fix(C180): select enxuto dos loaders, teto do id in da origem e paridade busca↔VM (Closes #1097)

docs/changelog/2026-09-16-c180.md

@@ -0,0 +1 @@
+**C180 — Acervo cortes: dívida pós-C174 — select enxuto, teto do `id in` e paridade busca↔VM (2026-09-16):** três achados do `/simplify` do C174 atacados em `src/utilities/speech/`, sem mudança visual. (1) **Select enxuto + DTO estreito:** `loadSpeechCutsForSpeech`/`loadSpeechCutsForSpeeches`/`loadSpeechCutOriginSpeechIds` deixam de trazer a linha inteira do corte; novo `SpeechCutSummaryViewModel` (`lib/speechCut.ts`) entrega só `id`/`status`/`title`/`description`/`durationSeconds` — o que "Cortes desta fala" e o aninhamento da busca renderizam — reusando o fallback de duração e a normalização `status` desconhecido→`failed`, e o `select` espelha os campos do mapper. (2) **Predicado único busca↔VM (bug corrigido):** `speechMatchesSearchQuery` puro em `lib/speechSearch.ts` passa a ser o espelho do ramo textual do `where`; o `keywords` do VM vira case-insensitive como o ILIKE `%q%` do Postgres, então uma fala cujo keyword casa em caixa diferente deixa de ser marcada como "Fala de origem" indevidamente. (3) **Teto do `id in` da origem:** `SPEECH_CUT_ORIGIN_SPEECH_ID_LIMIT = 200` + `originSpeechIdsOfCuts` (dedupe + cap puro) limitam a query de origem e o `id in` do acervo, com degradação honesta (menos resultados origin-only, nunca resultado errado); gatilho de revisão ~5–10 mil cortes (C168). Sem collection, Consent ou migration; loaders seguem `user` + `overrideAccess:false`; `/corte/<id>`, o acervo depth-1 e o dialog/player intocados. Testes: unit de `toSpeechCutSummaryViewModel`, `originSpeechIdsOfCuts` (dedupe/cap/`limit: 0`) e `speechMatchesSearchQuery` (acento/caixa); int do loader com chaves exatas, do keyword case-insensitive e da origem opção B; e2e curado `campaignSpeechAcervo` + `campaignSpeechCut` verde.

docs/plans/acervo-cortes-divida-pos-c174-impl.md

@@ -0,0 +1,99 @@
+# Impl: Acervo cortes — dívida pós-C174: select enxuto dos loaders, teto do `id in` da origem e paridade busca↔VM
+
+Status: aprovado
+Atualizado em: 2026-09-16
+Issue: #1097
+Intenção: docs/plans/acervo-cortes-divida-pos-c174.md
+Appetite restante: herdado (~0,5–1 dia)
+
+## Leitura da intenção
+
+- **Outcome:** cortar o custo de query dos três loaders de corte e o risco de drift entre a busca (SQL) e o VM, sem mudar comportamento visível.
+- **O que NÃO negociar:** nenhuma superfície visual nova; `lib/` nunca importa `utilities/`; sem módulo gêmeo; sem migration; `loadSpeechCutAcervoPageData` (depth 1), `loadSpeechCutDetailPageData`/`findSpeechCutForActor` e `/corte/<id>` intocados.
+- **O que reavaliar (surpresas do explorer):**
+  - `SpeechNestedCuts` com `showDescription` renderiza `description` (`SpeechNestedCuts.tsx:62-64`) → o DTO estreito PRECISA de `description`.
+  - As superfícies usam `durationSeconds` + `formatSpeechClock`, não `durationLabel` (`formatSpeechSpan`) → manter `durationSeconds`; NÃO incluir `durationLabel`.
+  - `failureMessage`/`error`/`step` não são lidos por nenhuma das duas superfícies → saem do DTO; mas a normalização `status` desconhecido → `'failed'` fica (senão `SpeechCutStatusBadge` quebra).
+  - `tests/int/speechCut.int.spec.ts:860-870` asserta `mediaUrl`/`publicPath` no retorno desse loader → muda com o novo shape.
+
+## Abordagem recomendada
+
+```mermaid
+flowchart LR
+  L[loaders: select enxuto + limit] --> S[toSpeechCutSummaryViewModel]
+  S --> A[SpeechCutsForSpeechSection]
+  S --> B[SpeechNestedCuts + SpeechListItemViewModel.cuts]
+  P[lib/speechSearch: predicado puro] --> VM[speechViewModels.speechMatchesSearchQuery]
+  P -. mesma semântica .-> W[speechListFilters Where]
+  C[SPEECH_CUT_ORIGIN_SPEECH_ID_LIMIT + originSpeechIdsOfCuts] --> L
+```
+
+**Opções consideradas:** A — DTO estreito + predicado puro compartilhado + cap do `id in` (as três fases) | B — só adicionar `select` mantendo `toSpeechCutViewModel` completo | C — F3 via subquery/raw e F2 via comparar por id já resolvido.
+**Recomendação:** A — remove o payload não usado no shape (não só no tráfego), unifica a semântica num só lugar puro e fecha o `in` com degradação honesta; menor caminho que cobre as três fases.
+**Rejeitadas:** B — `select` enxuto com VM completo ainda mapeia `mediaUrl`/`youtubeVideoId`/`failureMessage` que ninguém lê e exigiria `error` no select; C — subquery exige SQL cru fora do repositório Payload (quebra depth) e comparar por id erra quando a fala casa por texto E é origem de um corte.
+
+### Componentes / mudanças
+
+- **`SpeechCutSummaryViewModel` + `toSpeechCutSummaryViewModel` + `cutDurationSeconds`** (`src/lib/speechCut.ts`): mapper puro que emite `{ id, status, title, description, durationSeconds }`, reusando o fallback `durationSeconds ?? end − start` (extraído de `toSpeechCutViewModel`) e a normalização de status. `toSpeechCutViewModel` passa a usar o helper extraído.
+- **`speechMatchesSearchQuery`** (`src/lib/speechSearch.ts`): predicado puro do ramo textual; `speechViewModels` deixa de ter `speechMatchesText` local e chama este.
+- **`SPEECH_CUT_ORIGIN_SPEECH_ID_LIMIT` + `originSpeechIdsOfCuts`** (`src/lib/speechCut.ts`): dedupe + cap puro dos ids de fala de origem.
+- **`speechCutPageData.ts`**: `speechCutSummarySelect` (status/title/description/durationSeconds/startSeconds/endSeconds) nos dois loaders de corte; `speechCutOriginSelect` (`speech`) + `limit` na origem; retornos passam a `SpeechCutSummaryViewModel`; origem usa `originSpeechIdsOfCuts`.
+- **`speechViewModels.ts` / `SpeechCutsForSpeechSection.tsx` / `SpeechNestedCuts.tsx`**: tipos de prop passam a `SpeechCutSummaryViewModel` (render inalterado).
+- **Migration:** sem migration — nenhuma collection/global/field muda; só `select`, `limit` e DTO de leitura.
+- **Access / Consent:** n/a — os loaders seguem `user` + `overrideAccess:false`; nenhum write path.
+- **UI:** Impeccable A; nenhuma superfície visual nova nem copy nova → nenhum trigger do `designer` (a/b/c/d) é acionado.
+
+### Dados → forma
+
+- Não se aplica (sem KPI/série/mapa; o dado já é lista).
+
+## Fases verificáveis
+
+1. **F1 — DTO estreito + `select`** (`lib/speechCut.ts`, `speechCutPageData.ts`, 3 componentes). Prova: `tests/unit/speechCut.unit.spec.ts` novo `describe('toSpeechCutSummaryViewModel')` (chaves exatas sem `mediaUrl`/`failureMessage`; fallback `end − start`; status desconhecido → `failed`); int `speechCut.int.spec.ts:860-870` atualizado para as chaves do summary.
+2. **F2 — predicado único** (`lib/speechSearch.ts`, `speechViewModels.ts`). Prova: `tests/unit/speechSearch.unit.spec.ts` `describe('speechMatchesSearchQuery')` (searchText acento/caixa; keyword case-insensitive; query vazia/`undefined` → true) + int novo em `speechCut.int.spec.ts` com `keywords` em caixa diferente do `q` (o falso "Fala de origem" de hoje vira `matchedTextSearch: true`).
+3. **F3 — cap do `id in`** (`lib/speechCut.ts`, `speechCutPageData.ts`). Prova: unit de `originSpeechIdsOfCuts` (dedupe, relação não resolvida, corta no `limit`); int existente `speechCut.int.spec.ts:883-915` mantém a fiação ponta-a-ponta.
+4. **Gates:** `pnpm gate:fast` na iteração; entrega com `pnpm push`. E2E local discricionário; o manifest dispara o conjunto curado `campaignSpeechAcervo` + `campaignSpeechCut` (`scripts/lib/e2e-affected-manifest.mjs:300-314`) e o CI/PR cobre o blast radius.
+
+## Rabbit holes / Não escopo (engenharia)
+
+- Não estreitar `loadSpeechCutAcervoPageData` (`speechCutPageData.ts:42-70`, depth 1 precisa de origem/media) nem `loadSpeechCutDetailPageData`/`findSpeechCutForActor`.
+- Não tocar `toSpeechCutViewModel` além de extrair o helper de duração; dialog/player/result card seguem com o VM completo.
+- Não mexer no índice/`searchText` de `speechCut` (gatilho de volume do C168) nem no rename `matchedTextSearch`.
+- Não introduzir `durationLabel`/`formatSpeechSpan` nas duas superfícies (mudaria a saída).
+- Lista fixa de "Explicitamente fora" da intenção permanece fechada.
+
+## Riscos e mitigação
+
+- **Select enxuto quebrar o VM:** o summary não lê `media`/`error`/`step`; risco morre por construção — unit do mapper + int de chaves.
+- **Predicado divergir do SQL:** documentar a semântica (`searchText` normalizado dos dois lados; `keywords` case-insensitive mas acento-significante, como o ILIKE `%q%`) e fixar com unit + int.
+- **Cap degradar a busca:** o cap só corta a superfície "fala de origem" (menos falas origin-only), nunca torna resultado errado — o flag vem do predicado, não dos ids. Volume-gatilho ~5–10 mil cortes.
+
+## Aceite de engenharia
+
+- [ ] Aceite de produto da intenção ainda coberto (nenhuma saída visual muda)
+- [ ] Invariantes AGENTS/engineering-standards (depth `lib/`→`utilities/`, sem top-level novo, sem migration, pt-BR/identificadores EN)
+- [ ] Testes de domínio previstos (unit dos mappers/predicado/cap + int de loader) — sem write path novo
+
+## Decisões de engenharia
+
+- **D1 — DTO:** novo `SpeechCutSummaryViewModel` + `toSpeechCutSummaryViewModel` em `lib/speechCut.ts`. Inclui `description` (exigido pelo `showDescription` do `SpeechNestedCuts`), mantém `durationSeconds` (a superfície usa `formatSpeechClock`) e a normalização `status` desconhecido → `'failed'` (senão o badge quebra). Rejeitados: manter `toSpeechCutViewModel` completo (mapeia campos não lidos) e um DTO sem `description` (perde a evidência da CENA 05). Saem `failureMessage`/`step`/`media*`/`publicPath`/`youtubeVideoId`/`publishedAt`.
+- **D2 — predicado único:** `speechMatchesSearchQuery` puro em `src/lib/speechSearch.ts` (lib não importa utilities; ambos os consumidores já importam `normalizeForSearch`), consumido pelo VM; o `where` segue única fonte do SQL. `searchText` casa normalizado/acento-insensível dos dois lados; `keywords` passa a `toLowerCase().includes()` para casar o ILIKE (case-insensitive) mantendo acento significante — corrige o falso "Fala de origem" de hoje. Rejeitados: comparar por id já resolvido (erra quando a fala casa por texto E é origem) e duplicar o predicado.
+- **D3 — teto:** `SPEECH_CUT_ORIGIN_SPEECH_ID_LIMIT = 200` como `limit` da query de origem + cap no `originSpeechIdsOfCuts`; degradação honesta = menos resultados "só por corte", nunca resultado errado. Rejeitadas: subquery/raw (fora do Payload, quebra depth) e deixar sem teto (o `id in` cresce sem limite com termo comum; gatilho ~5–10 mil cortes/C168).
+
+## Explicitamente fora / Adiado com gatilho (triage do /simplify)
+
+- **Select ↔ mapper ligados por convenção (S1):** `speechCutSummarySelect` e `SpeechCutSummaryRecord` são mantidos em passo por comentário cruzado + o int exact-keys; um campo novo sem select key faz o int/unit falharem (não é `undefined` silencioso). Link compile-time (chave tipada derivando os dois) fica adiado — gatilho: o mapper ganhar um 2º consumidor ou o int deixar de cobrir o shape.
+- **Paridade predicado ↔ `where` só por teste de branch (S6):** `speechListFilters.unit.spec.ts` prende o conjunto exato de branches (`toEqual`), então adicionar/remover branch falha e força revisitar o predicado (doc aponta o par). Teste de paridade automático adiado — gatilho: uma 3ª dimensão textual entrar na busca de falas.
+- **`originSpeechIdsOfCuts` com 1 call site (S3):** mantido como dono puro do cap/dedupe (testável sem DB); o `limit` default duplica o da query de propósito (invariante). Gatilho para reavaliar: 3º call site ou 2º loader precisar dos ids de origem.
+- **Int exact-keys (S10):** mantido de propósito — é o único guard do contrato loader-select↔mapper (S1); reavaliar só se S1 ganhar link compile-time.
+- **Já resolvido no simplify (não reabrir):** `SpeechCutSummaryRecord` via `Pick` (S2); doc do `speechSearch` (S4); JSDoc do `originSpeechIdsOfCuts` (S5); export morto do record (S7); off-by-one de `limit: 0` (S8); tipo `CutDurationRecord & Pick` (S9); nome/fronteira do teste (S11).
+
+## Self-score (decision-quality ≥4)
+
+1. **Decisões caras com rejeitadas — 4/5:** D1–D3 nomeiam opção, porquê e rejeitadas, com shape de DTO, semântica de ILIKE e teto resolvidos.
+2. **Cabe no appetite — 5/5:** 5 arquivos de produção + 5 de teste; reusa mappers/loaders/int existentes.
+3. **Rabbit holes nomeados — 5/5:** lista fixa herdada + os loaders que NÃO podem ser estreitados.
+4. **Depth check — 5/5:** `utilities/speech/` segue dono; predicado e mappers puros em `lib/`; nenhum gêmeo.
+5. **Intenção de produto — 5/5:** zero mudança visível; só custo de query, paridade e teto.
+
+**Score: 5/5 (mínimo 4).**

src/components/campaign/speech/SpeechCutsForSpeechSection.tsx

@@ -6,7 +6,7 @@ import { SpeechExcerptCta } from '@/components/campaign/speech/SpeechExcerptCta'
 import { Button } from '@/components/ui/button'
 import { campaignSpeechCutDetailHref } from '@/lib/campaignPaths'
 import { formatSpeechClock } from '@/lib/speechClock'
-import type { SpeechCutViewModel } from '@/lib/speechCut'
+import type { SpeechCutSummaryViewModel } from '@/lib/speechCut'
 
 const SECTION_TITLE = 'Cortes desta fala'
 const SECTION_DESCRIPTION = 'Abra um corte existente para editar ou reaproveitar.'
@@ -38,7 +38,7 @@ export const SpeechCutsForSpeechSection = ({
   cuts,
   excerptSelectionAvailable,
 }: {
-  cuts: readonly SpeechCutViewModel[]
+  cuts: readonly SpeechCutSummaryViewModel[]
   excerptSelectionAvailable: boolean
 }) => {
   if (cuts.length === 0) {

src/components/campaign/speech/SpeechNestedCuts.tsx

@@ -4,14 +4,14 @@ import { SpeechCutStatusBadge } from '@/components/campaign/speech/SpeechCutStat
 import { SpeechHighlightParts } from '@/components/campaign/speech/SpeechHighlightParts'
 import { campaignSpeechCutDetailHref } from '@/lib/campaignPaths'
 import { formatSpeechClock } from '@/lib/speechClock'
-import type { SpeechCutViewModel } from '@/lib/speechCut'
+import type { SpeechCutSummaryViewModel } from '@/lib/speechCut'
 import { splitHighlightedParts } from '@/lib/speechHighlight'
 
 const countLabel = (count: number): string =>
   count === 1 ? '1 corte desta fala' : `${count} cortes desta fala`
 
 /** Touch-sized outlined button on mobile; the compact text link on `md+`. */
-const CutAction = ({ cut }: { cut: SpeechCutViewModel }) => (
+const CutAction = ({ cut }: { cut: SpeechCutSummaryViewModel }) => (
   <Link
     href={campaignSpeechCutDetailHref(cut.id)}
     aria-label={`Abrir corte: ${cut.title}`}
@@ -34,7 +34,7 @@ export const SpeechNestedCuts = ({
   query,
   showDescription = false,
 }: {
-  cuts: readonly SpeechCutViewModel[]
+  cuts: readonly SpeechCutSummaryViewModel[]
   caption?: string
   /** The search term, highlighted in the title. */
   query?: string

src/lib/speechCut.ts

@@ -263,12 +263,42 @@ export type SpeechCutViewModel = {
   failureMessage: string | null
 }
 
+/**
+ * C180 — the narrow view of a cut for the discovery surfaces ("Cortes desta
+ * fala" and the search nesting): everything they render, nothing they don't
+ * (`media*`/`publicPath`/`youtubeVideoId`/`failureMessage`/`step` stay out).
+ */
+export type SpeechCutSummaryViewModel = {
+  id: number
+  status: SpeechCutStatus
+  title: string
+  description: string
+  /** Stored span, stable even if the speech row changes; falls back to end − start. */
+  durationSeconds: number
+}
+
+/**
+ * The shape the summary mapper needs. Payload does not reflect `select` in the
+ * result type, so this must stay field-for-field with `speechCutSummarySelect`
+ * in `utilities/speech/speechCutPageData.ts`.
+ */
+type SpeechCutSummaryRecord = Pick<
+  SpeechCutRecordForView,
+  'id' | 'status' | 'title' | 'description' | 'startSeconds' | 'endSeconds' | 'durationSeconds'
+>
+
 const mediaOf = (media: SpeechCutRecordForView['media']): CutMediaRecord | null =>
   typeof media === 'object' && media !== null ? media : null
 
 const youtubeUrlOf = (speech: SpeechCutRecordForView['speech']): string | null =>
   typeof speech === 'object' && speech !== null ? (speech.youtubeUrl ?? null) : null
 
+/** Stored span when present, else the raw [start, end] fallback, floored at 0. */
+const cutDurationSeconds = (record: SpeechCutSummaryRecord): number =>
+  typeof record.durationSeconds === 'number' && Number.isFinite(record.durationSeconds)
+    ? Math.max(0, Math.round(record.durationSeconds))
+    : Math.max(0, Math.round(record.endSeconds) - Math.round(record.startSeconds))
+
 /**
  * Internal/status wire view of a cut. `error` and `createdBy` never pass
  * through here: the public page and the dialog receive only what they render —
@@ -278,10 +308,7 @@ export const toSpeechCutViewModel = (record: SpeechCutRecordForView): SpeechCutV
   const media = mediaOf(record.media)
   const status = isSpeechCutStatus(record.status) ? record.status : 'failed'
   const step = isSpeechCutStep(record.step) ? record.step : null
-  const durationSeconds =
-    typeof record.durationSeconds === 'number' && Number.isFinite(record.durationSeconds)
-      ? Math.max(0, Math.round(record.durationSeconds))
-      : Math.max(0, Math.round(record.endSeconds) - Math.round(record.startSeconds))
+  const durationSeconds = cutDurationSeconds(record)
 
   return {
     id: record.id,
@@ -303,6 +330,48 @@ export const toSpeechCutViewModel = (record: SpeechCutRecordForView): SpeechCutV
   }
 }
 
+/**
+ * C180 — narrow view of a cut for the discovery surfaces, reusing the same
+ * duration fallback and status normalization as the full view (an unknown
+ * status still degrades to `failed`, so the badge never receives a stray value).
+ */
+export const toSpeechCutSummaryViewModel = (
+  record: SpeechCutSummaryRecord,
+): SpeechCutSummaryViewModel => ({
+  id: record.id,
+  status: isSpeechCutStatus(record.status) ? record.status : 'failed',
+  title: record.title,
+  description: record.description,
+  durationSeconds: cutDurationSeconds(record),
+})
+
+/**
+ * C180 — volume ceiling for the origin `id in` the acervo ORs into its textual
+ * `where`. A common term can match thousands of cuts; the cap degrades the
+ * "origin-only" surface (fewer origin speeches), never the result's correctness
+ * — the "Fala de origem" flag comes from the predicate, not from these ids.
+ * Revisit when the catalog approaches ~5–10k cuts (same trigger as C168).
+ */
+export const SPEECH_CUT_ORIGIN_SPEECH_ID_LIMIT = 200
+
+/**
+ * Deduped origin speech ids of a batch of cuts, capped at `limit`. Depth 0
+ * keeps an unresolved relation as the raw id; a populated relation contributes
+ * its `.id`, and a missing/null relation is skipped.
+ */
+export const originSpeechIdsOfCuts = (
+  cuts: readonly { speech?: number | CutSpeechRecord | null }[],
+  limit = SPEECH_CUT_ORIGIN_SPEECH_ID_LIMIT,
+): number[] => {
+  const ids = new Set<number>()
+  for (const cut of cuts) {
+    if (ids.size >= limit) break
+    const speechId = typeof cut.speech === 'number' ? cut.speech : cut.speech?.id
+    if (typeof speechId === 'number') ids.add(speechId)
+  }
+  return [...ids]
+}
+
 /** Origin of a cut in the library: the speech it was cut from, with its link. */
 export type SpeechCutOriginViewModel = {
   id: number

src/lib/speechSearch.ts

@@ -14,6 +14,31 @@ export const normalizeForSearch = (value: string): string =>
     .replace(/\s+/g, ' ')
     .trim()
 
+type SpeechSearchableRecord = {
+  /** Normalized segment text persisted by the segment hook (C153). */
+  searchText?: string | null
+  keywords?: readonly string[] | null
+}
+
+/**
+ * C180 — pure mirror of the textual `where` branch
+ * (`speechListFilters.buildSpeechTextBranches`): normalized `searchText` LIKE
+ * the normalized query, OR a raw official keyword containing the query.
+ * `keywords` compare case-insensitively (accent still significant) to match the
+ * ILIKE `%q%` Postgres runs. Adding or removing a branch there means changing
+ * this predicate and its test in the same commit.
+ */
+export const speechMatchesSearchQuery = (
+  speech: SpeechSearchableRecord,
+  query: string | undefined,
+): boolean => {
+  const trimmed = query?.trim()
+  if (!trimmed) return true
+  if (normalizeForSearch(speech.searchText ?? '').includes(normalizeForSearch(trimmed))) return true
+  const lowered = trimmed.toLowerCase()
+  return (speech.keywords ?? []).some((keyword) => keyword.toLowerCase().includes(lowered))
+}
+
 /**
  * Dedupes values by their normalized form, preserving the first spelling.
  * Used by the facet classifier before persisting people/program mentions.

src/utilities/speech/speechCutPageData.ts

@@ -3,10 +3,12 @@ import 'server-only'
 import type { Payload } from 'payload'
 
 import {
+  originSpeechIdsOfCuts,
+  SPEECH_CUT_ORIGIN_SPEECH_ID_LIMIT,
   toSpeechCutLibraryItemViewModel,
-  toSpeechCutViewModel,
+  toSpeechCutSummaryViewModel,
   type SpeechCutLibraryItemViewModel,
-  type SpeechCutViewModel,
+  type SpeechCutSummaryViewModel,
 } from '@/lib/speechCut'
 import type { CampaignUser } from '@/payload-types'
 import { createEntityNotFoundError } from '@/utilities/entityNotFound'
@@ -21,6 +23,26 @@ import {
 
 type SpeechCutListSearchParams = Record<string, string | string[] | undefined>
 
+/**
+ * C180 — lean `select` for the discovery surfaces: only the fields the summary
+ * view model renders, so the loaders stop hauling media/error/step rows. Keep
+ * field-for-field with `SpeechCutSummaryRecord` (`lib/speechCut.ts`); Payload
+ * does not reflect `select` in the result type, so the int test is the guard.
+ */
+const speechCutSummarySelect = {
+  status: true,
+  title: true,
+  description: true,
+  durationSeconds: true,
+  startSeconds: true,
+  endSeconds: true,
+  // Kept so the batch loader can group rows by their origin speech.
+  speech: true,
+} as const
+
+/** The origin loader only needs the speech relation to dedupe the origins. */
+const speechCutOriginSelect = { speech: true } as const
+
 export const SpeechCutNotFoundError = createEntityNotFoundError(
   'SpeechCut',
   'Corte não encontrado.',
@@ -72,15 +94,15 @@ export const loadSpeechCutAcervoPageData = async (
 /**
  * C174 — every cut made from one speech, newest first, for the "Cortes desta
  * fala" section of the speech detail. Depth 0 is enough: the origin here IS the
- * page, and the section renders title/status/duration only (`error` still feeds
- * the honest `failureMessage`, never the raw cause). No page cap: the number of
- * cuts per speech is bounded by human effort.
+ * page, and the section renders title/status/duration only — the summary select
+ * keeps the query to what it shows. No page cap: the number of cuts per speech
+ * is bounded by human effort.
  */
 export const loadSpeechCutsForSpeech = async (
   payload: Payload,
   user: CampaignUser,
   speechId: number,
-): Promise<SpeechCutViewModel[]> => {
+): Promise<SpeechCutSummaryViewModel[]> => {
   const result = await payload.find({
     collection: 'speechCut',
     where: { speech: { equals: speechId } },
@@ -88,24 +110,26 @@ export const loadSpeechCutsForSpeech = async (
     limit: 0,
     pagination: false,
     sort: '-createdAt',
+    select: speechCutSummarySelect,
     user,
     overrideAccess: false,
   })
 
-  return result.docs.map((cut) => toSpeechCutViewModel(cut))
+  return result.docs.map((cut) => toSpeechCutSummaryViewModel(cut))
 }
 
 /**
  * C174 — every cut of every speech on the acervo page, grouped by speech id, so
  * the result list nests them under the speech without a query per row. Same
- * precedent as `loadSegmentsForSpeeches`; depth 0 keeps the origin as the id.
+ * precedent as `loadSegmentsForSpeeches`; depth 0 keeps the origin as the id and
+ * the summary select carries only what the nested card renders.
  */
 export const loadSpeechCutsForSpeeches = async (
   payload: Payload,
   user: CampaignUser,
   speechIds: readonly number[],
-): Promise<Map<number, SpeechCutViewModel[]>> => {
-  const bySpeech = new Map<number, SpeechCutViewModel[]>()
+): Promise<Map<number, SpeechCutSummaryViewModel[]>> => {
+  const bySpeech = new Map<number, SpeechCutSummaryViewModel[]>()
   if (speechIds.length === 0) return bySpeech
 
   const result = await payload.find({
@@ -115,6 +139,7 @@ export const loadSpeechCutsForSpeeches = async (
     limit: 0,
     pagination: false,
     sort: '-createdAt',
+    select: speechCutSummarySelect,
     user,
     overrideAccess: false,
   })
@@ -123,7 +148,7 @@ export const loadSpeechCutsForSpeeches = async (
     const speechId = typeof cut.speech === 'number' ? cut.speech : cut.speech?.id
     if (typeof speechId !== 'number') continue
     const list = bySpeech.get(speechId) ?? []
-    list.push(toSpeechCutViewModel(cut))
+    list.push(toSpeechCutSummaryViewModel(cut))
     bySpeech.set(speechId, list)
   }
   return bySpeech
@@ -133,6 +158,10 @@ export const loadSpeechCutsForSpeeches = async (
  * C174 (option B) — the origin speech ids of the cuts whose own title/description
  * match `q`. The acervo search uses them to surface a speech through a cut even
  * when the speech text itself does not contain the term.
+ *
+ * C180 — the query is capped at `SPEECH_CUT_ORIGIN_SPEECH_ID_LIMIT`: a common
+ * term must not let the acervo's `id in [...]` grow without bound. The cap only
+ * trims origin-only hits; it never makes a returned row wrong.
  */
 export const loadSpeechCutOriginSpeechIds = async (
   payload: Payload,
@@ -143,18 +172,14 @@ export const loadSpeechCutOriginSpeechIds = async (
     collection: 'speechCut',
     where: buildSpeechCutListWhere({ page: 1, q }),
     depth: 0,
-    limit: 0,
+    limit: SPEECH_CUT_ORIGIN_SPEECH_ID_LIMIT,
     pagination: false,
+    select: speechCutOriginSelect,
     user,
     overrideAccess: false,
   })
 
-  const ids = new Set<number>()
-  for (const cut of result.docs) {
-    const speechId = typeof cut.speech === 'number' ? cut.speech : cut.speech?.id
-    if (typeof speechId === 'number') ids.add(speechId)
-  }
-  return [...ids]
+  return originSpeechIdsOfCuts(result.docs)
 }
 
 /** One cut with its origin and stored media, for the library detail page. */

src/utilities/speech/speechListFilters.ts

@@ -54,7 +54,12 @@ const buildSpeechTextBranches = (q: string): Where[] => [
   { keywords: { contains: q } },
 ]
 
-/** The textual branch: normalized speech text OR a raw official keyword. */
+/**
+ * The textual branch: normalized speech text OR a raw official keyword. The
+ * keyword `contains` compiles to ILIKE `%q%` — case-insensitive, accent still
+ * significant. The view-model mirror is `lib/speechSearch.speechMatchesSearchQuery`
+ * (C180); a semantics change here has to land there too.
+ */
 const buildSpeechTextWhere = (q: string): Where => ({ or: buildSpeechTextBranches(q) })
 
 export const buildSpeechListWhere = (state: SpeechListState): Where => {

src/utilities/speech/speechViewModels.ts

@@ -5,7 +5,7 @@
  */
 import { CAMPAIGN_COMMUNICATION_ACERVO } from '@/lib/campaignPaths'
 import { formatSpeechClock, formatSpeechDate, formatSpeechSpan } from '@/lib/speechClock'
-import type { SpeechCutViewModel } from '@/lib/speechCut'
+import type { SpeechCutSummaryViewModel } from '@/lib/speechCut'
 import type { SpeechExcerptSegment } from '@/lib/speechExcerpt'
 import type { SpeechScope, SpeechTopic } from '@/lib/speechFacets'
 import {
@@ -16,7 +16,7 @@ import {
   type SpeechHighlightedExcerpt,
   type SpeechHighlightPart,
 } from '@/lib/speechHighlight'
-import { normalizeForSearch } from '@/lib/speechSearch'
+import { normalizeForSearch, speechMatchesSearchQuery } from '@/lib/speechSearch'
 import {
   correctedExcerptOffsetSeconds,
   excerptOffsetSeconds,
@@ -74,7 +74,7 @@ export type SpeechListItemViewModel = {
    */
   officialTextUrl: string | null
   /** C174 — the cuts already made from this speech, newest first. */
-  cuts: SpeechCutViewModel[]
+  cuts: SpeechCutSummaryViewModel[]
   /**
    * C174 — false when the speech surfaced only because one of its cuts matched
    * the term; the card then swaps the excerpt for the honest origin note.
@@ -196,19 +196,7 @@ export const parseSpeechSeekSeconds = (raw: string | undefined): number | null =
   return Number.isFinite(seconds) && seconds >= 0 ? seconds : null
 }
 
-type SpeechCuts = readonly SpeechCutViewModel[]
-
-/**
- * C174 — does the speech itself match `q`? Mirrors the textual `where` branch
- * (`speechListFilters.buildSpeechTextBranches`): normalized `searchText` LIKE
- * and raw keyword contains. Keep the two spellings in step.
- */
-const speechMatchesText = (speech: SpeechListRecord, query: string | undefined): boolean => {
-  const trimmed = query?.trim()
-  if (!trimmed) return true
-  if (normalizeForSearch(speech.searchText ?? '').includes(normalizeForSearch(trimmed))) return true
-  return (speech.keywords ?? []).some((keyword) => keyword.includes(trimmed))
-}
+type SpeechCuts = readonly SpeechCutSummaryViewModel[]
 
 export const toSpeechListItemViewModel = ({
   speech,
@@ -251,7 +239,7 @@ export const toSpeechListItemViewModel = ({
     watchHref: buildWatchHref(speech.id, matchedSegment, q),
     officialTextUrl: speech.officialTextUrl ?? null,
     cuts: [...cuts],
-    matchedTextSearch: speechMatchesText(speech, query),
+    matchedTextSearch: speechMatchesSearchQuery(speech, query),
   }
 }
 

tests/int/speechCut.int.spec.ts

@@ -863,9 +863,15 @@ describe('speech cut discovery (C174)', () => {
     const forSpeech = await loadSpeechCutsForSpeech(payload, communicator, speech)
     expect(forSpeech.map((cut) => cut.id)).toEqual([newer, older])
     expect(forSpeech.map((cut) => cut.id)).not.toContain(unrelated)
-    // Depth 0: the row carries no media/file internals.
-    expect(forSpeech[0]?.mediaUrl).toBeNull()
-    expect(forSpeech[0]?.publicPath).toBe(`/corte/${newer}`)
+    // C180 — the summary view carries only what the section renders (no media/file internals).
+    expect(Object.keys(forSpeech[0] ?? {}).sort()).toEqual([
+      'description',
+      'durationSeconds',
+      'id',
+      'status',
+      'title',
+    ])
+    expect(forSpeech[0]?.durationSeconds).toBe(20)
 
     const grouped = await loadSpeechCutsForSpeeches(payload, communicator, [speech, otherSpeech])
     expect(grouped.get(speech)?.map((cut) => cut.id)).toEqual([newer, older])
@@ -916,4 +922,19 @@ describe('speech cut discovery (C174)', () => {
     expect(row).toBeDefined()
     expect(row?.cuts).toHaveLength(0)
   })
+
+  it('flags a speech whose keyword matches case-insensitively as a regular result', async () => {
+    // C180 — the SQL `contains` is ILIKE; the view-model mirror follows the same
+    // case-insensitive semantics instead of showing the origin-only note wrongly.
+    const marker = `saudepublica${randomUUID().slice(0, 8)}`
+    const speech = await createSpeech({
+      keywords: [`SAÚDE ${marker.toUpperCase()}`],
+      segments: [{ startSeconds: 0, endSeconds: 2.7, text: `Discurso ${randomUUID()}` }],
+    })
+    const { communicator } = await createUsers()
+
+    const data = await loadSpeechAcervoPageData(payload, communicator, { q: `saúde ${marker}` })
+    const row = data.rows.find((candidate) => candidate.id === speech)
+    expect(row?.matchedTextSearch).toBe(true)
+  })
 })

tests/unit/speechCut.unit.spec.ts

@@ -6,6 +6,7 @@ import {
   buildSpeechCutFallbackMetadata,
   buildSpeechCutFfmpegArgs,
   buildSpeechCutShare,
+  originSpeechIdsOfCuts,
   SPEECH_CUT_FAILURE_GENERATING,
   SPEECH_CUT_FAILURE_INTERRUPTED,
   SPEECH_CUT_FAILURE_SPEECH_GONE,
@@ -14,6 +15,7 @@ import {
   speechCutFailureMessage,
   speechCutPublicPath,
   speechCutStepStates,
+  toSpeechCutSummaryViewModel,
   toSpeechCutViewModel,
 } from '@/lib/speechCut'
 
@@ -175,6 +177,69 @@ describe('toSpeechCutViewModel', () => {
   })
 })
 
+describe('toSpeechCutSummaryViewModel', () => {
+  it('emits only the fields the discovery surfaces render', () => {
+    const summary = toSpeechCutSummaryViewModel({
+      id: 7,
+      status: 'published',
+      title: 'Acesso a medicamentos',
+      description: 'Descrição',
+      startSeconds: 43,
+      endSeconds: 118,
+      durationSeconds: 75,
+    })
+    expect(summary).toEqual({
+      id: 7,
+      status: 'published',
+      title: 'Acesso a medicamentos',
+      description: 'Descrição',
+      durationSeconds: 75,
+    })
+    expect(Object.keys(summary).sort()).toEqual([
+      'description',
+      'durationSeconds',
+      'id',
+      'status',
+      'title',
+    ])
+  })
+
+  it('falls back to end − start and degrades an unknown status to failed', () => {
+    const summary = toSpeechCutSummaryViewModel({
+      id: 1,
+      status: 'weird',
+      title: 't',
+      description: 'd',
+      startSeconds: 10,
+      endSeconds: 30,
+    })
+    expect(summary.status).toBe('failed')
+    expect(summary.durationSeconds).toBe(20)
+  })
+})
+
+describe('originSpeechIdsOfCuts', () => {
+  it('dedupes resolved ids and skips null/missing relations', () => {
+    expect(
+      originSpeechIdsOfCuts([
+        { speech: 7 },
+        { speech: 7 },
+        { speech: 9 },
+        { speech: null },
+        { speech: { id: 11 } },
+        {},
+      ]),
+    ).toEqual([7, 9, 11])
+  })
+
+  it('caps the list at the limit and returns nothing at limit 0', () => {
+    const cuts = Array.from({ length: 10 }, () => ({ speech: 1 }))
+    cuts.push({ speech: 2 }, { speech: 3 })
+    expect(originSpeechIdsOfCuts(cuts, 2)).toEqual([1, 2])
+    expect(originSpeechIdsOfCuts(cuts, 0)).toEqual([])
+  })
+})
+
 describe('speechCutFailureMessage', () => {
   it('maps the named causes of the first stage to honest pt-BR copy', () => {
     expect(speechCutFailureMessage({ error: SPEECH_CUT_FAILURE_SPEECH_GONE })).toBe(

tests/unit/speechSearch.unit.spec.ts

@@ -2,7 +2,11 @@
 
 import { describe, expect, it } from 'vitest'
 
-import { normalizeForSearch, uniqueByNormalizedForm } from '@/lib/speechSearch'
+import {
+  normalizeForSearch,
+  speechMatchesSearchQuery,
+  uniqueByNormalizedForm,
+} from '@/lib/speechSearch'
 
 describe('normalizeForSearch', () => {
   it('strips accents, lowercases and collapses whitespace', () => {
@@ -21,3 +25,25 @@ describe('uniqueByNormalizedForm', () => {
     expect(uniqueByNormalizedForm(['', '  ', 'Bahia'])).toEqual(['Bahia'])
   })
 })
+
+describe('speechMatchesSearchQuery', () => {
+  it('matches the normalized search text accent- and case-insensitively', () => {
+    const speech = { searchText: 'a saude publica baiana', keywords: [] }
+    expect(speechMatchesSearchQuery(speech, 'SAÚDE Pública')).toBe(true)
+    expect(speechMatchesSearchQuery(speech, 'educação')).toBe(false)
+  })
+
+  it('matches an official keyword case-insensitively, accent still significant', () => {
+    const speech = { searchText: '', keywords: ['Farmácia Popular', 'SUS'] }
+    expect(speechMatchesSearchQuery(speech, 'farmácia popular')).toBe(true)
+    expect(speechMatchesSearchQuery(speech, 'FARMÁCIA')).toBe(true)
+    // the SQL is ILIKE `%q%`: a different accent would not match there either
+    expect(speechMatchesSearchQuery(speech, 'farmacia')).toBe(false)
+  })
+
+  it('treats an empty or missing query as a match', () => {
+    const speech = { searchText: null, keywords: null }
+    expect(speechMatchesSearchQuery(speech, undefined)).toBe(true)
+    expect(speechMatchesSearchQuery(speech, '   ')).toBe(true)
+  })
+})