `CampaignThemeRetryButton` (movido para `shared/`), chip de tema no card e o vazio honesto do modo tema com "Limpar filtros" + "Trocar para termo exato"; no /simplify o seletor de modo e o rótulo de chip viraram donos compartilhados (`CampaignSearchModeControl`, `selectedFilterTriggerLabel`). Access/Consent intocados (gate `communicationCatalog` fail-closed, sem PII nova, sem rota/collection nova). Verificação: unit (URL canônica/deep-link/ordenação, `where` das facetas+tema+gate de duração, omnibox, corpus novo), int (job grava facetas e degrada sem chave, curadoria `manual` preservada inclusive no escape autenticado, ano derivado/limpo, loader filtra/ordena/degrada o tema), e2e do `campaignSpeechAcervo` estendido (barra de facetas, estreitamento + vazio, ordenação por duração, tema degradando) e `campaignSpeechAcervo` no conjunto curado high-risk do PR (a migration torna a entrega high-risk), `pnpm gate:fast`, `pnpm test:int`, `pnpm build` e os e2e locais. Crítica final do `designer` (trigger c) **certificada em tier primário** (`Design tier: openai/gpt-5.6-sol`); o débito do "X" de limpar do omnibox que quebra linha no mobile com chip longo (pré-existente do shell compartilhado, reproduzido na Câmara) ficou fora e virou a Issue C221, com os demais adiamentos anotados com gatilho no `*-impl.md`.

docs/plans/acervo-paridade-busca-filtros-impl.md

@@ -0,0 +1,209 @@
+# Impl: C219 — Paridade de busca e filtros entre as fontes do acervo
+
+Status: aprovado
+Atualizado em: 2026-09-24
+Issue: #1295
+Intenção: docs/plans/acervo-paridade-busca-filtros.md
+Appetite restante: herdado (~2–3 dias eng)
+
+## Leitura da intenção
+
+- **Outcome:** "Gravações enviadas" passa a oferecer os mesmos gestos de garimpo da Câmara — busca `termo|tema`, filtros por Ano, Tema, Alcance, Município citado, Duração (e Pessoa, que segue exclusiva), chips ativos, contagem, ordenação e vazio honesto — com classificação automática e proveniência auditável sobre a transcrição já existente, sem tocar no player/download nem no contrato da Câmara.
+- **O que NÃO negociar:**
+  - "Pessoa" continua **exclusiva** das gravações (rótulo derivado, nunca segundo cadastro de pessoa — C200).
+  - Gate do acervo **fail-closed** (`requireCampaignPageActor({gate:'communicationCatalog'})` + `canReadRecording`/`canReadCommunicationCatalog`); `advisor`/`leader` negados.
+  - Deep-link atual **byte-compatível**: `?source=enviadas&q&person&page` continua canonicalizando para exatamente o mesmo href; `mode=termo` e o sort default nunca serializados.
+  - **Sem lista única** entre fontes; **sem redesenho da barra da Câmara** (ela é a referência).
+  - Classificação **automática e revisável, nunca inventa** — falha degrada para o que existir e é sinalizada; proveniência é rótulo auditável (`gazetteer|llm|manual`), nunca score.
+  - Buckets de duração idênticos à Câmara ("Até 2 min" · "2 a 5 min" · "Mais de 5 min" · "Sem duração"); **Ano derivado de `recordedAt`**, nunca um segundo campo digitado.
+- **O que reavaliar (hipóteses da intenção):**
+  - "Direção no codebase" aponta `speechClassifier` e `expandSpeechSearchTheme` corretamente, mas o seam real do job é `runRecordingJob` (roda via `after()`, sem request) — a classificação entra ali, **fora** da transação de save, e o backfill das linhas existentes é CLI.
+  - `classifySpeech` é reusável como está, mas pede `transcript` **cru** (o `searchText` da gravação é normalizado; o input tem de ser a concatenação dos `text` dos segmentos).
+  - `expandSpeechSearchTheme` já tem o parâmetro `corpus` desde S28 — a extensão é **um prompt novo**, não mecanismo novo.
+  - Gravações **não têm `keywords`**: a perna textual delas é só `searchText` (a Câmara tem `searchText` OR `keywords`).
+  - `Recording` **não tem** os campos de faceta/proveniência de `Speech`: schema novo + migration são inevitáveis (a intenção já previa "facets + proveniência").
+
+## Abordagem recomendada
+
+```mermaid
+flowchart LR
+  UP[upload / retry] --> JOB[runRecordingJob]
+  JOB --> ASR[extrair + transcrever + diarizar]
+  ASR --> CLS{classifyRecordingFacets<br/>classify injetável}
+  CLS -- "sem DEEPINFRA_API_KEY / erro" --> GZ[gazetteer: topics/scopes/municípios]
+  CLS -- ok --> LLM[gazetteer ∪ LLM validado]
+  GZ --> TX[(transação: segments + recording)]
+  LLM --> TX
+  TX --> RDY[ready + facets + classifiedBy]
+
+  CLI["scripts/backfill-recording-facets.mjs<br/>pnpm recordings:classify"] -. mesma função .-> CLS
+
+  Q["assessoria: ?source=enviadas&q&mode&year&topic&scope&municipality&duration&sort&person&page"] --> URL[recordingListUrl<br/>parse + canônico]
+  URL --> MODE{mode=tema?}
+  MODE -- não --> WHERE[buildRecordingListWhere<br/>facetas AND + texto OR + pessoa]
+  MODE -- sim --> EXP[expandRecordingSearchTheme<br/>corpus 'recording']
+  EXP -- null --> DEG[literal + aviso discreto]
+  EXP -- termos --> WHERE
+  DEG --> WHERE
+  WHERE --> FIND["payload.find<br/>user · overrideAccess:false"]
+  FIND --> VM[recordingViewModels<br/>excerto + destaque + chips]
+  VM --> UI[RecordingAcervoFilters + lista + contagem + ordenação]
+```
+
+**Opções consideradas | Recomendação | Rejeitadas (decisões caras):**
+
+| #   | Decisão                       | Opções                                                                                                                                                                                                                                                                                                                           | Recomendação                                                                                                                                                   | Rejeitadas                                                                                                                                                                                                                                            |
+| --- | ----------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| i   | Onde a classificação roda     | A) seam no `runRecordingJob` (injetável) + CLI backfill · B) hook `afterChange` da collection · C) on-demand no loader da busca · D) hook por `recordingSegment`                                                                                                                                                                 | **A** — o job já tem os segmentos em memória no fim da transcrição; o seam injetável é precedente (`catalogContentPiece`); o CLI cobre as linhas já existentes | B (sem transcrição no momento do hook, roda em caminho quente e re-executa por `step`); C (LLM por request, leitura viraria escrita, faceta inexistente não filtra); D (N chamadas por gravação, unidade errada)                                      |
+| ii  | Schema em `Recording`         | A) espelhar `Speech` (`year` derivado + `topics`/`scopes`/`classifiedBy`/`mentionedMunicipalities` + `preserveManualFacets`) com migration `add_recording_facets` · B) facetas por segmento · C) sem schema (derivar na busca) · D) ano sem coluna (range sobre `recordedAt`)                                                    | **A** — mesma forma de `where`/opções/canônico da Câmara e curadoria manual preservada; a migration é aditiva                                                  | B (unidade errada — corte explícito da intenção); C (não filtra/pagina, recalcula a cada busca e morre o "revisável"); D (dois shapes de filtro para o mesmo conceito; a coluna derivada é o que a Câmara tem e o que a intenção chama de "derivado") |
+| iii | Contrato de duração + omnibox | A) reusar donos da Câmara (buckets/labels/tipo de `speechListUrl`, `durationBucketWhere` exportado de `speechListFilters`, parsers de ano/município movidos para `campaignListUrl`) + adaptador novo `recordingOmnibox` · B) extrair módulo compartilhado novo do acervo e migrar a Câmara · C) duplicar no domínio de gravações | **A** — a paridade manda que o bucket tenha UM dono; o adaptador por domínio já é padrão (`municipalityOmnibox`, `municipalityUpdateOmnibox`)                  | B (churn na Câmara sem volatilidade; extrair quando C216 chegar — 3º consumidor); C (twin: mudança de bucket divergiria as duas fontes)                                                                                                               |
+| iv  | Busca por tema                | A) `ThemeSearchCorpus` ganha `'recording'` + prompt próprio + `expandRecordingSearchTheme` · B) reusar o corpus `'speech'` · C) só busca literal                                                                                                                                                                                 | **A** — o mecanismo é um só (S28); o prompt de discursos da Câmara não descreve plenárias enviadas/debates                                                     | B (prompt errado degrada a expansão); C (rejeitada no gate: a ponte foi confirmada)                                                                                                                                                                   |
+| v   | UI                            | A) `RecordingAcervoFilters` novo sobre os shells compartilhados (absorve o `RecordingPersonFilter`) · B) generalizar `SpeechAcervoFilters` configurável · C) estender `SpeechAcervoFilters` com `source`                                                                                                                         | **A** — shells compartilhados são os donos das mecânicas; a Câmara não é tocada                                                                                | B (API genérica rasa e mexe na Câmara — anti-goal); C (acopla os dois contratos num componente)                                                                                                                                                       |
+| vi  | Ordenação                     | A) param `sort` com valores pt-BR (`recentes` default omitido, `duracao_maior`, `duracao_menor`); sort de duração só sobre linhas com duração medida · B) par genérico `sort`/`dir` (C117) · C) sort só de UI                                                                                                                    | **A** — contrato fechado e legível no deep link; default preserva `-createdAt` (zero regressão)                                                                | B (o acervo não é tabela; a Câmara não tem esse contrato); C (quebra "paginação preservada no deep link" e back/forward)                                                                                                                              |
+
+**Recomendação (geral):** uma única passada de classificação no job (com o mesmo `classifySpeech`, injetável e com degradação para gazetteer), um CLI idempotente para o acervo já enviado, e a paridade de filtros construída sobre o contrato de URL, o `where`, o adaptador de omnibox e os shells que a Câmara já usa — mudando **só** a fonte de gravações.
+
+**Rejeitadas (geral):** busca global/lista única (anti-goal); pgvector/embeddings (C192 já rejeitou); facetas por trecho/falante (corte da intenção); qualquer alteração na barra/contrato da Câmara.
+
+### Componentes / mudanças
+
+- **`Recording`** (`src/collections/Recording.ts`): novos campos `year` (number, index, readOnly, derivado), `topics` (select hasMany, index, opções `SPEECH_TOPICS`), `scopes` (select hasMany, index, opções `SPEECH_SCOPES`), `classifiedBy` (select, index, **opcional, sem default**, valores `SPEECH_CLASSIFICATION_SOURCES`), `mentionedMunicipalities` (relationship hasMany → `municipality`, index). Hooks: `deriveRecordingYear` (beforeValidate, espelho de `deriveSpeechYear` — `recordedAt` string `slice(0,4)`, com fallback `originalDoc`) e `preserveManualFacets` (beforeChange, espelho de `Speech.ts:43-57`, `FACET_FIELDS = topics|scopes|classifiedBy|mentionedMunicipalities`; escrita sem `req.user` — job/CLI — nunca sobrescreve `manual`). `admin.defaultColumns` ganha `topics` e `classifiedBy` (superfície de revisão da classificação no admin). `admin.description` registra que a classificação é automática sobre o início da transcrição.
+- **Migration** (`pnpm migrate:create add_recording_facets` → `src/migrations/20260924_120000_add_recording_facets.ts` + `.json` + `index.ts`): tabelas de join `recording_topics`/`recording_scopes` (enums `enum_recording_topics`/`enum_recording_scopes`/`enum_recording_classified_by`), tabela `recording_rels` (`mentionedMunicipalities`), colunas `year`/`classified_by`. **Hand-edit** (precedente `20260913_150000_backfill_speech_search_text_trgm_index`): backfill determinístico de `year` a partir de `recorded_at` (`WHERE recorded_at IS NOT NULL`). Nada destrutivo; `recording_search_text_trgm_idx` já existe (não mexer).
+- **`src/payload-types.ts`**: regenerado (`pnpm generate:types`) — `Recording` ganha os campos.
+- **`classifyRecordingFacets`** (`src/utilities/recordings/recordingClassification.ts`, novo, `import 'server-only'`): `({ payload, transcript, classify = classifySpeech })` → `{ topics, scopes, classifiedBy, mentionedMunicipalities } | null`. Reusa `SpeechFacetInput`/`SpeechClassificationResult`, `classifySpeech` (DeepInfra + validação `lib/speechFacets`) e resolve `municipalities[].slug → municipality.id` numa query (bypass justificado — geografia read-only — precedente `speechImport.ts:176-196`). Nunca lança: transcrição vazia ou throw defensivo devolvem `null` e o job segue sem facets.
+- **`runRecordingJob`** (`src/utilities/recordings/recordingJob.ts`): ganha `classify` injetável (default `classifySpeech`), chamado **depois** de diarizar e **antes** de `markStep('saving')` (o LLM não entra na transação); transcript = `keyedSegments.map(s => s.text).join(' ')`. O update transacional final (`status:'ready'`, `searchText`, `speakerLabels`) passa a incluir `year` (quando houver `recordedAt`), `topics`, `scopes`, `classifiedBy`, `mentionedMunicipalities`.
+- **`scripts/backfill-recording-facets.mjs`** (novo) + `package.json` `"recordings:classify"`: reusa `classifyRecordingFacets`; default = linhas `ready` com `classifiedBy` ausente (`--all` reclassifica não-`manual`), `--dry-run`, `--limit N`; `assertLocalDatabase` + `RECORDING_CLASSIFY_CONFIRM=1` para alvo remoto/produção (precedente `CAMARA_IMPORT_CONFIRM`); relatório (classificadas/puladas-manual/falhas/tokens/custo) e idempotência por `classifiedBy`.
+- **URL** (`src/utilities/recordings/recordingListUrl.ts`): `RecordingListState` ganha `mode?` (reusa `SpeechSearchMode`), `years?`, `topics?`, `scopes?`, `municipalities?`, `durations?` (reusa `SpeechDurationBucket`) e `sort?`; `recordingListParamNames` ganha `mode|year|topic|scope|municipality|duration|sort`. Ordem canônica: `source, q, mode, year[], topic[], scope[], municipality[], duration[], sort, person[], page` — os parâmetros novos não perturbam o href dos deep links atuais. Serialização: `mode=tema` só com `q`; `sort` só quando ≠ `recentes`; `page` só > 1. Novos: `RECORDING_SORT_OPTIONS`/`recordingSortLabels` ("Mais recentes" · "Duração (maior)" · "Duração (menor)"), `recordingSortKeys(state)` (`-createdAt` default / `-durationSeconds` / `durationSeconds`) e `buildRecordingFiltersKey(state)` (espelho de `buildSpeechFiltersKey`, usado como `key` do remount da barra).
+- **`campaignListUrl.ts`**: recebe `parseYearValues`/`parseMunicipalityValues` (movidos de `speechListUrl.ts`, que passa a importá-los) — um dono para a validação estrutural de ano/município nas duas fontes.
+- **`speechListFilters.ts`**: exporta `durationBucketWhere` (sem mudança de comportamento) para o `where` das gravações.
+- **Filtros** (`src/utilities/recordings/recordingListFilters.ts`): `buildRecordingListWhere(state, themeTerms = [])` — facetas AND (`year in`, `topics in`, `scopes in`, `mentionedMunicipalities in`, OR de `durationBucketWhere`, OR de `speakerNames contains`), perna textual OR de `{ searchText: { like: normalizeForSearch(term) } }` para `[q, ...themeTerms]` (sem `keywords`), e `{ durationSeconds: { exists: true } }` quando `sort` é de duração (o Postgres ordena NULLS FIRST em DESC; sem isso "Duração (maior)" abriria com as linhas sem duração).
+- **Loader** (`src/utilities/recordings/recordingPageData.ts`): `RecordingFilterOptions` ganha `years` e `municipalities` (via `loadMunicipalityLabelsByIds`, precedente `loadSpeechFilterOptions`); `loadRecordingsPageData(payload, user, params, expandTheme = expandRecordingSearchTheme)` com o gate de tema (`mode=tema && q && canReadCommunicationCatalog(role)` antes de qualquer chamada externa) e retorno `themeUnavailable`/`themeApplied`; `sort` aplicado; `loadMatchedSegments` passa a receber os termos (`[q, ...themeTerms]`) e casa `or` de `searchText like`; `select` da lista ganha `topics` (chip do card).
+- **View models** (`src/utilities/recordings/recordingViewModels.ts`): `toRecordingListItemViewModel` ganha `themeTerms`; o excerto prefere a evidência de tema (espelho C192: primeiro termo que casa em `searchText`, segmento via `pickMatchingSegment`, `buildHighlightedExcerpt(..., { phrase: true })`) e cai no literal quando não há; `watchHref` mantém `?t=&q=` com o **`q` literal** (byte-compat; espelho de `buildWatchHref`); `RecordingListItemViewModel` ganha `topics: { value, label }[]` (labels de `speechTopicLabels`).
+- **Omnibox** (`src/utilities/recordings/recordingOmnibox.ts`, novo, puro/client-safe): chips (`q`, `year`, `topic`, `scope`, `municipality`, `duration`, `person`), seeds (Tema/Alcance/Duração visíveis com query vazia; Ano/Município citado/Pessoa pesquisáveis por palavra-chave), `applyRecordingOmniboxSuggestion`/`removeRecordingOmniboxChip`/`applyRecordingSearchMode`/`clearRecordingOmnibox` (preservando `source: 'enviadas'` no clear) sobre `lib/campaignListOmnibox.ts`; `person` reusa `toggleRecordingPerson`/`removeRecordingPerson` (dedupe case-insensitive já dono) e o prefixo `person:` lê o valor por `slice` (o rótulo pode conter `:`).
+- **Tema** (`src/utilities/ai/expandSpeechSearchTheme.ts`): `ThemeSearchCorpus` ganha `'recording'`; novo prompt de corpus (plenárias/debates/material próprio enviados pela equipe), `themeExpansionRules('gravação')`; wrapper `expandRecordingSearchTheme` (mesmo timeout/degradação/`normalizeSpeechThemeTerms`).
+- **UI (Impeccable B — port de `docs/plans/acervo-paridade-busca-filtros-ui-design.html`):**
+  - `src/components/campaign/recording/RecordingAcervoFilters.tsx` (novo, client) — **cenas 01/02**: form `campaignListOmniboxFormClassName` com label "Buscar nas gravações enviadas", `CampaignListOmnibox` (chips dentro do campo, como na Câmara — B184/B127), seletor segmentado `role="group"` `grid grid-cols-2 rounded-lg bg-muted p-1` + `aria-pressed` + `min-h-11 md:min-h-9` (com o estado degradado `themeUnavailable`), e as seis facetas como `CampaignHeaderFilterPopover triggerVariant="chip"` (`Ano`, `Tema`, `Alcance`, `Município citado`, `Duração`, `Pessoa`) com `selectedTriggerLabel` dobrando a seleção. A linha "Ativos:" do artefato é realizada pelos chips do próprio omnibox (shell compartilhado), sem segunda fileira.
+  - `src/components/campaign/recording/RecordingSortSelect.tsx` (novo, client) — **cena 01**: label "Ordenar por" + `select` nativo (`rounded-md border border-input`) via `useCampaignListFilterNavigation`; em sort de duração, hint "Gravações sem duração aparecem apenas em Mais recentes." A contagem fica no `CampaignListFooter` (padrão de todas as listas — sem duplicar a linha).
+  - `src/components/campaign/recording/RecordingThemeFallbackNotice.tsx` (novo) — **cena 04/C192**: `Alert variant="pending"` com a copy de gravações + `CampaignThemeRetryButton`; move `SpeechThemeRetryButton` → `src/components/campaign/shared/CampaignThemeRetryButton.tsx` (componente genérico, zero copy própria) e atualiza o import da página da Câmara.
+  - `src/components/campaign/recording/RecordingResultCard.tsx` — **cena 01 (card)**: chip de tema (`bg-muted`) ao lado do chip de pessoa; destaque do termo continua via `SpeechHighlightParts`.
+  - `src/app/(campaign)/campanha/(app)/comunicacao/acervo/page.tsx` — branch `enviadas`: `key={buildRecordingFiltersKey(data.state)}`, barra nova, aviso de tema, `RecordingSortSelect`, `RecordingStatusRefresher` (intocado), lista/vazio/footer; vazio honesto da **cena 03** (título "Nenhuma gravação encontrada para este tema", descrição do artefato, "Limpar filtros" + "Trocar para termo exato") e `hasFilters` por `buildRecordingFiltersKey`. O `RecordingsSearchForm` inline e o `RecordingPersonFilter` somem (absorvidos).
+  - **Cena 04 (sem permissão)** permanece o redirect fail-closed do gate — a cena é ilustrativa, não vira superfície nova. **Cena 05 (carregando)** é o `CampaignListPendingBoundary`/`CampaignListResults` existente (dima o resultado, não o controle).
+  - Remover `src/components/campaign/recording/RecordingPersonFilter.tsx`.
+- **Manifest e2e** (`scripts/lib/e2e-affected-manifest.mjs`): `src/collections/Recording.ts` entra no entry da vertical C154/C199 (`campaignSpeechAcervo`) — hoje um diff só nesse arquivo cairia no smoke da home; e `campaignSpeechAcervo` entra em `E2E_CURATED_SPECS` (precedente C167/S19: a migration torna todo PR da entrega high-risk), com o pin de congelamento atualizado em `tests/unit/e2eAffectedManifest.unit.spec.ts`.
+- **Access / Consent:** **nada muda**. Sem chave `Consent` nova (acervo interno; sem PII nova), sem collection nova, sem rota nova. O classificador fala com a DeepInfra (mesmo provedor/chave que já transcreve a gravação) e o tema com a DeepSeek (só o texto digitado, como na Câmara).
+
+### Dados → forma (se aplicável)
+
+N/A justificado. O item não apresenta métrica, agregado nem score: resultados e facetas são recorte editorial. A proveniência é rótulo auditável (`Gazetteer`/`LLM`/`Manual`) no admin — nunca número na UI; "contagem de resultados" é paginação, não métrica de vaidade.
+
+## Fases verificáveis
+
+1. **Tracer / schema + server** (~1 dia) — `Recording` com os campos/hooks + migration `add_recording_facets` (com backfill de `year`) + `pnpm generate:types` + `recordingClassification` + seam `classify` no `runRecordingJob` + `scripts/backfill-recording-facets.mjs`.
+   - Prova: `tests/int/recording.int.spec.ts` — job end-to-end grava `topics`/`scopes`/`classifiedBy`/`mentionedMunicipalities`/`year` (classificador injetado determinístico + caminho gazetteer sem `DEEPINFRA_API_KEY`); curadoria `manual` não é sobrescrita por escrita sem `user`; retry reclassifica; `pnpm migrate` local verde.
+2. **URL / filtros / loader** (~0,5 dia) — params e estado novos, `buildRecordingFiltersKey`, `recordingSortKeys`, `where` com facetas + tema + gate de sort, loader com opções (`years`/`municipalities`) e injeção do tema, corpus `'recording'`.
+   - Prova: `tests/unit/recordingListUrl.unit.spec.ts` (deep-link byte-compat, canônico, enums exaustivos, `sort` default omitido), `tests/unit/recordingListFilters.unit.spec.ts` (facetas AND, tema OR, duração, `exists` do sort), `tests/unit/recordingOmnibox.unit.spec.ts` (novo), `tests/unit/expandSpeechSearchTheme.unit.spec.ts` (corpus novo), `tests/int/recording.int.spec.ts` (loader filtra/ordena/tema e opções).
+3. **UI port** (~0,5–1 dia) — `RecordingAcervoFilters`, `RecordingSortSelect`, `RecordingThemeFallbackNotice`, `CampaignThemeRetryButton` movido, chip de tema no card, página; remoção do `RecordingPersonFilter`.
+   - Prova: `tests/e2e/campaignSpeechAcervo.e2e.spec.ts` (describe de gravações estendido: barra com chips, filtro que estreita + vazio honesto, modo tema degradando, sort, contagem; o teste de "Pessoa" passa a exercitar a barra nova); `pnpm dev` manual com a cena 01 do artefato.
+4. **Testes/gates** (~0,5 dia) — `pnpm gate:fast`; depois a cascata cheia (`pnpm lint`, `pnpm format:check`, `tsc --noEmit`, `pnpm exec knip`, `pnpm check:cycles`, `pnpm test`, `pnpm build`); `pnpm push`; entrada em `docs/changelog/<data>-c219.md`; PR ready com auto-merge (o check `checks` é o gate; o e2e curado roda `campaignSpeechAcervo`).
+
+## Rabbit holes / Não escopo (engenharia)
+
+- **Generalizar a barra da Câmara** (`SpeechAcervoFilters` configurável / componente `AcervoFilters` genérico) — anti-goal; shells compartilhados já são os donos.
+- **Extrair o adaptador de omnibox/helpers de estado para um genérico** — 2 call sites; extrair quando C216 ("Falas na internet") chegar (3º).
+- **Módulo compartilhado novo do contrato do acervo** (mover buckets/parsers para um terceiro) — churn na Câmara sem volatilidade; deferido para C216.
+- **pgvector/embeddings/índice vetorial** — já rejeitado no C192.
+- **Classificar por segmento/falante** ou reclassificar a taxonomia (18 temas fixos) — cortes de produto.
+- **Segundo cadastro de pessoa / inferir falantes** — "Pessoa" segue rótulo (C200).
+- **Mudar o default de ordenação para `-recordedAt`** — regressão do comportamento atual (`-createdAt`); fora do aceite.
+- **Busca global entre fontes / lista única** — anti-goal.
+- **UI de curadoria de facetas além do form do admin** — o form da collection é a superfície de revisão; qualquer coisa além disso é outro item.
+- **Cache/rerank da expansão de tema** — deferido no C192 (sem cache no v1).
+
+**Adiado com gatilho (triage do /simplify desta entrega):**
+
+- **Extração dos helpers de faceta/estado restantes** (`facetRows`/`clearFacet`/`runAction` e o aviso de fallback por domínio) — gatilho: C216 (3º consumidor do acervo), junto do adaptador de omnibox (S2/S3).
+- **N+1 do backfill** (`scripts/backfill-recording-facets.mjs` consulta os segmentos de cada gravação em queries separadas) — gatilho: acervo na casa das centenas de gravações ou `--all` perceptivelmente lento (S4).
+- **Ternários aninhados do ramo de gravações** em `acervo/page.tsx` — gatilho: extração do 3º consumidor na C216, quando o ramo for compartilhado (S5).
+- **"X" de limpar do omnibox quebra linha no mobile com chip longo** — débito do shell compartilhado, pré-existente e reproduzido na Câmara; registrado como Issue própria (C221, `docs/plans/omnibox-limpar-mobile.md`) — fora deste item (S1).
+
+## Riscos e mitigação
+
+- **Deep-link/redirect:** os nomes novos entram em `recordingListParamNames` (antes seriam "unsupported" e redirecionariam); o serializer precisa reproduzir `?source=enviadas&q=…&person=…&page=…` **byte a byte** (ordem canônica acima; `page` só >1). Mitigação: pins no `recordingListUrl.unit.spec.ts` (incluindo `mode=tema` sem `q` canonicalizando fora e `sort` default omitido).
+- **`select` allowlist:** esquecer `topics` no `recordingListSelect` some com o chip do card silenciosamente. Mitigação: o campo é exigido pelo input do view model (typecheck) + int test do loader.
+- **Manifest e2e:** `src/collections/Recording.ts` está fora do manifest (um diff só nele cairia no smoke da home); `src/payload-types.ts` é `HIGH_RISK_EXACT` → o PR roda o e2e **curado**, que hoje não inclui `campaignSpeechAcervo`. Mitigação: adicionar o arquivo ao entry da vertical e `campaignSpeechAcervo` ao `E2E_CURATED_SPECS` + pin (precedente C167/S19); a suíte full continua no `verify` do deploy.
+- **Ausência de `DEEPINFRA_API_KEY`:** `classifySpeech` degrada para gazetteer (topics/scopes/municípios do léxico) e nunca lança; o CLI reporta `llm: false` e o `mergeFacetClassification` garante que `classifiedBy: 'llm'` só sai com resposta validada.
+- **Gravações já existentes:** `classifiedBy` **nulo = "não classificada"** (honesto; nunca inventar `gazetteer` por default); ficam visíveis e sem faceta até o `pnpm recordings:classify` rodar; o CLI é idempotente e não toca `manual`.
+- **Transcrição longa:** o prompt do LLM trunca em 12k chars (dono: `classifySpeech`) — classificação pelo início da transcrição, registrada no `admin.description`; a busca (trgm) não é afetada. O custo é 1 chamada por gravação `ready` (mesmo provedor/chave do ASR), nunca por request.
+- **Sort × duração nula:** sem o gate `durationSeconds exists`, o Postgres colocaria as linhas sem duração no topo de "Duração (maior)"; o gate + hint no controle mantêm a ordem honesta. Int test cobre.
+- **Matriz de papéis:** intocada — gate da página e `canReadRecording` idênticos; a expansão de tema só roda depois de `canReadCommunicationCatalog(role)`; os bypasses do job/CLI são internos e justificados.
+- **Migration em produção:** aditiva (colunas nulas + joins), aplicada pelo `pnpm build` do deploy; o backfill de `year` é SQL determinístico; `down` gerado cobre rollback.
+- **Ciclos/convenções:** `recordings → speech` não pode criar ciclo (`speech` não importa `recordings`) — `pnpm check:cycles`; `recordingClassification.ts` declara `import 'server-only'` (regra do `codebaseConventions.unit.spec.ts`).
+
+## Aceite de engenharia
+
+- [ ] Aceite de produto da intenção ainda coberto (ano, duração, tema, alcance, município citado, pessoa exclusiva, contagem, chips, ordenação, vazio honesto, tema degradando honestamente).
+- [ ] Invariantes AGENTS/engineering-standards: `user` + `overrideAccess: false` em toda leitura; bypass do job/CLI justificado; sem `Consent`/PII nova; `leader`/`advisor` fail-closed; sem top-level novo em `utilities/`; copy pt-BR / identificadores em inglês; sem `openai/*` (OPS123).
+- [ ] Migrations: só arquivos novos (nunca editar antigas); `pnpm migrate` local; `pnpm generate:types`; prod aplica no build.
+- [ ] Testes de domínio previstos: unit (URL/where/omnibox/corpus), int (job escreve facetas, curadoria manual preservada, loader filtra/ordena/tema), e2e (barra/filtros/ordenação/vazio) — onde os write paths e o contrato de URL mudam.
+- [ ] Manifest/e2e atualizado (`src/collections/Recording.ts` na vertical; `campaignSpeechAcervo` no curado + pin).
+- [ ] Gates: `tsc --noEmit`, `pnpm lint` (0 warnings), `pnpm format:check`, `pnpm exec knip`, `pnpm check:cycles`, `pnpm test`, `pnpm build`; `pnpm push`; entrada em `docs/changelog/`.
+
+## Decisões de engenharia
+
+**(i) Onde a classificação roda**
+
+- **Opções:** A) seam no `runRecordingJob` (injetável) + CLI backfill | B) hook `afterChange` da collection | C) on-demand no loader da busca | D) hook por `recordingSegment`.
+- **Recomendação:** **A** — o job já tem `keyedSegments` em memória ao fim da transcrição, roda fora do request e faz o save transacional; o `classify` injetável (default `classifySpeech`) é o mesmo padrão de `catalogContentPiece` e mantém os testes offline; o CLI cobre as linhas anteriores à entrega com a mesma função.
+- **Rejeitadas:** B porque o hook dispara em cada transição de `status`/`step` (sem transcrição no momento útil), prenderia um LLM de 60s dentro do ciclo de escrita e re-executaria a cada passo; C porque pagaria o LLM em toda página, transformaria leitura em escrita e não filtraria pelo que ainda não foi classificado; D porque faria N chamadas por gravação e classificaria por trecho — unidade rejeitada pela intenção.
+
+**(ii) Schema em `Recording`**
+
+- **Opções:** A) espelhar `Speech` (`year` derivado, `topics`/`scopes`/`classifiedBy`/`mentionedMunicipalities`, `preserveManualFacets`) com a migration `add_recording_facets` | B) facetas por segmento | C) sem schema, derivar na busca | D) ano por range sobre `recordedAt` (sem coluna).
+- **Recomendação:** **A** — mantém `where`, opções de faceta e canônico idênticos aos da Câmara (a paridade é o objetivo), a curadoria manual preservada (hook espelho) e o "revisável" no admin. `year` derivado no `beforeValidate` com backfill SQL na migration; `classifiedBy` opcional (ver vii).
+- **Rejeitadas:** B porque a unidade da Câmara é a fala e o corte da intenção é a gravação; C porque não filtra com `where`/paginação, recalcula a cada busca e perde a revisão; D porque criaria um segundo shape de filtro para o mesmo conceito (e teria de repetir a regra de fuso em cada branch), enquanto a coluna derivada é literalmente o "derivado da data já informada" da intenção.
+
+**(iii) Reuso do contrato de duração e do omnibox**
+
+- **Opções:** A) reusar os donos da Câmara (buckets/labels/tipo em `speechListUrl`, `durationBucketWhere` exportado de `speechListFilters`, parsers de ano/município em `campaignListUrl`) + `recordingOmnibox` novo | B) extrair módulo compartilhado novo do acervo e migrar a Câmara | C) duplicar buckets/where no domínio de gravações.
+- **Recomendação:** **A** — um dono para o bucket (mudança na Câmara reflete nas gravações, que é a paridade pedida) e o adaptador por domínio segue o padrão já existente (`municipalityOmnibox`, `municipalityUpdateOmnibox`); o que é genérico (`lib/campaignListOmnibox`) já é compartilhado.
+- **Rejeitadas:** B porque mover o contrato de referência para um terceiro neutro gera churn na Câmara sem volatilidade real — gatilho de revisitação: a chegada da C216 (3º consumidor); C porque duplicar o bucket/where faz as duas fontes divergirem em silêncio, exatamente o que o item existe para impedir.
+
+**(iv) Corpus da busca por tema**
+
+- **Opções:** A) `ThemeSearchCorpus` ganha `'recording'` + prompt próprio + wrapper | B) reusar o corpus `'speech'` | C) só busca literal.
+- **Recomendação:** **A** — `expandSearchTheme(theme, corpus)` é o mecanismo único desde S28; o que muda é o prompt (plenárias/debates/material próprio enviados pela equipe). Degradação e `normalizeSpeechThemeTerms` são herdados.
+- **Rejeitadas:** B porque o prompt descreve discursos da Câmara e degradaria a expansão para gravações; C porque a pergunta de produto foi confirmada no gate (estender a ponte é a paridade pedida).
+
+**(v) UI: componente novo × generalização**
+
+- **Opções:** A) `RecordingAcervoFilters` novo sobre os shells compartilhados | B) generalizar `SpeechAcervoFilters` configurável | C) estender `SpeechAcervoFilters` com `source`.
+- **Recomendação:** **A** — os donos das mecânicas são os shells (`CampaignListOmnibox`, `CampaignHeaderFilterPopover`, `useCampaignListFilterNavigation`, `CampaignListPending*`, `CampaignListEmptyState/Footer`); o componente de domínio só monta os facets da fonte, como o da Câmara.
+- **Rejeitadas:** B porque uma API genérica configurável é pass-through raso e mexeria na Câmara (anti-goal); C porque acoplaria os dois contratos de URL num único componente e twinaria a Câmara dentro dele.
+
+**(vi) Contrato de ordenação**
+
+- **Opções:** A) `sort` com valores pt-BR (`recentes` default omitido, `duracao_maior`, `duracao_menor`) + exclusão de linhas sem duração nos sorts de duração | B) `sort`/`dir` genérico (C117) | C) sort só de UI.
+- **Recomendação:** **A** — contrato fechado (o canonicalizador redireciona valor desconhecido), default preserva `-createdAt` (zero regressão) e o gate `durationSeconds exists` evita NULLS FIRST no Postgres; o controle avisa que gravações sem duração aparecem só em "Mais recentes".
+- **Rejeitadas:** B porque o acervo não é tabela e a Câmara não tem esse contrato (a paridade pede o gesto, não o par genérico); C porque quebraria "paginação preservada no deep link" e o back/forward.
+
+**(vii) Proveniência nula nas gravações existentes**
+
+- **Opções:** A) `classifiedBy` opcional, sem default — nulo = "não classificada" | B) `required` com `defaultValue: 'gazetteer'` (espelho exato de `Speech`).
+- **Recomendação:** **A** — o default de B afirmaria que o gazetteer classificou linhas que nunca passaram por ele (viola o "nunca inventa"); nulo é o estado honesto até o `pnpm recordings:classify`, e o seletor do CLI fica trivial (`classifiedBy` ausente). Os **valores** persistidos são os mesmos da Câmara (`gazetteer|llm|manual`) — a paridade é do vocabulário, não do `required`.
+- **Rejeitadas:** B porque a migration carimbaria proveniência falsa nas linhas antigas e o backfill não teria como distinguir "classificada" de "nunca classificada".
+
+**Decisões baratas (com gatilho de revisitação):**
+
+- **Excerto no modo tema:** evidência de tema vence o excerto (espelho C192) e o `watchHref` mantém o `q` literal no `?t=&q=` (byte-compat; espelho `buildWatchHref`). Revisitar se a Câmara mudar a regra.
+- **`CampaignThemeRetryButton` movido para `shared/`** (componente genérico, 1 import atualizado na Câmara) — o aviso em si fica por domínio (copy).
+- **Helpers de estado do omnibox (`withPageReset`/`toggleValue`/`removeValue`)** duplicados no adaptador de gravações; extrair na C216 (3º).
+- **Contagem no `CampaignListFooter`** (não na linha do sort, como no artefato) — padrão de todas as listas; sem duplicar a linha.
+- **Chip de tema no card** usa `bg-muted` (mesma linguagem do chip de pessoa) — sem selo "Tema" (o artefato não pede).
+
+**Self-score (0–5, gate ≥4): 5/5.**
+
+1. **Decisões caras com rejeitadas:** 5/5 — as seis decisões caras (seam, schema, reuso de contrato, corpus, UI, sort) e a de proveniência nula têm opções/recomendação/rejeitadas; as baratas ficam com gatilho.
+2. **Appetite:** 5/5 — sem infra/provedor novo, sem collection/rota nova, reusando classificador, ponte de tema, shells e a transação do job; cabe nos ~2–3 dias.
+3. **Rabbit holes nomeados:** 5/5 — barra genérica, módulo compartilhado precoce, embeddings, classificação por trecho, segundo cadastro, mudança de default de sort, lista única.
+4. **Depth check:** 5/5 — reusa `classifySpeech`/`lib/speechFacets`, `expandSearchTheme`, `campaignListOmnibox`, shells de lista, `loadMunicipalityLabelsByIds`, `SpeechHighlightParts`, `withPayloadTransaction`; nada de camada cerimonial.
+5. **Intenção preservada:** 5/5 — Pessoa exclusiva, gate fail-closed, deep-link byte-compatível, sem lista única, sem redesenho da Câmara, classificação auditável e degradando honestamente.

docs/plans/omnibox-limpar-mobile.md

@@ -0,0 +1,36 @@
+# C221 — "Limpar" do omnibox não pode quebrar linha no mobile
+
+Status: rascunho
+Atualizado em: 2026-09-24
+Issue: (a registrar)
+Priority: P3
+Impeccable: B — ajuste de encaixe do shell de filtros existente (não é fluxo novo)
+Appetite: ~2h eng (fill-in)
+
+## Intenção
+
+Débito capturado na crítica de design da C219 (`docs/plans/acervo-paridade-busca-filtros.md`): no mobile, quando o chip de busca do omnibox compartilhado é longo, o botão global "Limpar" (o "X" circular dentro do campo, `md:hidden`) quebra para uma linha própria e aparece órfão abaixo do chip, criando uma faixa vazia entre o campo e o seletor de modo. Reproduzido igualmente na fonte Câmara e na fonte "Gravações enviadas" — é comportamento pré-existente do `CampaignListOmnibox`, não regressão da C219.
+
+Evidência: `/tmp/opencode/c219-shots/06-mobile-empty.png` (gravações) e `/tmp/opencode/c219-shots/11-mobile-camara-long-chip.png` (Câmara) na sessão da C219; o código é o cluster `flex-wrap` do campo em `src/components/campaign/shared/CampaignListOmnibox.tsx` (chips + input + botão de limpar na mesma linha flexível).
+
+## Objetivo e aceite
+
+- Com um chip de busca longo no mobile, o "X" de limpar permanece visualmente vinculado ao campo/chips (não quebra para uma linha órfã), em todas as listas que usam o shell.
+- Comportamento, acessibilidade (`aria-label="Limpar"`, anel de foco, região `sr-only`) e desktop intocados.
+- Sem mudança de contrato de URL/filtros.
+
+## Direção no codebase (hipótese)
+
+- `src/components/campaign/shared/CampaignListOmnibox.tsx` — cluster do campo (`flex min-h-10 w-full flex-wrap items-center gap-1.5 ...`) com chips, input e o botão de limpar `md:hidden`.
+- Consumidores para regressão visual: `SpeechAcervoFilters`, `RecordingAcervoFilters`, listas de municípios/apoiadores (qualquer `<form>` com `campaignListOmniboxFormClassName`).
+- Estratégias possíveis (decidir no impl): manter o botão fora do fluxo dos chips (posição absoluta no campo), agrupar chips+input num contêiner que quebra e ancorar o X na linha do input, ou limitar a largura do chip no mobile. A mais simples que preserve o layout desktop.
+
+## Fora de escopo
+
+- Redesenho da barra da Câmara ou dos shells de lista.
+- Mudanças nos adaptadores de omnibox por domínio.
+
+## Aceite de engenharia
+
+- [ ] Regressão visual mobile (390) em pelo menos uma lista de cada vertical que usa o shell.
+- [ ] Nenhuma mudança de markup/ARIA que quebre os e2e existentes das listas.

package.json

@@ -90,6 +90,7 @@
     "camara:pilot": "cross-env NODE_OPTIONS=\"--no-deprecation --import=tsx/esm\" node scripts/pilot-camara-speeches.mjs",
     "camara:import": "cross-env NODE_OPTIONS=\"--no-deprecation --import=tsx/esm --import=./scripts/seed-loader.mjs\" node scripts/import-camara-speeches.mjs",
     "falas-web:import": "cross-env NODE_OPTIONS=\"--no-deprecation --import=tsx/esm --import=./scripts/seed-loader.mjs\" node scripts/import-web-speeches.mjs",
+    "recordings:classify": "cross-env NODE_OPTIONS=\"--no-deprecation --import=tsx/esm --import=./scripts/seed-loader.mjs\" node scripts/backfill-recording-facets.mjs",
     "typecheck": "tsc --noEmit",
     "typecheck:tsc6": "tsc6 --noEmit",
     "prepare": "husky",

scripts/backfill-recording-facets.mjs

@@ -0,0 +1,162 @@
+/**
+ * C219 — facet backfill of the uploaded recordings ("Gravações enviadas").
+ *
+ * The transcription job classifies every recording it finishes; this CLI
+ * classifies the rows that already existed (and can reclassify on demand) with
+ * the SAME function the job uses (`classifyRecordingFacets` → `classifySpeech`:
+ * offline gazetteer + one validated LLM refinement on Deep Infra). It never
+ * invents provenance: a row is only written when the classifier produced a
+ * result, a `manual` curation is never touched, and the default selection is
+ * exactly the rows with no `classifiedBy` (never classified).
+ *
+ * Idempotent: re-running skips what it already classified (default mode);
+ * `--all` reclassifies everything except `manual` rows.
+ *
+ * The local-DB guard refuses a non-local DATABASE_URL, and any write run
+ * targeting production (or a remote/override DB) additionally requires
+ * `RECORDING_CLASSIFY_CONFIRM=1` (same family as `CAMARA_IMPORT_CONFIRM`).
+ *
+ * Usage:
+ *   pnpm recordings:classify --dry-run
+ *   pnpm recordings:classify --limit 5
+ *   pnpm recordings:classify
+ *   pnpm recordings:classify --all
+ *   RECORDING_CLASSIFY_CONFIRM=1 pnpm recordings:classify --all
+ */
+import { getPayload } from 'payload'
+
+import { assertLocalDatabase } from './assert-local-database.mjs'
+import { dieWithLabel, isTruthyEnv, loadCliEnv, requiresWriteConfirm } from './lib/cli.mjs'
+
+loadCliEnv()
+
+const die = dieWithLabel('recordings:classify')
+const WRITE_CONFIRM_FLAG = 'RECORDING_CLASSIFY_CONFIRM'
+
+const parseArgs = (argv) => {
+  const options = { dryRun: false, all: false, limit: null }
+  for (const arg of argv) {
+    if (arg === '--dry-run') options.dryRun = true
+    else if (arg === '--all') options.all = true
+    else if (arg.startsWith('--limit=')) {
+      const limit = Number(arg.slice('--limit='.length))
+      if (!Number.isSafeInteger(limit) || limit <= 0) die(`--limit inválido: ${arg}`)
+      options.limit = limit
+    } else die(`argumento desconhecido: ${arg}`)
+  }
+  return options
+}
+
+const options = parseArgs(process.argv.slice(2))
+
+assertLocalDatabase(
+  'recordings:classify',
+  'A classificação escreve em `recording`; aponte DATABASE_URL para o banco local.',
+)
+
+if (requiresWriteConfirm() && !isTruthyEnv(process.env[WRITE_CONFIRM_FLAG])) {
+  die(
+    `Este run escreve em um banco remoto/de produção. Se é isso mesmo, re-execute com:\n  ${WRITE_CONFIRM_FLAG}=1 pnpm recordings:classify …`,
+  )
+}
+
+const config = (await import('../src/payload.config.ts')).default
+const { classifyRecordingFacets, recordingFacetWriteData } =
+  await import('../src/utilities/recordings/recordingClassification.ts')
+
+const payload = await getPayload({ config })
+
+const where = options.all
+  ? { and: [{ status: { equals: 'ready' } }, { classifiedBy: { not_equals: 'manual' } }] }
+  : { and: [{ status: { equals: 'ready' } }, { classifiedBy: { exists: false } }] }
+
+const targets = await payload.find({
+  collection: 'recording',
+  where,
+  depth: 0,
+  limit: options.limit ?? 0,
+  pagination: false,
+  sort: 'createdAt',
+  select: { title: true, classifiedBy: true },
+  // Intentional admin bypass: the backfill CLI is a trusted actor with no session.
+  overrideAccess: true,
+})
+
+const report = {
+  targets: targets.docs.length,
+  classified: 0,
+  skippedEmptyTranscript: 0,
+  failed: 0,
+  llmUsed: 0,
+  totalTokens: 0,
+  estimatedCostUsd: 0,
+}
+
+console.log(
+  `[recordings:classify] ${report.targets} gravação(ões) alvo${options.dryRun ? ' (dry-run)' : ''}.`,
+)
+
+for (const recording of targets.docs) {
+  const segments = await payload.find({
+    collection: 'recordingSegment',
+    where: { recording: { equals: recording.id } },
+    depth: 0,
+    limit: 0,
+    pagination: false,
+    sort: 'order',
+    select: { text: true },
+    // Intentional admin bypass: reconstructing the transcript the job owns.
+    overrideAccess: true,
+  })
+  const transcript = segments.docs.map((segment) => segment.text).join(' ')
+
+  const classification = await classifyRecordingFacets({ payload, transcript })
+  if (!classification) {
+    report.skippedEmptyTranscript += 1
+    console.warn(`[recordings:classify] #${recording.id} sem transcrição utilizável — pulada.`)
+    continue
+  }
+
+  if (classification.llm.used) report.llmUsed += 1
+  if (classification.llm.totalTokens !== null) report.totalTokens += classification.llm.totalTokens
+  if (classification.llm.estimatedCostUsd !== null) {
+    report.estimatedCostUsd += classification.llm.estimatedCostUsd
+  }
+
+  if (options.dryRun) {
+    report.classified += 1
+    console.log(
+      `[recordings:classify] (dry-run) #${recording.id} "${recording.title}" → ` +
+        `${classification.topics.join(', ') || '—'} · ${classification.classifiedBy}`,
+    )
+    continue
+  }
+
+  try {
+    await payload.update({
+      collection: 'recording',
+      id: recording.id,
+      data: recordingFacetWriteData(classification),
+      // Intentional admin bypass: the backfill CLI is a trusted actor with no session.
+      overrideAccess: true,
+    })
+    report.classified += 1
+    console.log(
+      `[recordings:classify] #${recording.id} "${recording.title}" → ` +
+        `${classification.topics.join(', ') || '—'} · ${classification.classifiedBy}`,
+    )
+  } catch (error) {
+    report.failed += 1
+    console.warn(
+      `[recordings:classify] #${recording.id} falhou ao gravar: ` +
+        `${error instanceof Error ? error.message : String(error)}`,
+    )
+  }
+}
+
+console.log(
+  `\n[recordings:classify] alvos=${report.targets} classificadas=${report.classified} ` +
+    `sem-transcrição=${report.skippedEmptyTranscript} falhas=${report.failed} ` +
+    `llm=${report.llmUsed} tokens=${report.totalTokens} custo≈$${report.estimatedCostUsd.toFixed(4)}`,
+)
+process.exit(report.failed > 0 ? 1 : 0)

scripts/lib/e2e-affected-manifest.mjs

@@ -47,6 +47,10 @@ export const E2E_CURATED_SPECS = [
   // and the vote share sheet are new contracts over the C211 collection
   // (curated only).
   'frontendConteudos',
+  // C219 — deliberate: the recordings acervo gained its own search/facet
+  // contract and the migration makes every PR of this delivery high-risk
+  // (curated only).
+  'campaignSpeechAcervo',
 ]
 
 /**
@@ -443,6 +447,10 @@ export const E2E_AFFECTED_MANIFEST = [
       'src/components/campaign/reels',
       // C199 — the uploaded recordings source and its private media owner.
       'src/components/campaign/recording',
+      // C219 — the recording collection owns the facet/provenance hooks the
+      // recordings list filters on (the PR is high-risk anyway; the entry keeps
+      // the mapping honest for the curated/local paths).
+      'src/collections/Recording.ts',
       // C211 — the Central de Conteúdos surfaces.
       'src/components/campaign/content',
       'src/utilities/speech',

src/app/(campaign)/campanha/(app)/comunicacao/acervo/page.tsx

@@ -5,9 +5,11 @@ import { redirect } from 'next/navigation'
 import { getPayload } from 'payload'
 
 import { AcervoSourceToggle } from '@/components/campaign/recording/AcervoSourceToggle'
-import { RecordingPersonFilter } from '@/components/campaign/recording/RecordingPersonFilter'
+import { RecordingAcervoFilters } from '@/components/campaign/recording/RecordingAcervoFilters'
 import { RecordingResultList } from '@/components/campaign/recording/RecordingResultList'
+import { RecordingSortSelect } from '@/components/campaign/recording/RecordingSortSelect'
 import { RecordingStatusRefresher } from '@/components/campaign/recording/RecordingStatusRefresher'
+import { RecordingThemeFallbackNotice } from '@/components/campaign/recording/RecordingThemeFallbackNotice'
 import { RecordingUploadDialog } from '@/components/campaign/recording/RecordingUploadDialog'
 import { CampaignListEmptyState } from '@/components/campaign/shared/CampaignListEmptyState'
 import { CampaignListFooter } from '@/components/campaign/shared/CampaignListFooter'
@@ -15,21 +17,22 @@ import {
   CampaignListPendingBoundary,
   CampaignListResults,
 } from '@/components/campaign/shared/CampaignListPending'
-import { CampaignSearchInput } from '@/components/campaign/shared/CampaignSearchInput'
+import { CampaignThemeRetryButton } from '@/components/campaign/shared/CampaignThemeRetryButton'
 import { CampaignPageShell } from '@/components/campaign/shell/CampaignPageShell'
 import { SpeechAcervoFilters } from '@/components/campaign/speech/SpeechAcervoFilters'
 import { SpeechRefineSearchButton } from '@/components/campaign/speech/SpeechRefineSearchButton'
 import { SpeechResultList } from '@/components/campaign/speech/SpeechResultList'
 import { SpeechThemeFallbackNotice } from '@/components/campaign/speech/SpeechThemeFallbackNotice'
-import { SpeechThemeRetryButton } from '@/components/campaign/speech/SpeechThemeRetryButton'
 import { Button } from '@/components/ui/button'
 import { campaignPageMetadataFromCatalog } from '@/lib/campaignPageChrome'
 import { CAMPAIGN_COMMUNICATION_ACERVO, CAMPAIGN_COMMUNICATION_CORTES } from '@/lib/campaignPaths'
 import { requireCampaignPageActor } from '@/utilities/campaignPageActor'
 import {
   buildAcervoSourceHref,
+  buildRecordingFiltersKey,
   buildRecordingListHref,
   parseAcervoSource,
+  recordingHasActiveFilters,
 } from '@/utilities/recordings/recordingListUrl'
 import { loadRecordingsPageData } from '@/utilities/recordings/recordingPageData'
 import { buildSpeechFiltersKey, buildSpeechListHref } from '@/utilities/speech/speechListUrl'
@@ -78,80 +81,93 @@ const AcervoHeader = ({ source }: { source: 'camara' | 'enviadas' }) => (
   </div>
 )
 
-const RecordingsSearchForm = ({
-  query,
-  people,
-}: {
-  query?: string
-  people?: readonly string[]
-}) => (
-  <form method="get" action={CAMPAIGN_COMMUNICATION_ACERVO} className="flex flex-col gap-1.5">
-    <input type="hidden" name="source" value="enviadas" />
-    {/* The "Pessoa" facet survives a new textual search. */}
-    {(people ?? []).map((person) => (
-      <input key={person} type="hidden" name="person" value={person} />
-    ))}
-    {/* The shared input owns the accessible label; this is its visible twin. */}
-    <p className="text-sm font-medium">Buscar nas gravações enviadas</p>
-    <CampaignSearchInput
-      id="recording-search"
-      name="q"
-      label="Buscar nas gravações enviadas"
-      defaultValue={query}
-      placeholder="Busque por palavra ou trecho da transcrição…"
-    />
-    <button type="submit" className="sr-only">
-      Buscar
-    </button>
-  </form>
-)
-
 const RecordingsSource = ({
   data,
 }: {
   data: Awaited<ReturnType<typeof loadRecordingsPageData>>
 }) => {
-  const hasSearch = Boolean(data.state.q)
-  const hasPeople = (data.state.people?.length ?? 0) > 0
-  const hasFilters = hasSearch || hasPeople
+  const hasFilters = recordingHasActiveFilters(data.state)
+  const themeMode = data.state.mode === 'tema'
+  const themeActive = themeMode && data.themeApplied && !data.themeUnavailable
 
   return (
     <CampaignListPendingBoundary>
-      <RecordingsSearchForm query={data.state.q} people={data.state.people} />
+      <RecordingAcervoFilters
+        key={buildRecordingFiltersKey(data.state)}
+        state={data.state}
+        filterOptions={data.filterOptions}
+        themeUnavailable={data.themeUnavailable}
+      />
 
-      <RecordingPersonFilter state={data.state} options={data.filterOptions.people} />
+      {data.themeUnavailable ? <RecordingThemeFallbackNotice /> : null}
 
       <RecordingStatusRefresher
         recordings={data.rows.map((row) => ({ id: row.id, status: row.status }))}
       />
 
       <CampaignListResults>
+        <div className="flex flex-wrap items-end justify-between gap-3">
+          <div>
+            <h2 className="text-sm font-semibold">
+              {data.themeUnavailable
+                ? 'Resultados por termo exato'
+                : themeActive
+                  ? 'Resultados por tema'
+                  : 'Gravações encontradas'}
+            </h2>
+            {themeMode ? (
+              <p className="mt-0.5 text-xs text-muted-foreground">
+                {data.themeUnavailable
+                  ? 'Comportamento atual do acervo.'
+                  : themeActive
+                    ? 'Confira o indício em cada gravação antes de abrir.'
+                    : 'Nenhum termo relacionado foi acrescentado; mostramos a busca literal.'}
+              </p>
+            ) : null}
+          </div>
+          <div className="flex flex-wrap items-end gap-2">
+            {data.themeUnavailable ? <CampaignThemeRetryButton /> : null}
+            <RecordingSortSelect state={data.state} />
+          </div>
+        </div>
+
         {data.rows.length > 0 ? (
           <RecordingResultList rows={data.rows} />
         ) : (
           <CampaignListEmptyState
-            className="border-solid"
+            className={themeActive ? 'border-solid' : undefined}
             icon={hasFilters ? SearchIcon : VideoIcon}
             title={
-              hasSearch
-                ? `Nenhuma gravação encontrada para "${data.state.q}"`
-                : hasPeople
-                  ? 'Nenhuma gravação encontrada com esse filtro'
-                  : 'Nenhuma gravação enviada ainda'
+              themeActive
+                ? 'Nenhuma gravação encontrada para este tema'
+                : data.state.q
+                  ? `Nenhuma gravação encontrada para "${data.state.q}"`
+                  : hasFilters
+                    ? 'Nenhuma gravação encontrada com esses filtros'
+                    : 'Nenhuma gravação enviada ainda'
             }
             description={
-              hasSearch
-                ? 'Tente outro termo ou limpe a busca para ver todas as gravações enviadas.'
-                : hasPeople
+              themeActive
+                ? 'Não encontramos uma gravação que corresponda ao sentido desta busca com os filtros atuais. Não vamos preencher a lista com resultados pouco relacionados.'
+                : hasFilters
                   ? 'Tente outro termo, remova filtros ou limpe a busca.'
                   : 'Envie um arquivo de vídeo para acompanhar a transcrição e pesquisar o conteúdo no acervo.'
             }
           >
-            {hasFilters ? (
+            {themeActive ? (
+              <div className="flex flex-wrap items-center justify-center gap-2">
+                <Button asChild className="min-h-11">
+                  <Link href={buildAcervoSourceHref('enviadas')}>Limpar filtros</Link>
+                </Button>
+                <Button asChild variant="outline" className="min-h-11">
+                  <Link href={buildRecordingListHref({ ...data.state, mode: undefined }, 1)}>
+                    Trocar para termo exato
+                  </Link>
+                </Button>
+              </div>
+            ) : hasFilters ? (
               <Button asChild variant="outline" className="min-h-11">
-                <Link href={buildAcervoSourceHref('enviadas')}>
-                  {hasPeople ? 'Limpar filtros' : 'Limpar busca'}
-                </Link>
+                <Link href={buildAcervoSourceHref('enviadas')}>Limpar filtros</Link>
               </Button>
             ) : (
               <RecordingUploadDialog triggerLabel="Enviar gravação" />
@@ -233,7 +249,7 @@ export default async function SpeechAcervoPage({ searchParams }: SpeechAcervoPag
                       : 'Nenhum termo relacionado foi acrescentado; mostramos a busca literal.'}
                 </p>
               </div>
-              {data.themeUnavailable ? <SpeechThemeRetryButton /> : null}
+              {data.themeUnavailable ? <CampaignThemeRetryButton /> : null}
             </div>
           ) : null}
 

src/collections/Recording.ts

@@ -1,4 +1,5 @@
 import type {
+  CollectionBeforeChangeHook,
   CollectionBeforeDeleteHook,
   CollectionBeforeValidateHook,
   CollectionConfig,
@@ -14,6 +15,7 @@ import {
   recordingStepLabels,
 } from '@/lib/recording'
 import { speakerNamesFromLabels } from '@/lib/recordingDiarization'
+import { SPEECH_CLASSIFICATION_SOURCES, SPEECH_SCOPES, SPEECH_TOPICS } from '@/lib/speechFacets'
 import {
   canDeleteRecording,
   canReadRecording,
@@ -67,6 +69,44 @@ const deriveSpeakerNames: CollectionBeforeValidateHook = ({ data, originalDoc })
   return data
 }
 
+/**
+ * C219 — the "Ano" facet is derived from the recording date the team already
+ * informed (`recordedAt`), never a second hand-typed field. A recording without
+ * a date honestly has no year (it stays out of the year filter).
+ */
+const deriveRecordingYear: CollectionBeforeValidateHook = ({ data, originalDoc }) => {
+  if (!data) return data
+  const recordedAt =
+    typeof data.recordedAt === 'string' ? data.recordedAt : (originalDoc?.recordedAt ?? null)
+  const year = Number(String(recordedAt ?? '').slice(0, 4))
+  // Clearing the date clears the derived year too — a stale year would keep the
+  // recording in an "Ano" facet the team just removed.
+  data.year = Number.isInteger(year) && year > 1900 ? year : null
+  return data
+}
+const FACET_FIELDS = ['topics', 'scopes', 'classifiedBy', 'mentionedMunicipalities'] as const
+
+/**
+ * C219 — a manual facet correction is curation: unauthenticated writes (the
+ * transcription job and the backfill CLI) never overwrite it. Authenticated
+ * admin edits stay free to change both the facets and the provenance.
+ */
+const preserveManualFacets: CollectionBeforeChangeHook = ({
+  data,
+  operation,
+  originalDoc,
+  req,
+}) => {
+  if (!data || operation !== 'update' || originalDoc?.classifiedBy !== 'manual') return data
+  if (req.user) return data
+
+  const next = data as Record<string, unknown>
+  for (const field of FACET_FIELDS) {
+    if (field in next) next[field] = originalDoc[field]
+  }
+  return data
+}
+
 export const Recording: CollectionConfig = {
   slug: 'recording',
   labels: {
@@ -76,9 +116,9 @@ export const Recording: CollectionConfig = {
   admin: {
     group: 'Comunicação',
     useAsTitle: 'title',
-    defaultColumns: ['title', 'recordedAt', 'status', 'durationSeconds', 'createdAt'],
+    defaultColumns: ['title', 'recordedAt', 'status', 'topics', 'classifiedBy'],
     description:
-      'Gravações próprias da equipe no acervo. O arquivo é privado; a transcrição é somente leitura.',
+      'Gravações próprias da equipe no acervo. O arquivo é privado; a transcrição é somente leitura. Tema, alcance e municípios citados são classificados automaticamente pelo início da transcrição e podem ser corrigidos aqui (a correção vira proveniência "Manual").',
   },
   access: {
     create: canUploadRecording,
@@ -87,8 +127,8 @@ export const Recording: CollectionConfig = {
     delete: canDeleteRecording,
   },
   hooks: {
-    beforeValidate: [deriveSpeakerNames],
-    beforeChange: [stampCampaignCreatedBy],
+    beforeValidate: [deriveSpeakerNames, deriveRecordingYear],
+    beforeChange: [preserveManualFacets, stampCampaignCreatedBy],
     beforeDelete: [deleteRecordingSegments],
   },
   fields: [
@@ -108,6 +148,16 @@ export const Recording: CollectionConfig = {
         description: 'Data em que a gravação foi feita (opcional).',
       },
     },
+    {
+      name: 'year',
+      type: 'number',
+      label: 'Ano',
+      index: true,
+      admin: {
+        readOnly: true,
+        description: 'Derivado da data da gravação.',
+      },
+    },
     {
       name: 'status',
       type: 'select',
@@ -155,6 +205,47 @@ export const Recording: CollectionConfig = {
         description: 'Concatenação normalizada dos segmentos (sem acentos, minúsculas).',
       },
     },
+    {
+      name: 'topics',
+      type: 'select',
+      label: 'Temas',
+      hasMany: true,
+      index: true,
+      options: SPEECH_TOPICS.map(({ value, label }) => ({ value, label })),
+    },
+    {
+      name: 'scopes',
+      type: 'select',
+      label: 'Alcance',
+      hasMany: true,
+      index: true,
+      options: SPEECH_SCOPES.map(({ value, label }) => ({ value, label })),
+    },
+    {
+      name: 'classifiedBy',
+      type: 'select',
+      label: 'Proveniência da classificação',
+      // Optional on purpose: a recording from before C219 was never classified,
+      // and stamping it `gazetteer` would claim a pass that never ran. Null is
+      // the honest "não classificada" until the backfill CLI runs.
+      index: true,
+      options: SPEECH_CLASSIFICATION_SOURCES.map((value) => ({
+        value,
+        label: value === 'gazetteer' ? 'Gazetteer' : value === 'llm' ? 'LLM' : 'Manual',
+      })),
+      admin: {
+        description:
+          'Automática e auditável: Gazetteer (léxico), LLM (refinamento validado) ou Manual (correção humana).',
+      },
+    },
+    {
+      name: 'mentionedMunicipalities',
+      type: 'relationship',
+      relationTo: 'municipality',
+      label: 'Municípios citados',
+      hasMany: true,
+      index: true,
+    },
     {
       name: 'speakerLabels',
       type: 'array',

src/components/campaign/recording/RecordingAcervoFilters.tsx

@@ -0,0 +1,296 @@
+'use client'
+
+import { useMemo, useOptimistic, useState } from 'react'
+
+import {
+  CampaignHeaderFilterPopover,
+  type CampaignHeaderFilterRow,
+} from '@/components/campaign/shared/CampaignHeaderFilterPopover'
+import {
+  CampaignListOmnibox,
+  campaignListOmniboxFormClassName,
+} from '@/components/campaign/shared/CampaignListOmnibox'
+import { CampaignSearchModeControl } from '@/components/campaign/shared/CampaignSearchModeControl'
+import { useCampaignListFilterNavigation } from '@/components/campaign/shared/useCampaignListFilterNavigation'
+import { selectedFilterTriggerLabel } from '@/lib/campaignListOmnibox'
+import { SPEECH_SCOPES, SPEECH_TOPICS } from '@/lib/speechFacets'
+import {
+  buildRecordingListHref,
+  toggleRecordingPerson,
+  type RecordingFilterOptions,
+  type RecordingListState,
+} from '@/utilities/recordings/recordingListUrl'
+import {
+  applyRecordingOmniboxSuggestion,
+  applyRecordingSearchMode,
+  buildRecordingOmniboxChips,
+  buildRecordingOmniboxSuggestionSeeds,
+  clearRecordingOmnibox,
+  filterRecordingOmniboxSuggestions,
+  removeRecordingOmniboxChip,
+  type RecordingOmniboxAction,
+} from '@/utilities/recordings/recordingOmnibox'
+import {
+  SPEECH_DURATION_BUCKETS,
+  speechDurationLabels,
+  speechScopeLabels,
+  speechTopicLabels,
+  type SpeechSearchMode,
+} from '@/utilities/speech/speechListUrl'
+
+const topicOptions = SPEECH_TOPICS.map(({ value, label }) => ({ value, label }))
+const scopeOptions = SPEECH_SCOPES.map(({ value, label }) => ({ value, label }))
+const durationOptions = SPEECH_DURATION_BUCKETS.map(({ value, label }) => ({ value, label }))
+
+type FacetKey = 'years' | 'topics' | 'scopes' | 'municipalities' | 'durations' | 'people'
+
+type RecordingFacet = {
+  id: string
+  label: string
+  /** Prefix of the omnibox suggestion id (`${prefix}:${value}`). */
+  prefix: string
+  stateKey: FacetKey
+  options: readonly { value: string; label: string }[]
+  selected: readonly string[]
+  labels?: Readonly<Record<string, string>>
+}
+
+/**
+ * C219 — the filter bar of "Gravações enviadas": same gestures as the Câmara
+ * acervo (search by term or theme, year, topic, scope, cited municipality,
+ * duration) plus the exclusive "Pessoa" facet. The mechanics (omnibox, chips,
+ * facet popovers, pending boundary) are the shared shells; this component only
+ * assembles the recordings facets. Ported from the approved hi-fi design
+ * (`docs/plans/acervo-paridade-busca-filtros-ui-design.html`, scenes 01/02).
+ */
+export const RecordingAcervoFilters = ({
+  state,
+  filterOptions,
+  themeUnavailable = false,
+}: {
+  state: RecordingListState
+  filterOptions: RecordingFilterOptions
+  /** C219 — the theme expansion is down; the selector reflects the fallback. */
+  themeUnavailable?: boolean
+}) => {
+  const { navigate, isPending } = useCampaignListFilterNavigation({
+    state,
+    toHref: (next) => buildRecordingListHref(next, 1),
+  })
+  const [query, setQuery] = useState('')
+  const [viewState, setOptimisticState] = useOptimistic(state)
+
+  const activeMode: SpeechSearchMode =
+    viewState.mode === 'tema' && !themeUnavailable ? 'tema' : 'termo'
+
+  const municipalityLabelsById = useMemo(() => {
+    const map = new Map<number, string>()
+    for (const option of filterOptions.municipalities) {
+      const id = Number(option.value)
+      if (Number.isSafeInteger(id) && id > 0) map.set(id, option.label)
+    }
+    return map
+  }, [filterOptions.municipalities])
+
+  const chips = useMemo(
+    () => buildRecordingOmniboxChips({ state: viewState, municipalityLabelsById }),
+    [viewState, municipalityLabelsById],
+  )
+
+  const suggestionSeeds = useMemo(
+    () =>
+      buildRecordingOmniboxSuggestionSeeds({
+        years: filterOptions.years,
+        municipalityOptions: filterOptions.municipalities,
+        personOptions: filterOptions.people,
+      }),
+    [filterOptions],
+  )
+
+  const suggestions = useMemo(
+    () => filterRecordingOmniboxSuggestions(suggestionSeeds, query),
+    [suggestionSeeds, query],
+  )
+
+  const runAction = (action: RecordingOmniboxAction) => {
+    if (action.kind === 'clear') {
+      setQuery('')
+      navigate(action.state)
+      return
+    }
+    navigate(action.state)
+  }
+
+  const runMode = (mode: SpeechSearchMode) => {
+    const next = applyRecordingSearchMode({ state: viewState, mode }).state
+    setOptimisticState(next)
+    navigate(next)
+  }
+
+  const facetRows = (
+    prefix: string,
+    options: readonly { value: string; label: string }[],
+    selected: readonly string[],
+  ): CampaignHeaderFilterRow[] =>
+    options.map((option) => {
+      const next = applyRecordingOmniboxSuggestion({
+        state: viewState,
+        suggestionId: `${prefix}:${option.value}`,
+      }).state
+      return {
+        value: option.value,
+        label: option.label,
+        href: buildRecordingListHref(next, 1),
+        selected: selected.includes(option.value),
+        checkbox: true,
+        onChoose: () => setOptimisticState(next),
+      }
+    })
+
+  const personRows = (): CampaignHeaderFilterRow[] =>
+    filterOptions.people.map((person) => {
+      const next = toggleRecordingPerson(viewState, person)
+      return {
+        value: person,
+        label: person,
+        href: buildRecordingListHref(next, 1),
+        selected: (viewState.people ?? []).some(
+          (selected) => selected.toLocaleLowerCase('pt-BR') === person.toLocaleLowerCase('pt-BR'),
+        ),
+        checkbox: true,
+        onChoose: () => setOptimisticState(next),
+      }
+    })
+
+  const clearFacet = (facet: FacetKey) => {
+    const next: RecordingListState = { ...viewState, page: 1, [facet]: undefined }
+    return { href: buildRecordingListHref(next, 1), onChoose: () => setOptimisticState(next) }
+  }
+
+  const years = (viewState.years ?? []).map(String)
+  const topics = viewState.topics ?? []
+  const scopes = viewState.scopes ?? []
+  const municipalities = (viewState.municipalities ?? []).map(String)
+  const durations = viewState.durations ?? []
+  const people = viewState.people ?? []
+
+  const facets: RecordingFacet[] = [
+    {
+      id: 'recording-filter-year',
+      label: 'Ano',
+      prefix: 'year',
+      stateKey: 'years',
+      options: filterOptions.years.map((year) => ({ value: String(year), label: String(year) })),
+      selected: years,
+    },
+    {
+      id: 'recording-filter-topic',
+      label: 'Tema',
+      prefix: 'topic',
+      stateKey: 'topics',
+      options: topicOptions,
+      selected: topics,
+      labels: speechTopicLabels,
+    },
+    {
+      id: 'recording-filter-scope',
+      label: 'Alcance',
+      prefix: 'scope',
+      stateKey: 'scopes',
+      options: scopeOptions,
+      selected: scopes,
+      labels: speechScopeLabels,
+    },
+    {
+      id: 'recording-filter-municipality',
+      label: 'Município citado',
+      prefix: 'municipality',
+      stateKey: 'municipalities',
+      options: filterOptions.municipalities,
+      selected: municipalities,
+      labels: Object.fromEntries(municipalityLabelsById),
+    },
+    {
+      id: 'recording-filter-duration',
+      label: 'Duração',
+      prefix: 'duration',
+      stateKey: 'durations',
+      options: durationOptions,
+      selected: durations,
+      labels: speechDurationLabels,
+    },
+  ]
+
+  return (
+    <form
+      role="search"
+      className={campaignListOmniboxFormClassName}
+      onSubmit={(event) => {
+        event.preventDefault()
+      }}
+    >
+      <CampaignListOmnibox
+        id="recording-omnibox"
+        label="Buscar nas gravações enviadas"
+        placeholder="Busque por palavra, trecho ou tema da transcrição…"
+        chips={chips}
+        suggestions={suggestions}
+        query={query}
+        onQueryChange={setQuery}
+        isPending={isPending}
+        onSelectSuggestion={(suggestionId) => {
+          runAction(applyRecordingOmniboxSuggestion({ state: viewState, suggestionId }))
+        }}
+        onCommitQuery={(text) => {
+          runAction(
+            applyRecordingOmniboxSuggestion({ state: viewState, suggestionId: `q:${text}` }),
+          )
+        }}
+        onRemoveChip={(chipId) => {
+          runAction(removeRecordingOmniboxChip({ state: viewState, chipId }))
+        }}
+        onClearAll={() => {
+          runAction(clearRecordingOmnibox(viewState))
+        }}
+      />
+
+      <CampaignSearchModeControl
+        activeMode={activeMode}
+        themeUnavailable={themeUnavailable}
+        onSelect={runMode}
+        relatedHint="Encontra gravações relacionadas pelo sentido."
+      />
+
+      <div
+        role="group"
+        aria-label="Filtros do acervo de gravações"
+        className="mt-2 flex gap-2 overflow-x-auto pb-1 md:flex-wrap md:overflow-visible"
+      >
+        {facets.map((facet) => (
+          <CampaignHeaderFilterPopover
+            key={facet.id}
+            id={facet.id}
+            label={facet.label}
+            triggerVariant="chip"
+            triggerLabel={selectedFilterTriggerLabel(facet.label, facet.selected, facet.labels)}
+            active={facet.selected.length > 0}
+            closeOnChoose={false}
+            optionRows={facetRows(facet.prefix, facet.options, facet.selected)}
+            clear={facet.selected.length ? clearFacet(facet.stateKey) : undefined}
+          />
+        ))}
+        <CampaignHeaderFilterPopover
+          id="recording-filter-person"
+          label="Pessoa"
+          triggerVariant="chip"
+          triggerLabel={selectedFilterTriggerLabel('Pessoa', people)}
+          active={people.length > 0}
+          closeOnChoose={false}
+          emptyLabel="Nenhuma pessoa identificada nas gravações."
+          optionRows={personRows()}
+          clear={people.length ? clearFacet('people') : undefined}
+        />
+      </div>
+    </form>
+  )
+}

src/components/campaign/recording/RecordingPersonFilter.tsx

@@ -1,87 +0,0 @@
-'use client'
-
-import { FunnelIcon, XIcon } from 'lucide-react'
-import { useOptimistic } from 'react'
-
-import {
-  CampaignHeaderFilterPopover,
-  type CampaignHeaderFilterRow,
-} from '@/components/campaign/shared/CampaignHeaderFilterPopover'
-import { useCampaignListFilterNavigation } from '@/components/campaign/shared/useCampaignListFilterNavigation'
-import { foldSpeakerName } from '@/lib/recordingDiarization'
-import {
-  buildRecordingListHref,
-  removeRecordingPerson,
-  toggleRecordingPerson,
-  type RecordingListState,
-} from '@/utilities/recordings/recordingListUrl'
-
-const PERSON_FILTER_ID = 'recording-filter-person'
-
-/**
- * C200 — the "Pessoa" facet of the uploaded recordings (design scene 3): one
- * removable chip per selected label plus the multi-select popover backed by the
- * labels actually present. Options come from the server (`speakerNames`), never
- * from a free-form directory — the label is text, not a `Contact`.
- */
-export const RecordingPersonFilter = ({
-  state,
-  options,
-}: {
-  state: RecordingListState
-  options: readonly string[]
-}) => {
-  const { navigate } = useCampaignListFilterNavigation({
-    state,
-    toHref: (next) => buildRecordingListHref(next, 1),
-  })
-  const [viewState, setOptimisticState] = useOptimistic(state)
-  const selected = viewState.people ?? []
-
-  const optionRows: CampaignHeaderFilterRow[] = options.map((option) => {
-    const next = toggleRecordingPerson(viewState, option)
-    return {
-      value: option,
-      label: option,
-      href: buildRecordingListHref(next, 1),
-      selected: selected.some((person) => foldSpeakerName(person) === foldSpeakerName(option)),
-      checkbox: true,
-      onChoose: () => setOptimisticState(next),
-    }
-  })
-
-  const clear = {
-    href: buildRecordingListHref({ ...viewState, page: 1, people: undefined }, 1),
-    onChoose: () => setOptimisticState({ ...viewState, page: 1, people: undefined }),
-  }
-
-  return (
-    <div role="group" aria-label="Filtrar por pessoa" className="flex flex-wrap items-center gap-2">
-      <CampaignHeaderFilterPopover
-        id={PERSON_FILTER_ID}
-        label="Pessoa"
-        triggerVariant="chip"
-        triggerLabel={selected.length > 1 ? `Pessoa: ${selected.length}` : 'Pessoa'}
-        active={selected.length > 0}
-        closeOnChoose={false}
-        emptyLabel="Nenhuma pessoa identificada nas gravações."
-        optionRows={optionRows}
-        clear={selected.length ? clear : undefined}
-      />
-
-      {selected.map((person) => (
-        <button
-          key={person}
-          type="button"
-          className="inline-flex h-8 min-h-8 items-center gap-1.5 rounded-lg border border-input bg-white px-3 text-xs font-semibold focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring"
-          onClick={() => navigate(removeRecordingPerson(viewState, person))}
-          aria-label={`Remover filtro de ${person}`}
-        >
-          <FunnelIcon className="size-3.5 shrink-0" aria-hidden="true" />
-          <span className="max-w-56 truncate">Pessoa: {person}</span>
-          <XIcon className="size-3.5 shrink-0" aria-hidden="true" />
-        </button>
-      ))}
-    </div>
-  )
-}

src/components/campaign/recording/RecordingResultCard.tsx

@@ -66,8 +66,16 @@ export const RecordingResultCard = ({ recording }: { recording: RecordingListIte
           </p>
         )}
 
-        {recording.matchedPersons.length > 0 ? (
+        {recording.topics.length > 0 || recording.matchedPersons.length > 0 ? (
           <div className="mt-2 flex flex-wrap items-center gap-2 text-xs">
+            {recording.topics.map((topic) => (
+              <span
+                key={topic.value}
+                className="inline-flex min-h-7 items-center rounded-md bg-muted px-2 py-1 text-xs font-semibold"
+              >
+                {topic.label}
+              </span>
+            ))}
             {recording.matchedPersons.map((person) => (
               <span
                 key={person}
@@ -77,7 +85,9 @@ export const RecordingResultCard = ({ recording }: { recording: RecordingListIte
                 {person}
               </span>
             ))}
-            <span className="text-muted-foreground">aparece nesta gravação</span>
+            {recording.matchedPersons.length > 0 ? (
+              <span className="text-muted-foreground">aparece nesta gravação</span>
+            ) : null}
           </div>
         ) : null}
       </div>

src/components/campaign/recording/RecordingSortSelect.tsx

@@ -0,0 +1,51 @@
+'use client'
+
+import { useCampaignListFilterNavigation } from '@/components/campaign/shared/useCampaignListFilterNavigation'
+import {
+  buildRecordingListHref,
+  RECORDING_SORT_OPTIONS,
+  recordingSortIsDuration,
+  type RecordingListState,
+  type RecordingSortKey,
+} from '@/utilities/recordings/recordingListUrl'
+
+/**
+ * C219 — the list ordering control (design scene 01): "Mais recentes" (the
+ * default, never serialized) plus the two duration orders. A duration order
+ * only lists rows with a measured duration, so the hint says it out loud.
+ */
+export const RecordingSortSelect = ({ state }: { state: RecordingListState }) => {
+  const { navigate, isPending } = useCampaignListFilterNavigation({
+    state,
+    toHref: (next) => buildRecordingListHref(next, 1),
+  })
+  const value: RecordingSortKey = state.sort ?? 'recentes'
+
+  return (
+    <div className="flex flex-col items-end gap-1">
+      <label className="flex items-center gap-2 text-xs text-muted-foreground">
+        Ordenar por
+        <select
+          value={value}
+          disabled={isPending}
+          className="min-h-10 rounded-md border border-input bg-white px-3 py-2 text-sm text-foreground disabled:opacity-70"
+          onChange={(event) => {
+            const next = event.target.value as RecordingSortKey
+            navigate({ ...state, page: 1, sort: next === 'recentes' ? undefined : next })
+          }}
+        >
+          {RECORDING_SORT_OPTIONS.map((option) => (
+            <option key={option.value} value={option.value}>
+              {option.label}
+            </option>
+          ))}
+        </select>
+      </label>
+      {recordingSortIsDuration(state) ? (
+        <p className="text-xs text-muted-foreground">
+          Gravações sem duração aparecem apenas em Mais recentes.
+        </p>
+      ) : null}
+    </div>
+  )
+}

src/components/campaign/recording/RecordingThemeFallbackNotice.tsx

@@ -0,0 +1,26 @@
+import { TriangleAlertIcon } from 'lucide-react'
+
+import { Alert, AlertDescription, AlertTitle } from '@/components/ui/Alert'
+
+/**
+ * C219 — the discreet fallback notice when `mode=tema` was asked on the
+ * recordings source but the theme expansion was unavailable. The list below
+ * shows the literal results; the notice keeps the user's request honest
+ * instead of hiding the downgrade (same contract as the Câmara acervo, C192).
+ */
+export const RecordingThemeFallbackNotice = () => (
+  <Alert
+    variant="pending"
+    data-testid="recording-theme-fallback"
+    className="gap-3 border-estimate-pending/80 bg-estimate-pending/55 px-4 py-3"
+  >
+    <TriangleAlertIcon aria-hidden="true" />
+    <AlertTitle className="text-sm font-medium">
+      A busca por tema está indisponível agora.
+    </AlertTitle>
+    <AlertDescription className="text-xs leading-5">
+      Você pediu “Por tema”. Para não interromper o trabalho, mostramos abaixo os resultados por
+      termo exato.
+    </AlertDescription>
+  </Alert>
+)

src/components/campaign/shared/CampaignSearchModeControl.tsx

@@ -0,0 +1,83 @@
+'use client'
+
+import { SparklesIcon, TriangleAlertIcon } from 'lucide-react'
+
+import { cn } from '@/lib/utils'
+import type { SpeechSearchMode } from '@/utilities/speech/speechListUrl'
+
+/** C192/C219 — segmented search-mode button (mobile full width, desktop inline). */
+const modeButtonClass = (active: boolean, tone: 'termo' | 'tema'): string =>
+  cn(
+    'min-h-11 rounded-md px-3 text-sm md:min-h-9',
+    active
+      ? cn(
+          'bg-white font-semibold shadow-sm ring-1 ring-border',
+          tone === 'tema' ? 'text-primary' : 'text-foreground',
+        )
+      : 'font-medium text-muted-foreground',
+  )
+
+/**
+ * C219 — the "Termo exato | Por tema" selector shared by the two acervo
+ * sources. The mechanics (mode semantics, degradation) belong to the URL
+ * contract and the loader; this component only renders the choice and the
+ * honest state of the theme bridge.
+ */
+export const CampaignSearchModeControl = ({
+  activeMode,
+  themeUnavailable = false,
+  onSelect,
+  relatedHint,
+}: {
+  activeMode: SpeechSearchMode
+  /** The theme expansion is down; the selector reflects the fallback. */
+  themeUnavailable?: boolean
+  onSelect: (mode: SpeechSearchMode) => void
+  /** Desktop hint of the corpus ("Encontra falas relacionadas pelo sentido."). */
+  relatedHint: string
+}) => (
+  <div className="mt-3 flex flex-col gap-3 border-t border-border pt-3 md:flex-row md:flex-wrap md:items-center md:justify-between">
+    <div className="flex w-full items-center gap-2 md:w-auto">
+      <span className="hidden text-xs font-medium text-muted-foreground md:inline">Buscar por</span>
+      <div
+        role="group"
+        aria-label="Modo de busca"
+        className="grid w-full grid-cols-2 rounded-lg bg-muted p-1 md:inline-flex md:w-auto"
+      >
+        <button
+          type="button"
+          aria-pressed={activeMode === 'termo'}
+          className={modeButtonClass(activeMode === 'termo', 'termo')}
+          onClick={() => onSelect('termo')}
+        >
+          Termo exato
+        </button>
+        <button
+          type="button"
+          aria-pressed={activeMode === 'tema'}
+          aria-disabled={themeUnavailable || undefined}
+          className={modeButtonClass(activeMode === 'tema', 'tema')}
+          onClick={() => {
+            if (themeUnavailable) return
+            onSelect('tema')
+          }}
+        >
+          <span className="inline-flex items-center gap-1.5">
+            {themeUnavailable ? (
+              <TriangleAlertIcon
+                className="size-3.5 text-estimate-pending-foreground"
+                aria-hidden="true"
+              />
+            ) : (
+              <SparklesIcon className="size-3.5" aria-hidden="true" />
+            )}
+            Por tema
+          </span>
+        </button>
+      </div>
+    </div>
+    <p className="hidden text-xs text-muted-foreground md:block">
+      {themeUnavailable ? 'Indisponível agora; mostramos o termo exato.' : relatedHint}
+    </p>
+  </div>
+)

src/components/campaign/shared/CampaignThemeRetryButton.tsx

@@ -5,8 +5,11 @@ import { useTransition } from 'react'
 
 import { Button } from '@/components/ui/button'
 
-/** C192 — re-runs the same `mode=tema` request after a degraded fallback. */
-export const SpeechThemeRetryButton = () => {
+/**
+ * C192/C219 — re-runs the same `mode=tema` request after a degraded fallback.
+ * Shared by both acervo sources; the notice copy stays per domain.
+ */
+export const CampaignThemeRetryButton = () => {
   const router = useRouter()
   const [isPending, startTransition] = useTransition()
 

src/components/campaign/speech/SpeechAcervoFilters.tsx

@@ -1,6 +1,5 @@
 'use client'
 
-import { SparklesIcon, TriangleAlertIcon } from 'lucide-react'
 import { useMemo, useOptimistic, useState } from 'react'
 
 import {
@@ -11,9 +10,10 @@ import {
   CampaignListOmnibox,
   campaignListOmniboxFormClassName,
 } from '@/components/campaign/shared/CampaignListOmnibox'
+import { CampaignSearchModeControl } from '@/components/campaign/shared/CampaignSearchModeControl'
 import { useCampaignListFilterNavigation } from '@/components/campaign/shared/useCampaignListFilterNavigation'
+import { selectedFilterTriggerLabel } from '@/lib/campaignListOmnibox'
 import { SPEECH_SCOPES, SPEECH_TOPICS } from '@/lib/speechFacets'
-import { cn } from '@/lib/utils'
 import {
   buildSpeechListHref,
   SPEECH_DURATION_BUCKETS,
@@ -53,28 +53,6 @@ type SpeechFacet = {
   labels?: Readonly<Record<string, string>>
 }
 
-const selectedTriggerLabel = (
-  base: string,
-  values: readonly string[],
-  labels?: Readonly<Record<string, string>>,
-): string => {
-  if (values.length === 0) return base
-  if (values.length === 1) return `${base}: ${labels?.[values[0]!] ?? values[0]}`
-  return `${base}: ${values.length}`
-}
-
-/** C192 — segmented search-mode button (mobile full width, desktop inline). */
-const modeButtonClass = (active: boolean, tone: 'termo' | 'tema'): string =>
-  cn(
-    'min-h-11 rounded-md px-3 text-sm md:min-h-9',
-    active
-      ? cn(
-          'bg-white font-semibold shadow-sm ring-1 ring-border',
-          tone === 'tema' ? 'text-primary' : 'text-foreground',
-        )
-      : 'font-medium text-muted-foreground',
-  )
-
 export const SpeechAcervoFilters = ({
   state,
   filterOptions,
@@ -257,54 +235,12 @@ export const SpeechAcervoFilters = ({
         }}
       />
 
-      <div className="mt-3 flex flex-col gap-3 border-t border-border pt-3 md:flex-row md:flex-wrap md:items-center md:justify-between">
-        <div className="flex w-full items-center gap-2 md:w-auto">
-          <span className="hidden text-xs font-medium text-muted-foreground md:inline">
-            Buscar por
-          </span>
-          <div
-            role="group"
-            aria-label="Modo de busca"
-            className="grid w-full grid-cols-2 rounded-lg bg-muted p-1 md:inline-flex md:w-auto"
-          >
-            <button
-              type="button"
-              aria-pressed={activeMode === 'termo'}
-              className={modeButtonClass(activeMode === 'termo', 'termo')}
-              onClick={() => runMode('termo')}
-            >
-              Termo exato
-            </button>
-            <button
-              type="button"
-              aria-pressed={activeMode === 'tema'}
-              aria-disabled={themeUnavailable || undefined}
-              className={modeButtonClass(activeMode === 'tema', 'tema')}
-              onClick={() => {
-                if (themeUnavailable) return
-                runMode('tema')
-              }}
-            >
-              <span className="inline-flex items-center gap-1.5">
-                {themeUnavailable ? (
-                  <TriangleAlertIcon
-                    className="size-3.5 text-estimate-pending-foreground"
-                    aria-hidden="true"
-                  />
-                ) : (
-                  <SparklesIcon className="size-3.5" aria-hidden="true" />
-                )}
-                Por tema
-              </span>
-            </button>
-          </div>
-        </div>
-        <p className="hidden text-xs text-muted-foreground md:block">
-          {themeUnavailable
-            ? 'Indisponível agora; mostramos o termo exato.'
-            : 'Encontra falas relacionadas pelo sentido.'}
-        </p>
-      </div>
+      <CampaignSearchModeControl
+        activeMode={activeMode}
+        themeUnavailable={themeUnavailable}
+        onSelect={runMode}
+        relatedHint="Encontra falas relacionadas pelo sentido."
+      />
 
       <div
         role="group"
@@ -317,7 +253,7 @@ export const SpeechAcervoFilters = ({
             id={facet.id}
             label={facet.label}
             triggerVariant="chip"
-            triggerLabel={selectedTriggerLabel(facet.label, facet.selected, facet.labels)}
+            triggerLabel={selectedFilterTriggerLabel(facet.label, facet.selected, facet.labels)}
             active={facet.selected.length > 0}
             closeOnChoose={false}
             optionRows={facetRows(facet.prefix, facet.options, facet.selected)}

src/lib/campaignListOmnibox.ts

@@ -18,6 +18,21 @@ export type CampaignListOmniboxSuggestion = {
   keywords?: readonly string[]
 }
 
+/**
+ * The trigger text of a facet chip: the dimension alone when nothing is
+ * selected, "Ano: 2026" for one value and "Tema: 3" for several — one spelling
+ * for every acervo filter bar (C154/C219).
+ */
+export const selectedFilterTriggerLabel = (
+  base: string,
+  values: readonly string[],
+  labels?: Readonly<Record<string, string>>,
+): string => {
+  if (values.length === 0) return base
+  if (values.length === 1) return `${base}: ${labels?.[values[0]!] ?? values[0]}`
+  return `${base}: ${values.length}`
+}
+
 const omniboxQueryMatches = (
   haystack: string,
   needle: string,

src/migrations/20260924_040210_add_recording_facets.ts

@@ -0,0 +1,76 @@
+import { MigrateUpArgs, MigrateDownArgs, sql } from '@payloadcms/db-postgres'
+
+/**
+ * C219 — facet + provenance columns of the uploaded recordings, mirroring the
+ * speech catalog (C153): `year` derived from the informed recording date,
+ * `topics`/`scopes` enums, `classifiedBy` (nullable: a row from before C219 was
+ * never classified) and the `mentionedMunicipalities` join. Additive only.
+ *
+ * The `year` backfill is deterministic SQL (never an LLM pass in a migration):
+ * it derives exactly what the `deriveRecordingYear` hook derives on the next
+ * write, so old rows join the "Ano" facet before the classification CLI runs.
+ */
+export async function up({ db, payload, req }: MigrateUpArgs): Promise<void> {
+  await db.execute(sql`
+   CREATE TYPE "public"."enum_recording_topics" AS ENUM('saude', 'educacao', 'cultura', 'esporte', 'seguranca-publica', 'meio-ambiente', 'economia-trabalho', 'direitos-humanos', 'infraestrutura', 'ciencia-tecnologia', 'politica-instituicoes', 'agricultura', 'habitacao-cidades', 'comunicacao-midia', 'igualdade-racial', 'mulheres-genero', 'juventude', 'pessoa-deficiencia');
+  CREATE TYPE "public"."enum_recording_scopes" AS ENUM('bahia', 'brasil', 'internacional');
+  CREATE TYPE "public"."enum_recording_classified_by" AS ENUM('gazetteer', 'llm', 'manual');
+  CREATE TABLE "recording_topics" (
+  	"order" integer NOT NULL,
+  	"parent_id" integer NOT NULL,
+  	"value" "enum_recording_topics",
+  	"id" serial PRIMARY KEY NOT NULL
+  );
+  
+  CREATE TABLE "recording_scopes" (
+  	"order" integer NOT NULL,
+  	"parent_id" integer NOT NULL,
+  	"value" "enum_recording_scopes",
+  	"id" serial PRIMARY KEY NOT NULL
+  );
+  
+  CREATE TABLE "recording_rels" (
+  	"id" serial PRIMARY KEY NOT NULL,
+  	"order" integer,
+  	"parent_id" integer NOT NULL,
+  	"path" varchar NOT NULL,
+  	"municipality_id" integer
+  );
+  
+  ALTER TABLE "recording" ADD COLUMN "year" numeric;
+  ALTER TABLE "recording" ADD COLUMN "classified_by" "enum_recording_classified_by";
+  ALTER TABLE "recording_topics" ADD CONSTRAINT "recording_topics_parent_fk" FOREIGN KEY ("parent_id") REFERENCES "public"."recording"("id") ON DELETE cascade ON UPDATE no action;
+  ALTER TABLE "recording_scopes" ADD CONSTRAINT "recording_scopes_parent_fk" FOREIGN KEY ("parent_id") REFERENCES "public"."recording"("id") ON DELETE cascade ON UPDATE no action;
+  ALTER TABLE "recording_rels" ADD CONSTRAINT "recording_rels_parent_fk" FOREIGN KEY ("parent_id") REFERENCES "public"."recording"("id") ON DELETE cascade ON UPDATE no action;
+  ALTER TABLE "recording_rels" ADD CONSTRAINT "recording_rels_municipality_fk" FOREIGN KEY ("municipality_id") REFERENCES "public"."municipality"("id") ON DELETE cascade ON UPDATE no action;
+  CREATE INDEX "recording_topics_order_idx" ON "recording_topics" USING btree ("order");
+  CREATE INDEX "recording_topics_parent_idx" ON "recording_topics" USING btree ("parent_id");
+  CREATE INDEX "recording_topics_value_idx" ON "recording_topics" USING btree ("value");
+  CREATE INDEX "recording_scopes_order_idx" ON "recording_scopes" USING btree ("order");
+  CREATE INDEX "recording_scopes_parent_idx" ON "recording_scopes" USING btree ("parent_id");
+  CREATE INDEX "recording_scopes_value_idx" ON "recording_scopes" USING btree ("value");
+  CREATE INDEX "recording_rels_order_idx" ON "recording_rels" USING btree ("order");
+  CREATE INDEX "recording_rels_parent_idx" ON "recording_rels" USING btree ("parent_id");
+  CREATE INDEX "recording_rels_path_idx" ON "recording_rels" USING btree ("path");
+  CREATE INDEX "recording_rels_municipality_id_idx" ON "recording_rels" USING btree ("municipality_id");
+   CREATE INDEX "recording_year_idx" ON "recording" USING btree ("year");
+   CREATE INDEX "recording_classified_by_idx" ON "recording" USING btree ("classified_by");
+   UPDATE "recording" SET "year" = EXTRACT(YEAR FROM "recorded_at") WHERE "recorded_at" IS NOT NULL;`)
+}
+
+export async function down({ db, payload, req }: MigrateDownArgs): Promise<void> {
+  await db.execute(sql`
+   ALTER TABLE "recording_topics" DISABLE ROW LEVEL SECURITY;
+  ALTER TABLE "recording_scopes" DISABLE ROW LEVEL SECURITY;
+  ALTER TABLE "recording_rels" DISABLE ROW LEVEL SECURITY;
+  DROP TABLE "recording_topics" CASCADE;
+  DROP TABLE "recording_scopes" CASCADE;
+  DROP TABLE "recording_rels" CASCADE;
+  DROP INDEX "recording_year_idx";
+  DROP INDEX "recording_classified_by_idx";
+  ALTER TABLE "recording" DROP COLUMN "year";
+  ALTER TABLE "recording" DROP COLUMN "classified_by";
+  DROP TYPE "public"."enum_recording_topics";
+  DROP TYPE "public"."enum_recording_scopes";
+  DROP TYPE "public"."enum_recording_classified_by";`)
+}

src/migrations/index.ts

@@ -82,6 +82,7 @@ import * as migration_20260923_032714_add_content_piece from './20260923_032714_
 import * as migration_20260923_054145_add_share_link_announcement from './20260923_054145_add_share_link_announcement';
 import * as migration_20260923_172437_add_content_events from './20260923_172437_add_content_events';
 import * as migration_20260923_190414_add_content_event_variant from './20260923_190414_add_content_event_variant';
+import * as migration_20260924_040210_add_recording_facets from './20260924_040210_add_recording_facets';
 import * as migration_20260924_042028_add_web_speech_origin from './20260924_042028_add_web_speech_origin';
 import * as migration_20260924_045625_add_content_piece_link_failure_reason from './20260924_045625_add_content_piece_link_failure_reason';
 
@@ -506,6 +507,11 @@ export const migrations = [
     down: migration_20260923_190414_add_content_event_variant.down,
     name: '20260923_190414_add_content_event_variant',
   },
+  {
+    up: migration_20260924_040210_add_recording_facets.up,
+    down: migration_20260924_040210_add_recording_facets.down,
+    name: '20260924_040210_add_recording_facets'
+  },
   {
     up: migration_20260924_042028_add_web_speech_origin.up,
     down: migration_20260924_042028_add_web_speech_origin.down,

src/payload-types.ts

@@ -1362,7 +1362,7 @@ export interface ContentEvent {
   createdAt: string;
 }
 /**
- * Gravações próprias da equipe no acervo. O arquivo é privado; a transcrição é somente leitura.
+ * Gravações próprias da equipe no acervo. O arquivo é privado; a transcrição é somente leitura. Tema, alcance e municípios citados são classificados automaticamente pelo início da transcrição e podem ser corrigidos aqui (a correção vira proveniência "Manual").
  *
  * This interface was referenced by `Config`'s JSON-Schema
  * via the `definition` "recording".
@@ -1374,6 +1374,10 @@ export interface Recording {
    * Data em que a gravação foi feita (opcional).
    */
   recordedAt?: string | null;
+  /**
+   * Derivado da data da gravação.
+   */
+  year?: number | null;
   status: 'uploading' | 'processing' | 'ready' | 'failed';
   /**
    * Progresso honesto da transcrição em andamento.
@@ -1391,6 +1395,34 @@ export interface Recording {
    * Concatenação normalizada dos segmentos (sem acentos, minúsculas).
    */
   searchText?: string | null;
+  topics?:
+    | (
+        | 'saude'
+        | 'educacao'
+        | 'cultura'
+        | 'esporte'
+        | 'seguranca-publica'
+        | 'meio-ambiente'
+        | 'economia-trabalho'
+        | 'direitos-humanos'
+        | 'infraestrutura'
+        | 'ciencia-tecnologia'
+        | 'politica-instituicoes'
+        | 'agricultura'
+        | 'habitacao-cidades'
+        | 'comunicacao-midia'
+        | 'igualdade-racial'
+        | 'mulheres-genero'
+        | 'juventude'
+        | 'pessoa-deficiencia'
+      )[]
+    | null;
+  scopes?: ('bahia' | 'brasil' | 'internacional')[] | null;
+  /**
+   * Automática e auditável: Gazetteer (léxico), LLM (refinamento validado) ou Manual (correção humana).
+   */
+  classifiedBy?: ('gazetteer' | 'llm' | 'manual') | null;
+  mentionedMunicipalities?: (number | Municipality)[] | null;
   /**
    * Rótulos humanos por agrupamento acústico (chave → nome). Identificação sempre humana; o acervo nunca sugere ou infere pessoas.
    */
@@ -2872,11 +2904,16 @@ export interface ContentEventSelect<T extends boolean = true> {
 export interface RecordingSelect<T extends boolean = true> {
   title?: T;
   recordedAt?: T;
+  year?: T;
   status?: T;
   step?: T;
   media?: T;
   durationSeconds?: T;
   searchText?: T;
+  topics?: T;
+  scopes?: T;
+  classifiedBy?: T;
+  mentionedMunicipalities?: T;
   speakerLabels?:
     | T
     | {

src/utilities/ai/expandSpeechSearchTheme.ts

@@ -15,17 +15,23 @@ const EXPAND_TIMEOUT_MS = 4000
 const MAX_OUTPUT_TOKENS = 200
 const MAX_THEME_LENGTH = 200
 
+const EXPAND_SUBJECT_BY_CORPUS: Record<ThemeSearchCorpus, string> = {
+  speech: 'na transcrição de falas',
+  contentPiece: 'na descrição ou na transcrição de uma peça',
+  recording: 'na transcrição de uma gravação enviada pela equipe',
+}
+
 const expandSchemaFor = (corpus: ThemeSearchCorpus) =>
   z.object({
     terms: z
       .array(z.string())
       .describe(
-        `Termos e expressões curtas (${MIN_SPEECH_THEME_TERMS} a ${MAX_SPEECH_THEME_TERMS}) que podem aparecer ${corpus === 'speech' ? 'na transcrição de falas' : 'na descrição ou na transcrição de uma peça'} sobre o tema. Lista vazia quando não houver termo confiável.`,
+        `Termos e expressões curtas (${MIN_SPEECH_THEME_TERMS} a ${MAX_SPEECH_THEME_TERMS}) que podem aparecer ${EXPAND_SUBJECT_BY_CORPUS[corpus]} sobre o tema. Lista vazia quando não houver termo confiável.`,
       ),
   })
 
 /** The term rules are one contract; only the noun of the corpus changes. */
-const themeExpansionRules = (subject: 'discurso' | 'peça'): string =>
+const themeExpansionRules = (subject: 'discurso' | 'peça' | 'gravação'): string =>
   `Devolva de ${MIN_SPEECH_THEME_TERMS} a ${MAX_SPEECH_THEME_TERMS} termos, cada um com no máximo ${MAX_SPEECH_THEME_TERM_LENGTH} caracteres, sem explicações, ` +
   `sem pontuação extra e sem termos genéricos que apareceriam em qualquer ${subject} (ex.: "Brasil", ` +
   '"povo", "governo"). Se não houver termo confiável, devolva a lista vazia. Nunca invente fatos.'
@@ -48,14 +54,34 @@ const CONTENT_PIECE_CORPUS_PROMPT =
   themeExpansionRules('peça')
 
 /**
- * S28 — the corpus a theme expansion serves: the internal speech acervo (C192)
- * and the public Central de Conteúdos (S28). The mechanism is one; only the
- * system prompt describes a different body of material and persona.
+ * C219 — the third corpus: the uploaded recordings of the acervo ("Gravações
+ * enviadas"), whose material is the team's own plenárias/debates rather than
+ * the Câmara's speeches, so the prompt describes that body of material.
  */
-export type ThemeSearchCorpus = 'speech' | 'contentPiece'
+const RECORDING_CORPUS_PROMPT =
+  'Você ajuda a assessoria de comunicação a achar, no acervo de gravações enviadas pela equipe ' +
+  '(plenárias, debates e materiais próprios do deputado Jorge Solla), trechos sobre um tema mesmo ' +
+  'quando a gravação não usa as palavras exatas da busca. ' +
+  'Receberá o tema digitado e deve devolver termos e expressões curtas que provavelmente aparecem ' +
+  'na transcrição de uma gravação sobre esse tema — sinônimos, nomes de programas, termos técnicos ' +
+  'e palavras relacionadas. Inclua a própria expressão do tema quando ela for útil. ' +
+  themeExpansionRules('gravação')
 
-const systemPromptFor = (corpus: ThemeSearchCorpus): string =>
-  corpus === 'contentPiece' ? CONTENT_PIECE_CORPUS_PROMPT : SPEECH_CORPUS_PROMPT
+/**
+ * S28/C219 — the corpus a theme expansion serves: the internal speech acervo
+ * (C192), the public Central de Conteúdos (S28) and the uploaded recordings
+ * (C219). The mechanism is one; only the system prompt describes a different
+ * body of material and persona.
+ */
+export type ThemeSearchCorpus = 'speech' | 'contentPiece' | 'recording'
+
+const SYSTEM_PROMPT_BY_CORPUS: Record<ThemeSearchCorpus, string> = {
+  speech: SPEECH_CORPUS_PROMPT,
+  contentPiece: CONTENT_PIECE_CORPUS_PROMPT,
+  recording: RECORDING_CORPUS_PROMPT,
+}
+
+const systemPromptFor = (corpus: ThemeSearchCorpus): string => SYSTEM_PROMPT_BY_CORPUS[corpus]
 
 export type ThemeSearchExpansion = { terms: string[] }
 
@@ -110,3 +136,7 @@ export const expandSpeechSearchTheme: SpeechThemeExpansionResolver = (theme) =>
 /** S28 — the public Central de Conteúdos expansion. */
 export const expandContentPieceSearchTheme: ThemeSearchExpansionResolver = (theme) =>
   expandSearchTheme(theme, 'contentPiece')
+
+/** C219 — the uploaded recordings expansion ("Gravações enviadas"). */
+export const expandRecordingSearchTheme: ThemeSearchExpansionResolver = (theme) =>
+  expandSearchTheme(theme, 'recording')

src/utilities/campaignListUrl.ts

@@ -82,6 +82,21 @@ export const strictDecimalInteger = (value: string | undefined): number | undefi
   return Number.isSafeInteger(number) ? number : undefined
 }
 
+/**
+ * Structural validation of the data-driven year facet (both acervo sources):
+ * a plausible calendar year, not a closed enum.
+ */
+export const parseYearValues = (raw: string | string[] | undefined): number[] =>
+  allParamValues(raw)
+    .map((token) => strictDecimalInteger(token))
+    .filter((year): year is number => year !== undefined && year >= 1900 && year <= 2100)
+
+/** Structural validation of a municipality-id facet (positive safe integer). */
+export const parseMunicipalityValues = (raw: string | string[] | undefined): number[] =>
+  allParamValues(raw)
+    .map((token) => strictDecimalInteger(token))
+    .filter((id): id is number => id !== undefined && id > 0)
+
 export const inspectRawListParams = (
   params: RawSearchParams,
   paramNameSet: ReadonlySet<string>,

src/utilities/recordings/recordingClassification.ts

@@ -0,0 +1,97 @@
+import 'server-only'
+
+import type { Payload } from 'payload'
+
+import type { MergedSpeechFacets, SpeechScope, SpeechTopic } from '@/lib/speechFacets'
+import type { SpeechFacetInput } from '@/lib/speechGazetteer'
+import type { Recording } from '@/payload-types'
+import {
+  classifySpeech,
+  type SpeechClassificationResult,
+} from '@/utilities/speech/speechClassifier'
+
+/**
+ * C219 — facet classification of one uploaded recording. Reuses the speech
+ * classifier owner (`classifySpeech`: gazetteer + validated LLM refinement)
+ * over the transcript the job already has in memory, so recordings gain the
+ * same topic/scope/município facets with the same provenance vocabulary. The
+ * classifier never throws; this wrapper is also defensive (empty transcript or
+ * a database failure resolve to `null`) because the recording must still
+ * become `ready` with its transcript.
+ */
+
+export type RecordingFacetClassifier = (
+  input: SpeechFacetInput,
+) => Promise<SpeechClassificationResult>
+
+export type RecordingClassification = {
+  topics: SpeechTopic[]
+  scopes: SpeechScope[]
+  classifiedBy: MergedSpeechFacets['classifiedBy']
+  mentionedMunicipalities: number[]
+  /** Provider usage of the refinement pass; the backfill CLI reports it. */
+  llm: SpeechClassificationResult['llm']
+}
+
+/**
+ * The facet write shape, shared by the two writers (the job and the backfill
+ * CLI) so a new facet field lands in one place.
+ */
+export const recordingFacetWriteData = (
+  classification: RecordingClassification,
+): Pick<Recording, 'topics' | 'scopes' | 'classifiedBy' | 'mentionedMunicipalities'> => ({
+  topics: classification.topics,
+  scopes: classification.scopes,
+  classifiedBy: classification.classifiedBy,
+  mentionedMunicipalities: classification.mentionedMunicipalities,
+})
+
+export const classifyRecordingFacets = async ({
+  payload,
+  transcript,
+  classify = classifySpeech,
+}: {
+  payload: Pick<Payload, 'find' | 'logger'>
+  transcript: string
+  classify?: RecordingFacetClassifier
+}): Promise<RecordingClassification | null> => {
+  const text = transcript.trim()
+  if (!text) return null
+
+  try {
+    const { facets, llm } = await classify({ transcript: text })
+
+    const slugs = [...new Set(facets.municipalities.map((entry) => entry.slug))]
+    const municipalities =
+      slugs.length === 0
+        ? []
+        : (
+            await payload.find({
+              collection: 'municipality',
+              where: { slug: { in: slugs } },
+              limit: slugs.length,
+              depth: 0,
+              select: { slug: true },
+              // Intentional admin bypass: resolving gazetteer slugs against the
+              // read-only município catalog (same as the speech import).
+              overrideAccess: true,
+            })
+          ).docs
+
+    return {
+      topics: facets.topics,
+      scopes: facets.scopes,
+      classifiedBy: facets.classifiedBy,
+      mentionedMunicipalities: municipalities.map((doc) => doc.id),
+      llm,
+    }
+  } catch (error) {
+    // Defensive: the classifier contract is "never throws"; a database failure
+    // while resolving the gazetteer slugs lands here. The recording still gets
+    // its transcript — but the failure is logged instead of vanishing.
+    payload.logger.warn(
+      `[recording-classification] falha ao classificar: ${error instanceof Error ? error.message : String(error)}`,
+    )
+    return null
+  }
+}

src/utilities/recordings/recordingJob.ts

@@ -39,6 +39,12 @@ import {
   type PayloadTransactionRequest,
 } from '@/utilities/payloadTransaction'
 import { downloadPrivateMediaToFile } from '@/utilities/privateMedia/privateMediaResponse'
+import {
+  classifyRecordingFacets,
+  recordingFacetWriteData,
+  type RecordingFacetClassifier,
+} from '@/utilities/recordings/recordingClassification'
+import { classifySpeech } from '@/utilities/speech/speechClassifier'
 
 /**
  * C199 — the transcription job: download the private recording, extract/split
@@ -128,6 +134,7 @@ export const runRecordingJob = async (
   recordingId: number,
   transcribe: ChunkTranscriber = deepInfraTranscribeSegments,
   diarize: SpeakerDiarizer | null = configuredSpeakerDiarizer(),
+  classify: RecordingFacetClassifier = classifySpeech,
 ): Promise<void> => {
   let tempDir: string | null = null
   let currentStep: RecordingStep = 'extracting'
@@ -237,6 +244,15 @@ export const runRecordingJob = async (
       }
     }
 
+    // C219 — facet classification runs OUTSIDE the save transaction (the LLM
+    // may take a minute) and never fails the recording: a null result keeps the
+    // transcript with no facets and no provenance claim.
+    const classification = await classifyRecordingFacets({
+      payload,
+      transcript: keyedSegments.map((segment) => segment.text).join(' '),
+      classify,
+    })
+
     await markStep('saving')
     await withPayloadTransaction(payload, async ({ req }) => {
       // Fresh reads inside the transaction: the labels a curator wrote while
@@ -310,6 +326,7 @@ export const runRecordingJob = async (
           searchText: recordingSearchText(segments),
           speakerLabels: labels,
           speakerLabelsDropped: dropped,
+          ...(classification ? recordingFacetWriteData(classification) : {}),
         },
         req,
       )

src/utilities/recordings/recordingListFilters.ts

@@ -1,22 +1,67 @@
 /**
- * C199 — Payload `where` of the uploaded-recordings list: one textual branch
- * over the normalized `recording.searchText` (the concatenation of the
- * transcript segments). C200 adds the "Pessoa" facet: any selected label that
- * appears in the recording's derived `speakerNames`. A recording still
- * processing has no `searchText` and is honestly out of the search until the
- * transcript is saved.
+ * C199/C219 — Payload `where` of the uploaded-recordings list. One textual
+ * branch over the normalized `recording.searchText` (the concatenation of the
+ * transcript segments), the "Pessoa" facet (any selected label present in the
+ * derived `speakerNames`) and the C219 parity facets — year, topic, scope,
+ * cited municipality and duration — AND-ed like the Câmara's. A duration sort
+ * also gates `durationSeconds` to exist, otherwise Postgres' NULLS FIRST would
+ * open "Duração (maior)" with the rows that have no duration.
  */
 import type { Where } from 'payload'
 
 import { normalizeForSearch } from '@/lib/speechSearch'
 import { collapseListWhereOrBranches } from '@/utilities/campaignListUrl'
-import type { RecordingListState } from '@/utilities/recordings/recordingListUrl'
+import {
+  recordingSortIsDuration,
+  type RecordingListState,
+} from '@/utilities/recordings/recordingListUrl'
+import { durationBucketWhere } from '@/utilities/speech/speechListFilters'
 
-export const buildRecordingListWhere = (state: RecordingListState): Where => {
+const buildRecordingFacetWhere = (state: RecordingListState): Where[] => {
+  const filters: Where[] = []
+
+  if (state.years?.length) filters.push({ year: { in: state.years } })
+  if (state.topics?.length) filters.push({ topics: { in: state.topics } })
+  if (state.scopes?.length) filters.push({ scopes: { in: state.scopes } })
+  if (state.municipalities?.length) {
+    filters.push({ mentionedMunicipalities: { in: state.municipalities } })
+  }
+  if (state.durations?.length) {
+    const branch = collapseListWhereOrBranches(state.durations.map(durationBucketWhere))
+    if (branch) filters.push(branch)
+  }
+  if (recordingSortIsDuration(state)) filters.push({ durationSeconds: { exists: true } })
+
+  return filters
+}
+
+/**
+ * The textual branch: the literal query plus (C219) every expanded theme term,
+ * each matching the normalized recording text. Recordings have no official
+ * keywords, so the speech `keywords` branch has no counterpart here.
+ */
+const buildRecordingTextWhere = (terms: readonly string[]): Where | undefined => {
   const branches: Where[] = []
+  const seen = new Set<string>()
 
-  const normalized = normalizeForSearch(state.q ?? '')
-  if (normalized) branches.push({ searchText: { like: normalized } })
+  for (const term of terms) {
+    const normalized = normalizeForSearch(term)
+    if (!normalized || seen.has(normalized)) continue
+    seen.add(normalized)
+    branches.push({ searchText: { like: normalized } })
+  }
+
+  return collapseListWhereOrBranches(branches)
+}
+
+export const buildRecordingListWhere = (
+  state: RecordingListState,
+  themeTerms: readonly string[] = [],
+): Where => {
+  const branches = buildRecordingFacetWhere(state)
+
+  const textBranch = buildRecordingTextWhere([...(state.q ? [state.q] : []), ...themeTerms])
+  if (textBranch) branches.push(textBranch)
 
   const peopleBranch = collapseListWhereOrBranches(
     (state.people ?? []).map((person) => ({ speakerNames: { contains: person } })),

src/utilities/recordings/recordingListUrl.ts

@@ -1,43 +1,108 @@
 /**
- * C199 — source switcher + list URL contract of the uploaded recordings. The
- * Câmara acervo keeps its own frozen contract (`speechListUrl`); the recordings
- * branch lives on the same `/campanha/comunicacao/acervo` page behind
- * `?source=enviadas` and owns its parses/serialization over the shared helpers.
+ * C199/C219 — source switcher + list URL contract of the uploaded recordings.
+ * The Câmara acervo keeps its own frozen contract (`speechListUrl`); the
+ * recordings branch lives on the same `/campanha/comunicacao/acervo` page
+ * behind `?source=enviadas` and owns its parses/serialization over the shared
+ * helpers. C219 brings the parity facets — search mode, year, topic, scope,
+ * cited municipality, duration and sort — reusing the Câmara taxonomy, the
+ * duration buckets and the structural year/municipality parsers. The original
+ * `?source=enviadas&q&person&page` deep links stay byte-identical: the new
+ * params are only serialized when selected and `mode=termo`/the default sort
+ * are never serialized.
  */
 import { CAMPAIGN_COMMUNICATION_ACERVO } from '@/lib/campaignPaths'
 import { isContactSearchQueryReady } from '@/lib/contactSearchQuery'
 import { RECORDING_SPEAKER_LABEL_MAX_LENGTH } from '@/lib/recording'
 import { foldSpeakerName } from '@/lib/recordingDiarization'
+import {
+  SPEECH_SCOPES,
+  SPEECH_TOPICS,
+  type SpeechScope,
+  type SpeechTopic,
+} from '@/lib/speechFacets'
 import {
   allParamValues,
   buildListHref,
   firstValue,
   normalizedText,
+  parseExhaustiveEnumParam,
+  parseMunicipalityValues,
+  parseYearValues,
   resolveListUrl,
   strictDecimalInteger,
   type RawSearchParams,
 } from '@/utilities/campaignListUrl'
+import {
+  SPEECH_DURATION_BUCKETS,
+  type SpeechDurationBucket,
+  type SpeechSearchMode,
+} from '@/utilities/speech/speechListUrl'
 
 export const recordingPageSize = 25
 
 /** Which acervo source the page is showing. `camara` is the default. */
 export type AcervoSource = 'camara' | 'enviadas'
 
+/**
+ * C219 — list ordering. `recentes` is the default and never serialized (the
+ * stored contract stays `-createdAt`); the duration orders only list rows with
+ * a measured duration, which is why the `where` gates `durationSeconds`.
+ */
+export const RECORDING_SORT_OPTIONS = [
+  { value: 'recentes', label: 'Mais recentes' },
+  { value: 'duracao_maior', label: 'Duração (maior)' },
+  { value: 'duracao_menor', label: 'Duração (menor)' },
+] as const
+
+export type RecordingSortKey = (typeof RECORDING_SORT_OPTIONS)[number]['value']
+
 const ACERVO_SOURCE_PARAM = 'source'
 const ACERVO_SOURCE_ENVIADAS = 'enviadas'
 const ACERVO_PERSON_PARAM = 'person'
 
-const recordingListParamNames = [ACERVO_SOURCE_PARAM, 'q', 'page', ACERVO_PERSON_PARAM] as const
+const recordingListParamNames = [
+  ACERVO_SOURCE_PARAM,
+  'q',
+  'mode',
+  'year',
+  'topic',
+  'scope',
+  'municipality',
+  'duration',
+  'sort',
+  ACERVO_PERSON_PARAM,
+  'page',
+] as const
 const recordingListParamNameSet = new Set<string>(recordingListParamNames)
+const recordingTopicSet = new Set<string>(SPEECH_TOPICS.map(({ value }) => value))
+const recordingScopeSet = new Set<string>(SPEECH_SCOPES.map(({ value }) => value))
+const recordingDurationSet = new Set<string>(SPEECH_DURATION_BUCKETS.map(({ value }) => value))
+const recordingSortSet = new Set<string>(RECORDING_SORT_OPTIONS.map(({ value }) => value))
 
 export type RecordingListState = {
   source: 'enviadas'
   page: number
   q?: string
+  /** C219 — `tema` asks the semantic bridge; `termo` (default) is never stored. */
+  mode?: SpeechSearchMode
+  years?: number[]
+  topics?: SpeechTopic[]
+  scopes?: SpeechScope[]
+  /** Cited municipalities (`mentionedMunicipalities`), not the uploader's. */
+  municipalities?: number[]
+  durations?: SpeechDurationBucket[]
+  sort?: RecordingSortKey
   /** Selected "Pessoa" labels; repeated `person` params, literal text. */
   people?: string[]
 }
 
+/** Server-loaded filter options (present years, cited municipalities, people). */
+export type RecordingFilterOptions = {
+  people: string[]
+  years: number[]
+  municipalities: { value: string; label: string }[]
+}
+
 /** `enviadas` only for the explicit value; anything else is the Câmara default. */
 export const parseAcervoSource = (params: RawSearchParams): AcervoSource =>
   firstValue(params[ACERVO_SOURCE_PARAM]) === ACERVO_SOURCE_ENVIADAS ? 'enviadas' : 'camara'
@@ -60,33 +125,99 @@ export const parseRecordingPeople = (raw: string | string[] | undefined): string
   return people
 }
 
+/** Only `tema` is meaningful; anything else (and `termo`) means the default. */
+const parseRecordingSearchMode = (
+  raw: string | string[] | undefined,
+): SpeechSearchMode | undefined => (firstValue(raw) === 'tema' ? 'tema' : undefined)
+
+/** Only the non-default duration orders are meaningful; anything else is `recentes`. */
+const parseRecordingSort = (raw: string | string[] | undefined): RecordingSortKey | undefined => {
+  const value = firstValue(raw)
+  return value && value !== 'recentes' && recordingSortSet.has(value)
+    ? (value as RecordingSortKey)
+    : undefined
+}
+
 export const parseRecordingListParams = (params: RawSearchParams): RecordingListState => {
   const rawPage = strictDecimalInteger(firstValue(params.page))
   const rawQ = normalizedText(firstValue(params.q))
   const q = rawQ && isContactSearchQueryReady(rawQ) ? rawQ : undefined
+  const mode = parseRecordingSearchMode(params.mode)
+  const years = parseYearValues(params.year)
+  const topics = parseExhaustiveEnumParam<SpeechTopic>(params.topic, recordingTopicSet)
+  const scopes = parseExhaustiveEnumParam<SpeechScope>(params.scope, recordingScopeSet)
+  const municipalities = parseMunicipalityValues(params.municipality)
+  const durations = parseExhaustiveEnumParam<SpeechDurationBucket>(
+    params.duration,
+    recordingDurationSet,
+  )
+  const sort = parseRecordingSort(params.sort)
   const people = parseRecordingPeople(params[ACERVO_PERSON_PARAM])
 
   return {
     source: 'enviadas',
     page: rawPage ?? 1,
     ...(q ? { q } : {}),
+    ...(mode ? { mode } : {}),
+    ...(years.length ? { years } : {}),
+    ...(topics.length ? { topics } : {}),
+    ...(scopes.length ? { scopes } : {}),
+    ...(municipalities.length ? { municipalities } : {}),
+    ...(durations.length ? { durations } : {}),
+    ...(sort ? { sort } : {}),
     ...(people.length > 0 ? { people } : {}),
   }
 }
 
-const recordingListSearchParams = (
+const recordingListStateToRawParams = (
   state: RecordingListState,
   page = state.page,
+): RawSearchParams => ({
+  [ACERVO_SOURCE_PARAM]: ACERVO_SOURCE_ENVIADAS,
+  page: String(page),
+  q: state.q,
+  mode: state.mode,
+  year: state.years?.map(String),
+  topic: state.topics,
+  scope: state.scopes,
+  municipality: state.municipalities?.map(String),
+  duration: state.durations,
+  sort: state.sort,
+  [ACERVO_PERSON_PARAM]: state.people,
+})
+
+/** Expects already-canonical state (from parse or a rule-preserving toggle). */
+const serializeCanonicalRecordingListSearchParams = (
+  canonicalState: RecordingListState,
 ): URLSearchParams => {
   const params = new URLSearchParams()
   // `source=enviadas` is always serialized — it is what selects the source.
   params.set(ACERVO_SOURCE_PARAM, ACERVO_SOURCE_ENVIADAS)
-  if (state.q) params.set('q', state.q)
-  for (const person of state.people ?? []) params.append(ACERVO_PERSON_PARAM, person)
-  if (page > 1) params.set('page', String(page))
+  if (canonicalState.q) params.set('q', canonicalState.q)
+  // `termo` is the default and never serialized; `tema` only makes sense with a
+  // query, so a theme param without `q` canonicalizes away.
+  if (canonicalState.mode === 'tema' && canonicalState.q) params.set('mode', 'tema')
+  for (const year of canonicalState.years ?? []) params.append('year', String(year))
+  for (const topic of canonicalState.topics ?? []) params.append('topic', topic)
+  for (const scope of canonicalState.scopes ?? []) params.append('scope', scope)
+  for (const municipality of canonicalState.municipalities ?? []) {
+    params.append('municipality', String(municipality))
+  }
+  for (const duration of canonicalState.durations ?? []) params.append('duration', duration)
+  // `recentes` is the default and never serialized.
+  if (canonicalState.sort && canonicalState.sort !== 'recentes') {
+    params.set('sort', canonicalState.sort)
+  }
+  for (const person of canonicalState.people ?? []) params.append(ACERVO_PERSON_PARAM, person)
+  if (canonicalState.page > 1) params.set('page', String(canonicalState.page))
   return params
 }
 
+const recordingListSearchParams = (state: RecordingListState, page = state.page): URLSearchParams =>
+  serializeCanonicalRecordingListSearchParams(
+    parseRecordingListParams(recordingListStateToRawParams(state, page)),
+  )
+
 /** The URL of one "Pessoa" facet option with `person` toggled (page resets). */
 export const toggleRecordingPerson = (
   state: RecordingListState,
@@ -114,6 +245,47 @@ export const removeRecordingPerson = (
 export const buildRecordingListHref = (state: RecordingListState, page: number): string =>
   buildListHref(state, recordingListSearchParams, CAMPAIGN_COMMUNICATION_ACERVO, page)
 
+/**
+ * C219 — the canonical query drives the filter bar remount (`key`), exactly
+ * like `buildSpeechFiltersKey`, so optimistic state resets when the URL
+ * changes.
+ */
+export const buildRecordingFiltersKey = (state: RecordingListState): string =>
+  recordingListSearchParams(state).toString()
+
+/** Payload sort order of the list: the default keeps the stored `-createdAt`. */
+export const recordingSortOrder = (state: RecordingListState): string => {
+  switch (state.sort) {
+    case 'duracao_maior':
+      return '-durationSeconds'
+    case 'duracao_menor':
+      return 'durationSeconds'
+    default:
+      return '-createdAt'
+  }
+}
+
+/** Duration orders only make sense over rows with a measured duration. */
+export const recordingSortIsDuration = (state: RecordingListState): boolean =>
+  state.sort === 'duracao_maior' || state.sort === 'duracao_menor'
+
+/**
+ * C219 — whether the actor narrowed the list (query/mode or any facet). The
+ * empty state uses it to tell "no recording was ever sent" from "the filters
+ * hid everything"; pagination is not a filter.
+ */
+export const recordingHasActiveFilters = (state: RecordingListState): boolean =>
+  Boolean(
+    state.q ||
+    state.mode ||
+    state.years?.length ||
+    state.topics?.length ||
+    state.scopes?.length ||
+    state.municipalities?.length ||
+    state.durations?.length ||
+    state.people?.length,
+  )
+
 /** Toggle hrefs of the source switcher (the Câmara side is the bare acervo). */
 export const buildAcervoSourceHref = (source: AcervoSource): string =>
   source === 'enviadas'

src/utilities/recordings/recordingOmnibox.ts

@@ -0,0 +1,286 @@
+/**
+ * C219 — recordings acervo omnibox adapter. Pure / client-safe: builds the
+ * active chips, the suggestion seeds (facet constants plus server-loaded
+ * years/municipalities/people) and the toggle/remove/clear actions over the
+ * URL state, mirroring `speechOmnibox` on the same shared helpers. "Pessoa"
+ * reuses the case-insensitive dedupe owned by the recordings URL contract.
+ */
+import {
+  createOmniboxSuggestionSeed,
+  filterOmniboxSuggestionSeeds,
+  type CampaignListOmniboxChip,
+  type CampaignListOmniboxSuggestion,
+} from '@/lib/campaignListOmnibox'
+import {
+  SPEECH_SCOPES,
+  SPEECH_TOPICS,
+  type SpeechScope,
+  type SpeechTopic,
+} from '@/lib/speechFacets'
+import {
+  removeRecordingPerson,
+  toggleRecordingPerson,
+  type RecordingListState,
+} from '@/utilities/recordings/recordingListUrl'
+import {
+  SPEECH_DURATION_BUCKETS,
+  speechDurationLabels,
+  speechScopeLabels,
+  speechTopicLabels,
+  type SpeechDurationBucket,
+  type SpeechSearchMode,
+} from '@/utilities/speech/speechListUrl'
+
+export type RecordingFilterOption = {
+  value: string
+  label: string
+}
+
+const MUNICIPALITY_CHIP_LABEL = 'Município citado'
+
+export type RecordingOmniboxAction =
+  | { kind: 'url'; state: RecordingListState }
+  | { kind: 'clear'; state: RecordingListState }
+
+const chipLabel = (dimension: string, value: string): string => `${dimension}: ${value}`
+
+const withPageReset = (state: RecordingListState): RecordingListState => ({ ...state, page: 1 })
+
+const toggleValue = <T>(values: T[] | undefined, value: T): T[] | undefined => {
+  const current = values ?? []
+  const next = current.includes(value)
+    ? current.filter((item) => item !== value)
+    : [...current, value]
+  return next.length ? next : undefined
+}
+
+const removeValue = <T>(values: T[] | undefined, value: T): T[] | undefined => {
+  const next = (values ?? []).filter((item) => item !== value)
+  return next.length ? next : undefined
+}
+
+const chipPrefix = (chipId: string): { key: string; value: string } => {
+  const separator = chipId.indexOf(':')
+  if (separator === -1) return { key: chipId, value: '' }
+  return { key: chipId.slice(0, separator), value: chipId.slice(separator + 1) }
+}
+
+type FacetMutation = 'toggle' | 'remove'
+
+/**
+ * ONE facet mutation for both omnibox actions: `toggle` (selecting a
+ * suggestion) and `remove` (dropping a chip) differ only in the values
+ * operation. Returns null when the id carries an invalid numeric value.
+ */
+const mutateFacet = (
+  state: RecordingListState,
+  key: string,
+  value: string,
+  mode: FacetMutation,
+): RecordingListState | null => {
+  const mutate = mode === 'toggle' ? toggleValue : removeValue
+  switch (key) {
+    case 'year': {
+      const year = Number(value)
+      if (!Number.isSafeInteger(year)) return null
+      return { ...state, years: mutate(state.years, year) }
+    }
+    case 'topic':
+      return { ...state, topics: mutate(state.topics, value as SpeechTopic) }
+    case 'scope':
+      return { ...state, scopes: mutate(state.scopes, value as SpeechScope) }
+    case 'municipality': {
+      const municipality = Number(value)
+      if (!Number.isSafeInteger(municipality)) return null
+      return { ...state, municipalities: mutate(state.municipalities, municipality) }
+    }
+    case 'duration':
+      return { ...state, durations: mutate(state.durations, value as SpeechDurationBucket) }
+    default:
+      return null
+  }
+}
+
+export const buildRecordingOmniboxChips = ({
+  state,
+  municipalityLabelsById,
+}: {
+  state: RecordingListState
+  municipalityLabelsById: ReadonlyMap<number, string>
+}): CampaignListOmniboxChip[] => {
+  const chips: CampaignListOmniboxChip[] = []
+
+  if (state.q) chips.push({ id: 'q', label: chipLabel('Busca', state.q) })
+  for (const year of state.years ?? []) {
+    chips.push({ id: `year:${year}`, label: chipLabel('Ano', String(year)) })
+  }
+  for (const topic of state.topics ?? []) {
+    chips.push({ id: `topic:${topic}`, label: chipLabel('Tema', speechTopicLabels[topic]) })
+  }
+  for (const scope of state.scopes ?? []) {
+    chips.push({ id: `scope:${scope}`, label: chipLabel('Alcance', speechScopeLabels[scope]) })
+  }
+  for (const municipality of state.municipalities ?? []) {
+    chips.push({
+      id: `municipality:${municipality}`,
+      label: chipLabel(
+        MUNICIPALITY_CHIP_LABEL,
+        municipalityLabelsById.get(municipality) ?? `Município #${municipality}`,
+      ),
+    })
+  }
+  for (const duration of state.durations ?? []) {
+    chips.push({
+      id: `duration:${duration}`,
+      label: chipLabel('Duração', speechDurationLabels[duration]),
+    })
+  }
+  for (const person of state.people ?? []) {
+    chips.push({ id: `person:${person}`, label: chipLabel('Pessoa', person) })
+  }
+
+  return chips
+}
+
+export const buildRecordingOmniboxSuggestionSeeds = ({
+  years,
+  municipalityOptions,
+  personOptions,
+}: {
+  years: readonly number[]
+  municipalityOptions: readonly RecordingFilterOption[]
+  personOptions: readonly string[]
+}) => {
+  const seeds = []
+
+  for (const { value, label } of SPEECH_TOPICS) {
+    seeds.push(
+      createOmniboxSuggestionSeed(
+        { id: `topic:${value}`, group: 'Tema', label, keywords: ['tema'] },
+        { emptyQueryVisible: true },
+      ),
+    )
+  }
+
+  for (const { value, label } of SPEECH_SCOPES) {
+    seeds.push(
+      createOmniboxSuggestionSeed(
+        { id: `scope:${value}`, group: 'Alcance', label, keywords: ['alcance'] },
+        { emptyQueryVisible: true },
+      ),
+    )
+  }
+
+  for (const { value, label } of SPEECH_DURATION_BUCKETS) {
+    seeds.push(
+      createOmniboxSuggestionSeed(
+        { id: `duration:${value}`, group: 'Duração', label, keywords: ['duracao', 'tempo'] },
+        { emptyQueryVisible: true },
+      ),
+    )
+  }
+
+  for (const year of years) {
+    seeds.push(
+      createOmniboxSuggestionSeed({
+        id: `year:${year}`,
+        group: 'Ano',
+        label: String(year),
+        keywords: ['ano'],
+      }),
+    )
+  }
+
+  for (const option of municipalityOptions) {
+    seeds.push(
+      createOmniboxSuggestionSeed({
+        id: `municipality:${option.value}`,
+        group: MUNICIPALITY_CHIP_LABEL,
+        label: option.label,
+        keywords: ['municipio', 'cidade', 'citado'],
+      }),
+    )
+  }
+
+  for (const person of personOptions) {
+    seeds.push(
+      createOmniboxSuggestionSeed({
+        id: `person:${person}`,
+        group: 'Pessoa',
+        label: person,
+        keywords: ['pessoa', 'falante'],
+      }),
+    )
+  }
+
+  return seeds
+}
+
+export const filterRecordingOmniboxSuggestions = (
+  seeds: ReturnType<typeof buildRecordingOmniboxSuggestionSeeds>,
+  query: string,
+): CampaignListOmniboxSuggestion[] => filterOmniboxSuggestionSeeds(seeds, query)
+
+export const applyRecordingOmniboxSuggestion = ({
+  state,
+  suggestionId,
+}: {
+  state: RecordingListState
+  suggestionId: string
+}): RecordingOmniboxAction => {
+  if (suggestionId.startsWith('q:')) {
+    const q = suggestionId.slice(2).trim()
+    return { kind: 'url', state: withPageReset({ ...state, q: q || undefined }) }
+  }
+
+  const { key, value } = chipPrefix(suggestionId)
+  if (key === 'person') {
+    return { kind: 'url', state: toggleRecordingPerson(withPageReset(state), value) }
+  }
+
+  const next = mutateFacet(withPageReset(state), key, value, 'toggle')
+  return { kind: 'url', state: next ?? state }
+}
+
+export const removeRecordingOmniboxChip = ({
+  state,
+  chipId,
+}: {
+  state: RecordingListState
+  chipId: string
+}): RecordingOmniboxAction => {
+  // Dropping the query also drops the theme mode: a theme without a query has
+  // nothing to expand.
+  if (chipId === 'q') {
+    return { kind: 'url', state: withPageReset({ ...state, q: undefined, mode: undefined }) }
+  }
+
+  const { key, value } = chipPrefix(chipId)
+  if (key === 'person') {
+    return { kind: 'url', state: removeRecordingPerson(withPageReset(state), value) }
+  }
+
+  const next = mutateFacet(withPageReset(state), key, value, 'remove')
+  return { kind: 'url', state: next ?? state }
+}
+
+/**
+ * C219 — the search mode toggle. `tema` only sticks with a query; `termo`
+ * (the default) clears the param so the URL stays canonical and shareable.
+ */
+export const applyRecordingSearchMode = ({
+  state,
+  mode,
+}: {
+  state: RecordingListState
+  mode: SpeechSearchMode
+}): RecordingOmniboxAction => ({
+  kind: 'url',
+  state: withPageReset({ ...state, mode: mode === 'tema' && state.q ? 'tema' : undefined }),
+})
+
+/** Clearing keeps `source=enviadas` — it is what selects this source. */
+export const clearRecordingOmnibox = (_state: RecordingListState): RecordingOmniboxAction => ({
+  kind: 'clear',
+  state: { source: 'enviadas', page: 1 },
+})

src/utilities/recordings/recordingPageData.ts

@@ -2,15 +2,24 @@ import 'server-only'
 
 import type { Payload } from 'payload'
 
+import { canReadCommunicationCatalog } from '@/lib/campaignRoles'
 import { foldSpeakerName } from '@/lib/recordingDiarization'
 import { RECORDING_NOT_FOUND_MESSAGE } from '@/lib/schemas/recording'
 import { normalizeForSearch } from '@/lib/speechSearch'
 import type { CampaignUser } from '@/payload-types'
+import {
+  expandRecordingSearchTheme,
+  type ThemeSearchExpansionResolver,
+} from '@/utilities/ai/expandSpeechSearchTheme'
+import { collapseListWhereOrBranches } from '@/utilities/campaignListUrl'
 import { createEntityNotFoundError } from '@/utilities/entityNotFound'
+import { loadMunicipalityLabelsByIds } from '@/utilities/loadNamesByIds'
 import { buildRecordingListWhere } from '@/utilities/recordings/recordingListFilters'
 import {
   recordingPageSize,
+  recordingSortOrder,
   resolveRecordingListUrl,
+  type RecordingFilterOptions,
   type RecordingListState,
 } from '@/utilities/recordings/recordingListUrl'
 import {
@@ -35,6 +44,8 @@ const recordingListSelect = {
   recordedAt: true,
   durationSeconds: true,
   speakerNames: true,
+  // C219 — the card's topic chip; the facet itself is not displayed.
+  topics: true,
 } as const
 
 const recordingDetailSelect = {
@@ -63,17 +74,12 @@ const toSegmentRecord = (segment: {
   speakerKey: segment.speakerKey ?? null,
 })
 
-/** C200 — the options the "Pessoa" facet offers. */
-export type RecordingFilterOptions = {
-  people: string[]
-}
-
 /**
- * C200 — distinct labels across the recordings the actor can read: the facet
- * only ever lists what is reachable, and the derived `speakerNames` column is
- * the single source. Small acervo now; if the list grows past a few thousand
- * rows this becomes a `SELECT DISTINCT` (registered as a revisit trigger in the
- * impl plan).
+ * C200/C219 — distinct labels/years/cited municipalities across the
+ * recordings the actor can read: a facet only ever lists what is reachable,
+ * and the derived columns are the single source. Municipality labels resolve
+ * through the justified `loadMunicipalityLabelsByIds` bypass — the ids come
+ * from recordings the actor was already authorized to read.
  */
 export const loadRecordingFilterOptions = async (
   payload: Payload,
@@ -84,13 +90,15 @@ export const loadRecordingFilterOptions = async (
     depth: 0,
     limit: 0,
     pagination: false,
-    select: { speakerNames: true },
+    select: { speakerNames: true, year: true, mentionedMunicipalities: true },
     user,
     overrideAccess: false,
   })
 
   const people: string[] = []
   const seen = new Set<string>()
+  const years = new Set<number>()
+  const municipalityIds = new Set<number>()
   for (const recording of result.docs) {
     for (const name of recording.speakerNames ?? []) {
       const trimmed = name.trim()
@@ -99,35 +107,56 @@ export const loadRecordingFilterOptions = async (
       seen.add(folded)
       people.push(trimmed)
     }
+    if (typeof recording.year === 'number') years.add(recording.year)
+    for (const value of recording.mentionedMunicipalities ?? []) {
+      municipalityIds.add(typeof value === 'number' ? value : value.id)
+    }
   }
 
-  return { people: people.sort((left, right) => left.localeCompare(right, 'pt-BR')) }
+  const labels = await loadMunicipalityLabelsByIds(payload, [...municipalityIds])
+  const municipalities = [...municipalityIds]
+    .flatMap((id) => {
+      const entry = labels.get(id)
+      return entry ? [{ value: String(id), label: entry.name }] : []
+    })
+    .sort((left, right) => left.label.localeCompare(right.label, 'pt-BR'))
+
+  return {
+    people: people.sort((left, right) => left.localeCompare(right, 'pt-BR')),
+    years: [...years].sort((left, right) => right - left),
+    municipalities,
+  }
 }
 
 /**
  * The excerpt segment of each listed recording: ONE matching segment per row
  * (the search runs `LIKE` on the joined `recording.searchText`, so a page of
- * hour-long recordings never pays for all of its segments). The rows whose
- * query only matches across the junction of two segments fall back together in
- * a single batched query, so a result never renders without an excerpt.
+ * hour-long recordings never pays for all of its segments). C219 passes the
+ * literal query plus every expanded theme term, so a theme result still gets
+ * the passage that surfaced it. The rows whose query only matches across the
+ * junction of two segments fall back together in a single batched query, so a
+ * result never renders without an excerpt.
  */
 const loadMatchedSegments = async (
   payload: Payload,
   user: CampaignUser,
   recordingIds: readonly number[],
-  q: string,
+  terms: readonly string[],
 ): Promise<Map<number, RecordingSegmentRecord[]>> => {
   const byRecording = new Map<number, RecordingSegmentRecord[]>()
-  const normalized = normalizeForSearch(q)
-  if (!normalized || recordingIds.length === 0) return byRecording
+  const normalizedTerms = [
+    ...new Set(terms.map((term) => normalizeForSearch(term)).filter((term) => term.length > 0)),
+  ]
+  const textBranch = collapseListWhereOrBranches(
+    normalizedTerms.map((term) => ({ searchText: { like: term } })),
+  )
+  if (!textBranch || recordingIds.length === 0) return byRecording
 
   await Promise.all(
     recordingIds.map(async (recordingId) => {
       const match = await payload.find({
         collection: 'recordingSegment',
-        where: {
-          and: [{ recording: { equals: recordingId } }, { searchText: { like: normalized } }],
-        },
+        where: { and: [{ recording: { equals: recordingId } }, textBranch] },
         depth: 0,
         limit: 1,
         pagination: false,
@@ -172,25 +201,44 @@ export type RecordingsPageData = {
   redirectHref?: string
   totalDocs: number
   totalPages: number
+  /**
+   * C219 — true when the actor asked for `mode=tema` but the expansion
+   * mechanism was unavailable; the page shows the discreet fallback notice.
+   */
+  themeUnavailable: boolean
+  /** C219 — true when the expansion actually contributed terms. */
+  themeApplied: boolean
 }
 
 export const loadRecordingsPageData = async (
   payload: Payload,
   user: CampaignUser,
   searchParams: Promise<RecordingListSearchParams> | RecordingListSearchParams,
+  expandTheme: ThemeSearchExpansionResolver = expandRecordingSearchTheme,
 ): Promise<RecordingsPageData> => {
   const rawSearchParams = await searchParams
   const canonicalUrl = resolveRecordingListUrl(rawSearchParams)
   const state = canonicalUrl.state
 
+  // C219 — only expand for an actor who may read the catalog (the `find` below
+  // is still the final, fail-closed barrier) and only with a query.
+  const themeRequested = state.mode === 'tema' && Boolean(state.q)
+  let themeTerms: readonly string[] = []
+  let themeUnavailable = false
+  if (themeRequested && canReadCommunicationCatalog(user.role)) {
+    const expansion = await expandTheme(state.q ?? '')
+    if (expansion) themeTerms = expansion.terms
+    else themeUnavailable = true
+  }
+
   const [result, filterOptions] = await Promise.all([
     payload.find({
       collection: 'recording',
       depth: 0,
       limit: recordingPageSize,
       page: state.page,
-      sort: '-createdAt',
-      where: buildRecordingListWhere(state),
+      sort: recordingSortOrder(state),
+      where: buildRecordingListWhere(state, themeTerms),
       select: recordingListSelect,
       user,
       overrideAccess: false,
@@ -200,12 +248,13 @@ export const loadRecordingsPageData = async (
 
   const resolvedUrl = resolveRecordingListUrl(rawSearchParams, result.totalPages)
   const recordings = result.docs
-  const segmentsByRecording = state.q
+  const matchedTerms = [...(state.q ? [state.q] : []), ...themeTerms]
+  const segmentsByRecording = matchedTerms.length
     ? await loadMatchedSegments(
         payload,
         user,
         recordings.map((recording) => recording.id),
-        state.q,
+        matchedTerms,
       )
     : new Map<number, RecordingSegmentRecord[]>()
 
@@ -215,6 +264,7 @@ export const loadRecordingsPageData = async (
         recording,
         matchedSegments: segmentsByRecording.get(recording.id) ?? [],
         query: state.q,
+        themeTerms,
         people: state.people,
       }),
     ),
@@ -223,6 +273,8 @@ export const loadRecordingsPageData = async (
     redirectHref: resolvedUrl.redirectHref ?? canonicalUrl.redirectHref,
     totalDocs: result.totalDocs,
     totalPages: result.totalPages,
+    themeUnavailable,
+    themeApplied: themeTerms.length > 0,
   }
 }
 

src/utilities/recordings/recordingViewModels.ts

@@ -11,6 +11,7 @@ import {
 } from '@/lib/recording'
 import { recordingSpeakerDefaultLabel, speakerNameMatches } from '@/lib/recordingDiarization'
 import { formatSpeechClock } from '@/lib/speechClock'
+import type { SpeechTopic } from '@/lib/speechFacets'
 import {
   buildHighlightedExcerpt,
   pickMatchingSegment,
@@ -18,6 +19,7 @@ import {
   type SpeechHighlightedExcerpt,
   type SpeechHighlightPart,
 } from '@/lib/speechHighlight'
+import { speechTopicLabels } from '@/utilities/speech/speechListUrl'
 
 /** The segment shape the recording detail transcript renders. */
 export type RecordingSegmentRecord = {
@@ -40,6 +42,8 @@ type RecordingBaseRecord = {
 export type RecordingListRecord = RecordingBaseRecord & {
   /** C200 — derived distinct labels; drives the card chip under a "Pessoa" filter. */
   speakerNames?: string[] | null
+  /** C219 — classified topics; the card shows them as chips. */
+  topics?: SpeechTopic[] | null
 }
 
 export type RecordingDetailRecord = RecordingBaseRecord & {
@@ -55,6 +59,8 @@ export type RecordingListItemViewModel = RecordingViewModel & {
   excerpt: SpeechHighlightedExcerpt | null
   /** Selected "Pessoa" labels that this recording indeed contains. */
   matchedPersons: string[]
+  /** C219 — classified topics with their pt-BR labels (card chips). */
+  topics: { value: SpeechTopic; label: string }[]
   /**
    * Detail link that seeks the player to the matching segment when the search
    * found one (`?t=`), carrying the term for transcript highlighting (`?q=`).
@@ -148,6 +154,24 @@ const withSeek = (href: string, segment: RecordingSegmentRecord | undefined, que
   return queryString ? `${href}?${queryString}` : href
 }
 
+/**
+ * C219 — the first expanded theme term that actually surfaced this recording:
+ * the passage that carries it, preferred for the excerpt so a theme result
+ * shows why it appeared. Returns undefined when no theme term applies.
+ */
+const pickThemeSegment = (
+  segments: readonly RecordingSegmentRecord[],
+  themeTerms: readonly string[],
+): { term: string; segment: RecordingSegmentRecord } | undefined => {
+  for (const term of themeTerms) {
+    const trimmed = term.trim()
+    if (!trimmed) continue
+    const segment = pickMatchingSegment(segments, trimmed)
+    if (segment) return { term: trimmed, segment }
+  }
+  return undefined
+}
+
 /**
  * List item of one recording. The excerpt only exists for a `ready` row and a
  * search term; the status cards of the other three states carry their own copy
@@ -157,20 +181,31 @@ export const toRecordingListItemViewModel = ({
   recording,
   matchedSegments,
   query,
+  themeTerms = [],
   people,
 }: {
   recording: RecordingListRecord
   /** Segments that matched the query (or the row's first one as fallback). */
   matchedSegments: readonly RecordingSegmentRecord[]
   query?: string
+  /** C219 — expanded theme terms; empty in the literal search. */
+  themeTerms?: readonly string[]
   /** C200 — the selected "Pessoa" labels of the list state. */
   people?: readonly string[]
 }): RecordingListItemViewModel => {
   const base = toRecordingViewModel(recording)
   const isReady = base.status === 'ready'
   const q = query?.trim()
-  const segment = isReady && q ? pickMatchingSegment(matchedSegments, q) : undefined
-  const excerpt = segment ? buildHighlightedExcerpt(segment.text, q ?? '') : null
+  const themeMatch = isReady ? pickThemeSegment(matchedSegments, themeTerms) : undefined
+  const segment =
+    themeMatch?.segment ?? (isReady && q ? pickMatchingSegment(matchedSegments, q) : undefined)
+  // The theme passage highlights the whole phrase as one band (the per-term
+  // split would break it into pieces), mirroring the speech acervo (C192).
+  const excerpt = themeMatch
+    ? buildHighlightedExcerpt(themeMatch.segment.text, themeMatch.term, { phrase: true })
+    : segment
+      ? buildHighlightedExcerpt(segment.text, q ?? '')
+      : null
   const speakerNames = recording.speakerNames ?? []
 
   return {
@@ -179,6 +214,7 @@ export const toRecordingListItemViewModel = ({
     matchedPersons: (people ?? []).filter((person) =>
       speakerNames.some((name) => speakerNameMatches(name, person)),
     ),
+    topics: (recording.topics ?? []).map((value) => ({ value, label: speechTopicLabels[value] })),
     watchHref: withSeek(base.detailHref, segment, q),
   }
 }

src/utilities/speech/speechListFilters.ts

@@ -13,7 +13,11 @@ import type { SpeechDurationBucket, SpeechListState } from '@/utilities/speech/s
 const DURATION_MEDIA_MIN_SECONDS = 120
 const DURATION_LONGA_MIN_SECONDS = 300
 
-const durationBucketWhere = (bucket: SpeechDurationBucket): Where => {
+/**
+ * The ONE duration-bucket predicate of the acervo: the recordings list (C219)
+ * reuses it so the buckets can never diverge between the two sources.
+ */
+export const durationBucketWhere = (bucket: SpeechDurationBucket): Where => {
   switch (bucket) {
     case 'curta':
       return { durationSeconds: { less_than: DURATION_MEDIA_MIN_SECONDS } }

src/utilities/speech/speechListUrl.ts

@@ -19,6 +19,8 @@ import {
   firstValue,
   normalizedText,
   parseExhaustiveEnumParam,
+  parseMunicipalityValues,
+  parseYearValues,
   resolveListUrl,
   strictDecimalInteger,
   type RawSearchParams,
@@ -95,16 +97,6 @@ const speechScopeSet = new Set<string>(SPEECH_SCOPES.map(({ value }) => value))
 const speechDurationSet = new Set<string>(SPEECH_DURATION_BUCKETS.map(({ value }) => value))
 const MAX_PHASE_LENGTH = 80
 
-const parseYearValues = (raw: string | string[] | undefined): number[] =>
-  allParamValues(raw)
-    .map((token) => strictDecimalInteger(token))
-    .filter((year): year is number => year !== undefined && year >= 1900 && year <= 2100)
-
-const parseMunicipalityValues = (raw: string | string[] | undefined): number[] =>
-  allParamValues(raw)
-    .map((token) => strictDecimalInteger(token))
-    .filter((id): id is number => id !== undefined && id > 0)
-
 /** Only `tema` is meaningful; anything else (and `termo`) means the default. */
 const parseSpeechSearchMode = (raw: string | string[] | undefined): SpeechSearchMode | undefined =>
   firstValue(raw) === 'tema' ? 'tema' : undefined

tests/e2e/campaignSpeechAcervo.e2e.spec.ts

@@ -534,6 +534,13 @@ test.describe('communication vertical (C154/C162)', () => {
           speakerKey?: string | null
         }[]
         speakerLabels?: { speakerKey: string; label: string }[]
+        /** C219 — parity facets. */
+        recordedAt?: string
+        durationSeconds?: number
+        topics?: SpeechTopic[]
+        scopes?: SpeechScope[]
+        classifiedBy?: 'gazetteer' | 'llm' | 'manual'
+        mentionedMunicipalities?: number[]
       },
     ) => {
       const bytes = Buffer.from(`recording-${input.marker}`)
@@ -552,9 +559,18 @@ test.describe('communication vertical (C154/C162)', () => {
         data: {
           title: `Gravação ${input.marker}`,
           status: input.status ?? 'ready',
-          recordedAt: '2026-09-01T00:00:00.000Z',
+          recordedAt: input.recordedAt ?? '2026-09-01T00:00:00.000Z',
           media: media.id,
           ...(input.speakerLabels ? { speakerLabels: input.speakerLabels } : {}),
+          ...(input.durationSeconds !== undefined
+            ? { durationSeconds: input.durationSeconds }
+            : {}),
+          ...(input.topics ? { topics: input.topics } : {}),
+          ...(input.scopes ? { scopes: input.scopes } : {}),
+          ...(input.classifiedBy ? { classifiedBy: input.classifiedBy } : {}),
+          ...(input.mentionedMunicipalities
+            ? { mentionedMunicipalities: input.mentionedMunicipalities }
+            : {}),
         },
         depth: 0,
       })
@@ -806,6 +822,135 @@ test.describe('communication vertical (C154/C162)', () => {
       expect(html).not.toContain(`/campanha/comunicacao/acervo/gravacoes/${other.id}`)
     })
 
+    test('the parity facets narrow, order by duration and render the filter bar (C219)', async ({
+      campaign,
+      campaignRequest,
+    }) => {
+      const marker = campaign.fixtures.value('gravacaofacetas')
+      const otherMarker = `${marker}b`
+      const municipality = await campaign.fixtures.claimMunicipality()
+
+      const { recording: health } = await createRecording(campaign, {
+        marker,
+        recordedAt: '2025-04-02T00:00:00.000Z',
+        durationSeconds: 90,
+        topics: ['saude'],
+        scopes: ['bahia'],
+        classifiedBy: 'llm',
+        mentionedMunicipalities: [municipality.id],
+        segments: [
+          { startSeconds: 12, endSeconds: 20, text: `A plenária discutiu a saúde ${marker}.` },
+        ],
+      })
+      const { recording: culture } = await createRecording(campaign, {
+        marker: otherMarker,
+        recordedAt: '2023-08-02T00:00:00.000Z',
+        durationSeconds: 400,
+        topics: ['cultura'],
+        segments: [
+          {
+            startSeconds: 12,
+            endSeconds: 20,
+            text: `A plenária discutiu a cultura ${otherMarker}.`,
+          },
+        ],
+      })
+      const { recording: noDuration } = await createRecording(campaign, {
+        marker: `${marker}c`,
+        segments: [{ startSeconds: 12, endSeconds: 20, text: `Sem duração ${marker}.` }],
+      })
+
+      const user = await campaign.fixtures.createCampaignUser('communicator')
+      const request = await campaignRequest(user, user.password)
+
+      // The bar renders the Câmara-parity facets, the mode selector and the sort.
+      const bar = rendered(
+        await (
+          await request.get(`/campanha/comunicacao/acervo?source=enviadas&q=${marker}`)
+        ).text(),
+      )
+      for (const label of ['Ano', 'Tema', 'Alcance', 'Município citado', 'Duração', 'Pessoa']) {
+        expect(bar).toContain(label)
+      }
+      expect(bar).toContain('Termo exato')
+      expect(bar).toContain('Por tema')
+      expect(bar).toContain('Ordenar por')
+      expect(bar).toContain('Mais recentes')
+
+      // Year + topic + cited municipality + duration narrow to the health row.
+      // The params ride in the canonical order: a hand-built URL out of order
+      // canonicalizes through a redirect the HTTP request helper does not follow.
+      const narrowed = rendered(
+        await (
+          await request.get(
+            `/campanha/comunicacao/acervo?source=enviadas&q=${marker}` +
+              `&year=2025&topic=saude&municipality=${municipality.id}&duration=curta`,
+          )
+        ).text(),
+      )
+      expect(narrowed).toContain(`/gravacoes/${health.id}?`)
+      expect(narrowed).not.toContain(`/gravacoes/${culture.id}?`)
+      expect(narrowed).not.toContain(`/gravacoes/${noDuration.id}?`)
+
+      // A contradictory combination is an honest empty state with the clear CTA.
+      const empty = rendered(
+        await (
+          await request.get(
+            `/campanha/comunicacao/acervo?source=enviadas&q=${marker}&year=2023&topic=saude`,
+          )
+        ).text(),
+      )
+      expect(empty).toContain('Nenhuma gravação encontrada')
+      expect(empty).toContain('Limpar filtros')
+
+      // Duration orders list the measured rows only, longest first.
+      const longestFirst = rendered(
+        await (
+          await request.get(
+            `/campanha/comunicacao/acervo?source=enviadas&q=${marker}&sort=duracao_maior`,
+          )
+        ).text(),
+      )
+      const cultureAt = longestFirst.indexOf(`/gravacoes/${culture.id}?`)
+      const healthAt = longestFirst.indexOf(`/gravacoes/${health.id}?`)
+      expect(cultureAt).toBeGreaterThan(-1)
+      expect(healthAt).toBeGreaterThan(cultureAt)
+      expect(longestFirst).not.toContain(`/gravacoes/${noDuration.id}?`)
+      expect(longestFirst).toContain('Gravações sem duração aparecem apenas em Mais recentes.')
+    })
+
+    test('theme mode degrades honestly on the recordings source too (C219)', async ({
+      campaign,
+      campaignRequest,
+    }) => {
+      const marker = campaign.fixtures.value('gravacaotemadeg')
+      await createRecording(campaign, { marker })
+
+      const user = await campaign.fixtures.createCampaignUser('communicator')
+      const request = await campaignRequest(user, user.password)
+
+      // The e2e environment blanks DEEPSEEK_API_KEY (playwright.config.ts), so
+      // the expansion is unavailable and the page must fall back with the notice.
+      const themed = await request.get(
+        `/campanha/comunicacao/acervo?source=enviadas&q=${marker}&mode=tema`,
+      )
+      expect(themed.status()).toBe(200)
+      const themedHtml = rendered(await themed.text())
+      expect(themedHtml).toContain('data-testid="recording-theme-fallback"')
+      expect(themedHtml).toContain('A busca por tema está indisponível agora.')
+      expect(themedHtml).toContain('Resultados por termo exato')
+      expect(themedHtml).toContain('Tentar por tema novamente')
+      expect(themedHtml).toContain(marker)
+      expect(themedHtml).toContain('<mark')
+
+      // The default mode stays untouched: no mode param means no notice.
+      const exact = await request.get(`/campanha/comunicacao/acervo?source=enviadas&q=${marker}`)
+      expect(exact.status()).toBe(200)
+      expect(rendered(await exact.text())).not.toContain(
+        'A busca por tema está indisponível agora.',
+      )
+    })
+
     test('labels a cluster through the JSON route and denies the non-acervo roles', async ({
       campaign,
       campaignRequest,

tests/int/recording.int.spec.ts

@@ -32,6 +32,7 @@ import {
   labelRecordingSpeakerForActor,
   retryRecordingForActor,
 } from '@/app/(campaign)/campanha/actions/recording'
+import { getMunicipalityCatalogEntry } from '@/lib/municipalityCatalog'
 import { RECORDING_MAX_BYTES } from '@/lib/recording'
 import {
   RECORDING_FORBIDDEN_MESSAGE,
@@ -41,6 +42,7 @@ import { normalizeForSearch } from '@/lib/speechSearch'
 import type { CampaignUser, RecordingMedia } from '@/payload-types'
 import config from '@/payload.config'
 import { hookFilledCreateData } from '@/utilities/hookFilledData'
+import { classifyRecordingFacets } from '@/utilities/recordings/recordingClassification'
 import {
   reapStaleRecording,
   RECORDING_STALE_MS,
@@ -100,6 +102,12 @@ const createRecording = async ({
   segments = [],
   speakerLabels,
   withMedia = true,
+  recordedAt,
+  durationSeconds,
+  topics,
+  scopes,
+  classifiedBy,
+  mentionedMunicipalities,
 }: {
   status?: 'uploading' | 'processing' | 'ready' | 'failed'
   title?: string
@@ -111,6 +119,12 @@ const createRecording = async ({
   }[]
   speakerLabels?: { speakerKey: string; label: string }[]
   withMedia?: boolean
+  recordedAt?: string
+  durationSeconds?: number
+  topics?: ('saude' | 'cultura' | 'educacao')[]
+  scopes?: ('bahia' | 'brasil' | 'internacional')[]
+  classifiedBy?: 'gazetteer' | 'llm' | 'manual'
+  mentionedMunicipalities?: number[]
 } = {}) => {
   const media = withMedia ? await createMedia(`${title}.mp4`) : null
   const recording = await payload.create({
@@ -120,6 +134,12 @@ const createRecording = async ({
       status,
       ...(media ? { media: media.id } : {}),
       ...(speakerLabels ? { speakerLabels } : {}),
+      ...(recordedAt ? { recordedAt } : {}),
+      ...(durationSeconds !== undefined ? { durationSeconds } : {}),
+      ...(topics ? { topics } : {}),
+      ...(scopes ? { scopes } : {}),
+      ...(classifiedBy ? { classifiedBy } : {}),
+      ...(mentionedMunicipalities ? { mentionedMunicipalities } : {}),
     },
     overrideAccess: true,
   })
@@ -431,6 +451,201 @@ describe('uploaded recordings (C199)', () => {
     expect(segments.docs[0]?.searchText).toBe('bom dia a todas')
   })
 
+  it('classifies the transcript with the injected classifier and derives the year', async () => {
+    const fixtures = campaignFixtures()
+    const municipality = await fixtures.getMunicipality()
+    const catalogEntry = getMunicipalityCatalogEntry(municipality.slug)
+    expect(catalogEntry).toBeDefined()
+
+    const { recording } = await createRecording({
+      status: 'processing',
+      title: 'Plenária sobre saúde',
+    })
+    await payload.update({
+      collection: 'recording',
+      id: recording.id,
+      data: { recordedAt: '2024-03-10T12:00:00.000Z' },
+      overrideAccess: true,
+    })
+
+    const restore = withEnv('FFMPEG_PATH', FAKE_FFMPEG)
+    try {
+      await runRecordingJob(
+        payload,
+        recording.id,
+        async () => ({
+          ok: true,
+          segments: [{ start: 1, end: 4, text: 'Falamos sobre a saúde no município' }],
+          durationSeconds: 120,
+        }),
+        null,
+        async (input) => {
+          expect(input.transcript).toContain('saúde no município')
+          return {
+            facets: {
+              topics: ['saude'],
+              scopes: ['bahia'],
+              municipalities: [catalogEntry!],
+              people: [],
+              programs: [],
+              projects: [],
+              classifiedBy: 'llm',
+            },
+            llm: { used: true, totalTokens: 42, estimatedCostUsd: 0.001, error: null },
+          }
+        },
+      )
+    } finally {
+      restore()
+    }
+
+    const updated = await payload.findByID({
+      collection: 'recording',
+      id: recording.id,
+      depth: 0,
+      overrideAccess: true,
+    })
+    expect(updated.status).toBe('ready')
+    expect(updated.year).toBe(2024)
+    expect(updated.topics).toEqual(['saude'])
+    expect(updated.scopes).toEqual(['bahia'])
+    expect(updated.classifiedBy).toBe('llm')
+    expect(updated.mentionedMunicipalities).toEqual([municipality.id])
+  })
+
+  it('never overwrites a manual facet curation with the job result', async () => {
+    const { recording } = await createRecording({
+      status: 'processing',
+      title: 'Curadoria manual',
+    })
+    await payload.update({
+      collection: 'recording',
+      id: recording.id,
+      data: { topics: ['cultura'], scopes: ['brasil'], classifiedBy: 'manual' },
+      overrideAccess: true,
+    })
+
+    const restore = withEnv('FFMPEG_PATH', FAKE_FFMPEG)
+    try {
+      await runRecordingJob(
+        payload,
+        recording.id,
+        async () => ({
+          ok: true,
+          segments: [{ start: 1, end: 4, text: 'Falamos sobre saúde' }],
+          durationSeconds: 120,
+        }),
+        null,
+        async () => ({
+          facets: {
+            topics: ['saude'],
+            scopes: ['bahia'],
+            municipalities: [],
+            people: [],
+            programs: [],
+            projects: [],
+            classifiedBy: 'llm',
+          },
+          llm: { used: true, totalTokens: 10, estimatedCostUsd: 0, error: null },
+        }),
+      )
+    } finally {
+      restore()
+    }
+
+    const updated = await payload.findByID({
+      collection: 'recording',
+      id: recording.id,
+      depth: 0,
+      overrideAccess: true,
+    })
+    expect(updated.status).toBe('ready')
+    expect(updated.topics).toEqual(['cultura'])
+    expect(updated.scopes).toEqual(['brasil'])
+    expect(updated.classifiedBy).toBe('manual')
+  })
+
+  it('lets an authenticated actor overwrite a manual curation (the escape hatch)', async () => {
+    const { recording } = await createRecording({
+      status: 'ready',
+      title: 'Curadoria corrigida no admin',
+      topics: ['cultura'],
+      scopes: ['brasil'],
+      classifiedBy: 'manual',
+    })
+
+    const updated = await payload.update({
+      collection: 'recording',
+      id: recording.id,
+      data: { topics: ['saude'], scopes: ['bahia'], classifiedBy: 'llm' },
+      user: communicator,
+      overrideAccess: false,
+    })
+
+    expect(updated.topics).toEqual(['saude'])
+    expect(updated.scopes).toEqual(['bahia'])
+    expect(updated.classifiedBy).toBe('llm')
+  })
+
+  it('keeps the transcript and claims no provenance when the classifier throws', async () => {
+    const { recording } = await createRecording({
+      status: 'processing',
+      title: 'Classificador fora do ar',
+    })
+
+    const restore = withEnv('FFMPEG_PATH', FAKE_FFMPEG)
+    try {
+      await runRecordingJob(
+        payload,
+        recording.id,
+        async () => ({
+          ok: true,
+          segments: [{ start: 1, end: 4, text: 'Bom dia a todas' }],
+          durationSeconds: 120,
+        }),
+        null,
+        async () => {
+          throw new Error('classificador fora do ar')
+        },
+      )
+    } finally {
+      restore()
+    }
+
+    const updated = await payload.findByID({
+      collection: 'recording',
+      id: recording.id,
+      depth: 0,
+      overrideAccess: true,
+    })
+    expect(updated.status).toBe('ready')
+    // The fake ffmpeg emits two chunks and the injected transcriber repeats the
+    // answer for both, exactly like the end-to-end job test above.
+    expect(updated.searchText).toBe('bom dia a todas bom dia a todas')
+    expect(updated.classifiedBy).toBeNull()
+    expect(updated.topics ?? []).toEqual([])
+  })
+
+  it('resolves the gazetteer municipality without the LLM key (degraded provenance)', async () => {
+    const fixtures = campaignFixtures()
+    const municipality = await fixtures.getMunicipality()
+    const catalogEntry = getMunicipalityCatalogEntry(municipality.slug)
+    expect(catalogEntry).toBeDefined()
+
+    const restore = withEnv('DEEPINFRA_API_KEY', undefined)
+    try {
+      const classification = await classifyRecordingFacets({
+        payload,
+        transcript: `Falamos em ${catalogEntry!.city} sobre saúde e educação`,
+      })
+      expect(classification?.classifiedBy).toBe('gazetteer')
+      expect(classification?.topics).toEqual(expect.arrayContaining(['saude', 'educacao']))
+      expect(classification?.mentionedMunicipalities).toContain(municipality.id)
+    } finally {
+      restore()
+    }
+  })
+
   it('marks a provider failure as failed, preserves the file and retries', async () => {
     const { recording } = await createRecording({ status: 'processing', title: 'Debate' })
 
@@ -604,6 +819,123 @@ describe('uploaded recordings (C199)', () => {
     expect(options.people).toContain(`Dep. ${marker}`)
   })
 
+  it('filters by the C219 facets and lists their options', async () => {
+    const fixtures = campaignFixtures()
+    const municipality = await fixtures.getMunicipality()
+    const marker = `facetas${Date.now().toString(36)}`
+
+    const { recording: health } = await createRecording({
+      title: `Plenária de saúde ${marker}`,
+      recordedAt: '2025-05-04T10:00:00.000Z',
+      durationSeconds: 90,
+      topics: ['saude'],
+      scopes: ['bahia'],
+      classifiedBy: 'llm',
+      mentionedMunicipalities: [municipality.id],
+      segments: [{ startSeconds: 0, endSeconds: 5, text: `Saúde em pauta ${marker}` }],
+    })
+    const { recording: culture } = await createRecording({
+      title: `Debate de cultura ${marker}`,
+      recordedAt: '2023-01-01T10:00:00.000Z',
+      durationSeconds: 400,
+      topics: ['cultura'],
+      segments: [{ startSeconds: 0, endSeconds: 5, text: `Cultura em pauta ${marker}` }],
+    })
+    const { recording: noDuration } = await createRecording({
+      title: `Material sem duração ${marker}`,
+      segments: [{ startSeconds: 0, endSeconds: 5, text: `Ainda processando ${marker}` }],
+    })
+
+    const load = (extra: Record<string, string>) =>
+      loadRecordingsPageData(payload, communicator, {
+        source: 'enviadas',
+        q: marker,
+        ...extra,
+      })
+
+    expect((await load({ year: '2025' })).rows.map((row) => row.id)).toEqual([health.id])
+    expect((await load({ topic: 'saude' })).rows.map((row) => row.id)).toEqual([health.id])
+    expect((await load({ scope: 'bahia' })).rows.map((row) => row.id)).toEqual([health.id])
+    expect(
+      (await load({ municipality: String(municipality.id) })).rows.map((row) => row.id),
+    ).toEqual([health.id])
+    expect((await load({ duration: 'curta' })).rows.map((row) => row.id)).toEqual([health.id])
+    expect((await load({ duration: 'longa' })).rows.map((row) => row.id)).toEqual([culture.id])
+    expect((await load({ duration: 'sem_duracao' })).rows.map((row) => row.id)).toEqual([
+      noDuration.id,
+    ])
+
+    const options = await loadRecordingFilterOptions(payload, communicator)
+    expect(options.years).toContain(2025)
+    expect(options.municipalities.map((option) => option.value)).toContain(String(municipality.id))
+  })
+
+  it('orders by duration only over rows with a measured duration', async () => {
+    const marker = `ordem${Date.now().toString(36)}`
+    const { recording: short } = await createRecording({
+      title: `Curta ${marker}`,
+      durationSeconds: 60,
+      segments: [{ startSeconds: 0, endSeconds: 5, text: `Trecho ${marker}` }],
+    })
+    const { recording: long } = await createRecording({
+      title: `Longa ${marker}`,
+      durationSeconds: 900,
+      segments: [{ startSeconds: 0, endSeconds: 5, text: `Trecho ${marker}` }],
+    })
+    await createRecording({
+      title: `Sem duração ${marker}`,
+      segments: [{ startSeconds: 0, endSeconds: 5, text: `Trecho ${marker}` }],
+    })
+
+    const load = (sort?: string) =>
+      loadRecordingsPageData(payload, communicator, {
+        source: 'enviadas',
+        q: marker,
+        ...(sort ? { sort } : {}),
+      })
+
+    expect((await load('duracao_maior')).rows.map((row) => row.id)).toEqual([long.id, short.id])
+    expect((await load('duracao_menor')).rows.map((row) => row.id)).toEqual([short.id, long.id])
+    // The default order keeps every row (the gate only applies to duration orders).
+    expect((await load()).rows.length).toBe(3)
+  })
+
+  it('searches by theme with the injected expansion and degrades honestly', async () => {
+    const marker = `tema${Date.now().toString(36)}`
+    const { recording } = await createRecording({
+      title: `Merenda ${marker}`,
+      segments: [
+        { startSeconds: 0, endSeconds: 6, text: `Falamos de alimentação escolar ${marker}.` },
+      ],
+    })
+
+    const expanded = await loadRecordingsPageData(
+      payload,
+      communicator,
+      { source: 'enviadas', q: 'merenda escolar', mode: 'tema' },
+      async (theme) => {
+        expect(theme).toBe('merenda escolar')
+        return { terms: ['alimentacao escolar'] }
+      },
+    )
+    expect(expanded.themeApplied).toBe(true)
+    expect(expanded.themeUnavailable).toBe(false)
+    const row = expanded.rows.find((item) => item.id === recording.id)
+    expect(row).toBeDefined()
+    expect(row?.excerpt?.parts.some((part) => part.highlighted)).toBe(true)
+
+    const degraded = await loadRecordingsPageData(
+      payload,
+      communicator,
+      { source: 'enviadas', q: 'merenda escolar', mode: 'tema' },
+      async () => null,
+    )
+    expect(degraded.themeApplied).toBe(false)
+    expect(degraded.themeUnavailable).toBe(true)
+    // Honest fallback: the literal phrase does not occur, so the row is out.
+    expect(degraded.rows.map((item) => item.id)).not.toContain(recording.id)
+  })
+
   it('labels one cluster through the action and rebuilds the facet names', async () => {
     const marker = `falante${Date.now().toString(36)}`
     const { recording } = await createRecording({

tests/unit/e2eAffectedManifest.unit.spec.ts

@@ -75,6 +75,10 @@ describe('E2E_AFFECTED_MANIFEST (OPS5)', () => {
       // route and the vote share sheet are new contracts over the C211
       // collection (curated only).
       'frontendConteudos',
+      // C219 — deliberate: the recordings acervo gained its own search/facet
+      // contract and the migration makes every PR of this delivery high-risk
+      // (curated only).
+      'campaignSpeechAcervo',
     ])
     const onDisk = new Set(specNamesOnDisk())
     for (const spec of E2E_CURATED_SPECS) {

tests/unit/expandSpeechSearchTheme.unit.spec.ts

@@ -12,6 +12,7 @@ vi.mock('@ai-sdk/deepseek', () => ({ deepSeek: deepSeekMock }))
 
 import {
   expandContentPieceSearchTheme,
+  expandRecordingSearchTheme,
   expandSearchTheme,
   expandSpeechSearchTheme,
 } from '@/utilities/ai/expandSpeechSearchTheme'
@@ -98,4 +99,15 @@ describe('expandSearchTheme (S28 — corpus)', () => {
     await expect(expandSearchTheme('defesa do SUS', 'contentPiece')).resolves.toBeNull()
     expect(generateObjectMock).toHaveBeenCalledTimes(1)
   })
+
+  it('uses the recordings prompt for the enviadas corpus (C219)', async () => {
+    generateObjectMock.mockResolvedValue({ object: { terms: ['merenda escolar'] } })
+
+    await expandRecordingSearchTheme('merenda')
+
+    const call = generateObjectMock.mock.calls[0]![0] as { system: string }
+    expect(call.system).toContain('gravações enviadas pela equipe')
+    expect(call.system).not.toContain('acervo de discursos')
+    expect(call.system).not.toContain('Central de Conteúdos')
+  })
 })

tests/unit/recordingListFilters.unit.spec.ts

@@ -41,4 +41,80 @@ describe('recording list filters (C199/C200)', () => {
       ],
     })
   })
+
+  it('ANDs the C219 facets (year, topic, scope, cited municipality)', () => {
+    expect(
+      buildRecordingListWhere({
+        source: 'enviadas',
+        page: 1,
+        years: [2026],
+        topics: ['saude'],
+        scopes: ['bahia'],
+        municipalities: [42],
+      }),
+    ).toEqual({
+      and: [
+        { year: { in: [2026] } },
+        { topics: { in: ['saude'] } },
+        { scopes: { in: ['bahia'] } },
+        { mentionedMunicipalities: { in: [42] } },
+      ],
+    })
+  })
+
+  it('maps each duration bucket to the shared Câmara predicate', () => {
+    expect(buildRecordingListWhere({ source: 'enviadas', page: 1, durations: ['curta'] })).toEqual({
+      durationSeconds: { less_than: 120 },
+    })
+    expect(buildRecordingListWhere({ source: 'enviadas', page: 1, durations: ['media'] })).toEqual({
+      durationSeconds: { greater_than_equal: 120, less_than: 300 },
+    })
+    expect(buildRecordingListWhere({ source: 'enviadas', page: 1, durations: ['longa'] })).toEqual({
+      durationSeconds: { greater_than_equal: 300 },
+    })
+    expect(
+      buildRecordingListWhere({ source: 'enviadas', page: 1, durations: ['sem_duracao'] }),
+    ).toEqual({ durationSeconds: { exists: false } })
+
+    // Two buckets OR together inside the AND.
+    expect(
+      buildRecordingListWhere({ source: 'enviadas', page: 1, durations: ['curta', 'longa'] }),
+    ).toEqual({
+      or: [
+        { durationSeconds: { less_than: 120 } },
+        { durationSeconds: { greater_than_equal: 300 } },
+      ],
+    })
+  })
+
+  it('ORs the literal query with the expanded theme terms, deduped', () => {
+    expect(
+      buildRecordingListWhere({ source: 'enviadas', page: 1, q: 'saúde' }, [
+        'acesso universal à saúde',
+        'saúde',
+      ]),
+    ).toEqual({
+      or: [{ searchText: { like: 'saude' } }, { searchText: { like: 'acesso universal a saude' } }],
+    })
+  })
+
+  it('gates rows without a measured duration out of the duration orders', () => {
+    expect(buildRecordingListWhere({ source: 'enviadas', page: 1, sort: 'duracao_maior' })).toEqual(
+      { durationSeconds: { exists: true } },
+    )
+
+    expect(
+      buildRecordingListWhere({
+        source: 'enviadas',
+        page: 1,
+        sort: 'duracao_menor',
+        q: 'merenda',
+      }),
+    ).toEqual({
+      and: [{ durationSeconds: { exists: true } }, { searchText: { like: 'merenda' } }],
+    })
+
+    // The default order carries no gate.
+    expect(buildRecordingListWhere({ source: 'enviadas', page: 1 })).toEqual({})
+  })
 })

tests/unit/recordingListUrl.unit.spec.ts

@@ -2,16 +2,22 @@
 
 import { describe, expect, it } from 'vitest'
 
+import { SPEECH_TOPICS } from '@/lib/speechFacets'
 import {
   buildAcervoSourceHref,
+  buildRecordingFiltersKey,
   buildRecordingListHref,
   parseAcervoSource,
   parseRecordingListParams,
   parseRecordingPeople,
+  recordingHasActiveFilters,
+  recordingSortIsDuration,
+  recordingSortOrder,
   removeRecordingPerson,
   resolveRecordingListUrl,
   toggleRecordingPerson,
 } from '@/utilities/recordings/recordingListUrl'
+import { SPEECH_DURATION_BUCKETS } from '@/utilities/speech/speechListUrl'
 
 describe('recording list URL (C199)', () => {
   it('detects the enviadas source and defaults everything else to the Câmara', () => {
@@ -44,8 +50,8 @@ describe('recording list URL (C199)', () => {
     const polluted = resolveRecordingListUrl({
       source: 'enviadas',
       q: 'merenda',
-      year: '2026',
-      topic: 'saude',
+      phase: 'Ordem do Dia',
+      foo: 'bar',
     })
     expect(polluted.redirectHref).toBe('/campanha/comunicacao/acervo?source=enviadas&q=merenda')
 
@@ -130,4 +136,128 @@ describe('recording list URL (C199)', () => {
       people: undefined,
     })
   })
+
+  it('parses the C219 parity facets and serializes them in canonical order', () => {
+    const canonical = resolveRecordingListUrl({
+      source: 'enviadas',
+      q: 'merenda',
+      mode: 'tema',
+      year: ['2026', '2024'],
+      topic: 'saude',
+      scope: 'bahia',
+      municipality: '42',
+      duration: 'curta',
+      sort: 'duracao_maior',
+      person: 'Solla',
+      page: '2',
+    })
+    expect(canonical.state).toEqual({
+      source: 'enviadas',
+      page: 2,
+      q: 'merenda',
+      mode: 'tema',
+      years: [2026, 2024],
+      topics: ['saude'],
+      scopes: ['bahia'],
+      municipalities: [42],
+      durations: ['curta'],
+      sort: 'duracao_maior',
+      people: ['Solla'],
+    })
+    expect(canonical.href).toBe(
+      '/campanha/comunicacao/acervo?source=enviadas&q=merenda&mode=tema&year=2026&year=2024' +
+        '&topic=saude&scope=bahia&municipality=42&duration=curta&sort=duracao_maior&person=Solla&page=2',
+    )
+    expect(canonical.redirectHref).toBeUndefined()
+  })
+
+  it('keeps the C199 deep link byte-identical (no new param is serialized)', () => {
+    const canonical = resolveRecordingListUrl({
+      source: 'enviadas',
+      q: 'merenda',
+      person: ['Solla', 'Presidente'],
+      page: '2',
+    })
+    expect(canonical.href).toBe(
+      '/campanha/comunicacao/acervo?source=enviadas&q=merenda&person=Solla&person=Presidente&page=2',
+    )
+    expect(canonical.redirectHref).toBeUndefined()
+  })
+
+  it('canonicalizes the default sort and a theme without a query away', () => {
+    const defaultSort = resolveRecordingListUrl({ source: 'enviadas', sort: 'recentes' })
+    expect(defaultSort.state.sort).toBeUndefined()
+    expect(defaultSort.href).toBe('/campanha/comunicacao/acervo?source=enviadas')
+    expect(defaultSort.redirectHref).toBe(defaultSort.href)
+
+    // The parsed state keeps `mode`, but the canonical href drops it and the
+    // redirect lands on the URL without the meaningless param (same as Câmara).
+    const themeWithoutQuery = resolveRecordingListUrl({ source: 'enviadas', mode: 'tema' })
+    expect(themeWithoutQuery.state.mode).toBe('tema')
+    expect(themeWithoutQuery.href).toBe('/campanha/comunicacao/acervo?source=enviadas')
+    expect(themeWithoutQuery.redirectHref).toBe(themeWithoutQuery.href)
+
+    // An unknown sort value can never widen the contract: it canonicalizes away.
+    const unknownSort = resolveRecordingListUrl({ source: 'enviadas', sort: 'aleatorio' })
+    expect(unknownSort.state.sort).toBeUndefined()
+    expect(unknownSort.redirectHref).toBe(unknownSort.href)
+  })
+
+  it('collapses an exhaustive enum selection to no filter (same rule as the Câmara)', () => {
+    const allTopics = resolveRecordingListUrl({
+      source: 'enviadas',
+      topic: SPEECH_TOPICS.map(({ value }) => value),
+    })
+    expect(allTopics.state.topics).toBeUndefined()
+    expect(allTopics.href).toBe('/campanha/comunicacao/acervo?source=enviadas')
+
+    const allDurations = resolveRecordingListUrl({
+      source: 'enviadas',
+      duration: SPEECH_DURATION_BUCKETS.map(({ value }) => value),
+    })
+    expect(allDurations.state.durations).toBeUndefined()
+    expect(allDurations.href).toBe('/campanha/comunicacao/acervo?source=enviadas')
+  })
+
+  it('drops structurally invalid year/municipality values on canonicalization', () => {
+    const canonical = resolveRecordingListUrl({
+      source: 'enviadas',
+      year: ['1800', 'abc', '2026'],
+      municipality: ['-3', '0', '7'],
+    })
+    expect(canonical.state.years).toEqual([2026])
+    expect(canonical.state.municipalities).toEqual([7])
+    expect(canonical.href).toBe(
+      '/campanha/comunicacao/acervo?source=enviadas&year=2026&municipality=7',
+    )
+  })
+
+  it('exposes the canonical filter key and the Payload sort of each order', () => {
+    expect(
+      buildRecordingFiltersKey({ source: 'enviadas', page: 1, q: 'merenda', topics: ['saude'] }),
+    ).toBe('source=enviadas&q=merenda&topic=saude')
+
+    expect(recordingSortOrder({ source: 'enviadas', page: 1 })).toBe('-createdAt')
+    expect(recordingSortOrder({ source: 'enviadas', page: 1, sort: 'duracao_maior' })).toBe(
+      '-durationSeconds',
+    )
+    expect(recordingSortOrder({ source: 'enviadas', page: 1, sort: 'duracao_menor' })).toBe(
+      'durationSeconds',
+    )
+    expect(recordingSortIsDuration({ source: 'enviadas', page: 1 })).toBe(false)
+    expect(recordingSortIsDuration({ source: 'enviadas', page: 1, sort: 'duracao_maior' })).toBe(
+      true,
+    )
+  })
+
+  it('tells an unfiltered page (including pagination) from a narrowed one', () => {
+    expect(recordingHasActiveFilters({ source: 'enviadas', page: 1 })).toBe(false)
+    expect(recordingHasActiveFilters({ source: 'enviadas', page: 4 })).toBe(false)
+    expect(recordingHasActiveFilters({ source: 'enviadas', page: 1, sort: 'duracao_maior' })).toBe(
+      false,
+    )
+    expect(recordingHasActiveFilters({ source: 'enviadas', page: 1, q: 'merenda' })).toBe(true)
+    expect(recordingHasActiveFilters({ source: 'enviadas', page: 1, years: [2026] })).toBe(true)
+    expect(recordingHasActiveFilters({ source: 'enviadas', page: 1, people: ['Solla'] })).toBe(true)
+  })
 })

tests/unit/recordingOmnibox.unit.spec.ts

@@ -0,0 +1,197 @@
+// @vitest-environment node
+
+import { describe, expect, it } from 'vitest'
+
+import {
+  applyRecordingOmniboxSuggestion,
+  applyRecordingSearchMode,
+  buildRecordingOmniboxChips,
+  buildRecordingOmniboxSuggestionSeeds,
+  clearRecordingOmnibox,
+  filterRecordingOmniboxSuggestions,
+  removeRecordingOmniboxChip,
+} from '@/utilities/recordings/recordingOmnibox'
+
+const labels = new Map<number, string>([[42, 'Feira de Santana']])
+
+describe('recordings omnibox (C219)', () => {
+  it('builds one chip per active filter with the facet labels', () => {
+    const chips = buildRecordingOmniboxChips({
+      state: {
+        source: 'enviadas',
+        page: 1,
+        q: 'merenda',
+        years: [2026],
+        topics: ['saude'],
+        scopes: ['bahia'],
+        municipalities: [42],
+        durations: ['curta'],
+        people: ['Dep. Jorge Solla'],
+      },
+      municipalityLabelsById: labels,
+    })
+    expect(chips.map((chip) => chip.label)).toEqual([
+      'Busca: merenda',
+      'Ano: 2026',
+      'Tema: Saúde',
+      'Alcance: Bahia',
+      'Município citado: Feira de Santana',
+      'Duração: Até 2 min',
+      'Pessoa: Dep. Jorge Solla',
+    ])
+  })
+
+  it('falls back to the id when the municipality label is not loaded', () => {
+    const chips = buildRecordingOmniboxChips({
+      state: { source: 'enviadas', page: 1, municipalities: [99] },
+      municipalityLabelsById: labels,
+    })
+    expect(chips).toEqual([{ id: 'municipality:99', label: 'Município citado: Município #99' }])
+  })
+
+  it('applies the query suggestion resetting the page and keeping the source', () => {
+    const action = applyRecordingOmniboxSuggestion({
+      state: { source: 'enviadas', page: 4 },
+      suggestionId: 'q:merenda escolar',
+    })
+    expect(action).toEqual({
+      kind: 'url',
+      state: { source: 'enviadas', page: 1, q: 'merenda escolar' },
+    })
+  })
+
+  it('toggles facets and people (person dedupes case-insensitively)', () => {
+    const base = { source: 'enviadas' as const, page: 3 }
+
+    expect(applyRecordingOmniboxSuggestion({ state: base, suggestionId: 'topic:saude' })).toEqual({
+      kind: 'url',
+      state: { source: 'enviadas', page: 1, topics: ['saude'] },
+    })
+
+    expect(applyRecordingOmniboxSuggestion({ state: base, suggestionId: 'year:2026' })).toEqual({
+      kind: 'url',
+      state: { source: 'enviadas', page: 1, years: [2026] },
+    })
+
+    expect(
+      applyRecordingOmniboxSuggestion({ state: base, suggestionId: 'municipality:42' }),
+    ).toEqual({ kind: 'url', state: { source: 'enviadas', page: 1, municipalities: [42] } })
+
+    const withPerson = applyRecordingOmniboxSuggestion({
+      state: base,
+      suggestionId: 'person:Dep. Jorge Solla',
+    })
+    expect(withPerson).toEqual({
+      kind: 'url',
+      state: { source: 'enviadas', page: 1, people: ['Dep. Jorge Solla'] },
+    })
+
+    const removedPerson = applyRecordingOmniboxSuggestion({
+      state: { source: 'enviadas', page: 1, people: ['Dep. Jorge Solla'] },
+      suggestionId: 'person:dep. jorge solla',
+    })
+    expect(removedPerson).toEqual({
+      kind: 'url',
+      state: { source: 'enviadas', page: 1, people: undefined },
+    })
+
+    // An invalid numeric id never mutates the state.
+    expect(applyRecordingOmniboxSuggestion({ state: base, suggestionId: 'year:abc' })).toEqual({
+      kind: 'url',
+      state: base,
+    })
+  })
+
+  it('removes chips (dropping the query also drops the theme mode)', () => {
+    expect(
+      removeRecordingOmniboxChip({
+        state: { source: 'enviadas', page: 2, q: 'merenda', mode: 'tema' },
+        chipId: 'q',
+      }),
+    ).toEqual({
+      kind: 'url',
+      state: { source: 'enviadas', page: 1, q: undefined, mode: undefined },
+    })
+
+    expect(
+      removeRecordingOmniboxChip({
+        state: { source: 'enviadas', page: 2, topics: ['saude', 'cultura'] },
+        chipId: 'topic:saude',
+      }),
+    ).toEqual({ kind: 'url', state: { source: 'enviadas', page: 1, topics: ['cultura'] } })
+
+    expect(
+      removeRecordingOmniboxChip({
+        state: { source: 'enviadas', page: 2, people: ['Solla'] },
+        chipId: 'person:Solla',
+      }),
+    ).toEqual({ kind: 'url', state: { source: 'enviadas', page: 1, people: undefined } })
+  })
+
+  it('keeps a person label with a colon intact through the chip round trip', () => {
+    const state = { source: 'enviadas' as const, page: 1, people: ['Ana: assessora'] }
+    expect(buildRecordingOmniboxChips({ state, municipalityLabelsById: labels })).toEqual([
+      { id: 'person:Ana: assessora', label: 'Pessoa: Ana: assessora' },
+    ])
+    expect(removeRecordingOmniboxChip({ state, chipId: 'person:Ana: assessora' })).toEqual({
+      kind: 'url',
+      state: { source: 'enviadas', page: 1, people: undefined },
+    })
+  })
+
+  it('toggles the theme mode only with a query and clears back to the source', () => {
+    expect(
+      applyRecordingSearchMode({
+        state: { source: 'enviadas', page: 2, q: 'merenda' },
+        mode: 'tema',
+      }),
+    ).toEqual({
+      kind: 'url',
+      state: { source: 'enviadas', page: 1, q: 'merenda', mode: 'tema' },
+    })
+
+    expect(
+      applyRecordingSearchMode({ state: { source: 'enviadas', page: 1 }, mode: 'tema' }),
+    ).toEqual({ kind: 'url', state: { source: 'enviadas', page: 1 } })
+
+    expect(
+      applyRecordingSearchMode({
+        state: { source: 'enviadas', page: 1, q: 'merenda', mode: 'tema' },
+        mode: 'termo',
+      }),
+    ).toEqual({ kind: 'url', state: { source: 'enviadas', page: 1, q: 'merenda' } })
+
+    expect(
+      clearRecordingOmnibox({ source: 'enviadas', page: 3, q: 'merenda', topics: ['saude'] }),
+    ).toEqual({ kind: 'clear', state: { source: 'enviadas', page: 1 } })
+  })
+
+  it('seeds the taxonomy up front and the data-driven options by keyword', () => {
+    const seeds = buildRecordingOmniboxSuggestionSeeds({
+      years: [2026],
+      municipalityOptions: [{ value: '42', label: 'Feira de Santana' }],
+      personOptions: ['Dep. Jorge Solla'],
+    })
+
+    const emptyQuery = filterRecordingOmniboxSuggestions(seeds, '')
+    expect(emptyQuery.some((suggestion) => suggestion.id === 'topic:saude')).toBe(true)
+    expect(emptyQuery.some((suggestion) => suggestion.id === 'scope:bahia')).toBe(true)
+    expect(emptyQuery.some((suggestion) => suggestion.id === 'duration:curta')).toBe(true)
+    // Years, municipalities and people are searchable, not shown up front.
+    expect(emptyQuery.some((suggestion) => suggestion.id === 'year:2026')).toBe(false)
+    expect(emptyQuery.some((suggestion) => suggestion.id === 'person:Dep. Jorge Solla')).toBe(false)
+
+    // The typed query itself is always the first suggestion; the facet seed
+    // follows it when it matches.
+    expect(filterRecordingOmniboxSuggestions(seeds, 'feira').map((s) => s.id)).toContain(
+      'municipality:42',
+    )
+    expect(filterRecordingOmniboxSuggestions(seeds, '2026').map((s) => s.id)).toContain('year:2026')
+    expect(filterRecordingOmniboxSuggestions(seeds, 'falante').map((s) => s.id)).toContain(
+      'person:Dep. Jorge Solla',
+    )
+    expect(filterRecordingOmniboxSuggestions(seeds, 'municipio').map((s) => s.id)).toContain(
+      'municipality:42',
+    )
+  })
+})