diff --git a/docs/changelog/2026-09-14-c162.md b/docs/changelog/2026-09-14-c162.md
new file mode 100644
index 00000000..6ccf6761
--- /dev/null
+++ b/docs/changelog/2026-09-14-c162.md
@@ -0,0 +1 @@
+**C162 — VOD do acervo: player resolve o trecho sob demanda e YouTube vira default quando existe (2026-09-14):** o detalhe da fala deixa de depender do MP4 efêmero gravado (`vod.camara.leg.br`, apagado pelo CDN) e escolhe a superfície pelas coordenadas da linha: **YouTube default** quando `youtubeUrl` parseia (embed `youtube-nocookie`, sem chamada à Câmara no render) e **capa "Assistir o trecho"** quando só há VOD armazenado, resolvendo no clique o trecho exato (`video-sob-demanda?idAudio=&trecho=<excerptTMs>`) com links **verificados antes de tocar/baixar**. Dono novo das regras puras: `src/lib/speechVod.ts` recebe `CAMARA_USER_AGENT`/`buildVodUrl`/`parseVodStatus`/`parseDurationToSeconds` (movidos de `scripts/lib/camaraSpeeches.mjs`, que passa a re-exportar; `camara:pilot` ganha o mesmo loader tsx do `camara:import`) e adiciona `parseYoutubeVideoId` (watch/`youtu.be`/embed/live/shorts) e `excerptOffsetSeconds` (epoch ms → BRT `Intl` menos o naive `eventStartAt`, literal da intenção 15:43:54−15:00=2634). Resolver server-only `speechVodResolver` faz **1 chamada bounded** (2 tentativas/15s, UA da Câmara) e probe Range `bytes=0-1023` em paralelo no playback e no download (fallback HEAD em 405/501; `text/html` = hash morto), devolvendo `pronto`/`gerando`/`indisponivel`; a action `resolveSpeechVodForActor` gateia `canReadSpeechCatalog` antes da leitura (`overrideAccess: false`, select só dos campos de elegibilidade) e recusa fala sem VOD armazenado sem tocar a rede; a rota `POST /campanha/comunicacao/acervo/resolver-vod` usa `campaignJsonMutationRoute` (mesma origem) com mensagens constantes em `src/lib/schemas/speechVod.ts`. O view model perde `downloadUrl`/`vodDownloadUrl` da lista e do detalhe (o link armazenado nunca é renderizado — é só sinal de elegibilidade) e ganha `youtubeVideoId`/`youtubeOffsetSeconds`/`vodResolvable`; `SpeechDetailPlayer` reescrito com superfície única por quadrante (iframe com seek por reload `start=offset+segmento` — degradação C admitida no gate, gatilho registrado —, `<video>` nativo 0-based pós-PRONTO, painéis honestos de `GERANDO`/falha/indisponível, `aria-busy`+live region no pending) e "Baixar vídeo (MP4)" abre aba síncrona e só aponta para URL verificada (nunca arquivo morto); o "Baixar" direto do card da lista sai (com o campo morto do view model). Sem migration/collection/`Consent`/S3. Testes: unit `speechVod` (11), `speechVodResolver` (6, fetch mockado), `speechDetailPlayer` (9, jsdom: quadrantes, pending, retry, seek) + `camaraSpeeches` verde pela re-exportação; int `speechAcervo` estendido (offset/vodResolvable, lista sem download); e2e `campaignSpeechAcervo` renegociado (11 verdes: card sem download e sem link VOD, ambos→YouTube com `start` somado, só VOD→capa, sem vídeo→indisponível sem retry, rota 401/403/400 sem rede). `gate:fast` (3061 unit) + e2e local verdes.
diff --git a/docs/plans/acervo-vod-sob-demanda-impl.md b/docs/plans/acervo-vod-sob-demanda-impl.md
new file mode 100644
index 00000000..760cab18
--- /dev/null
+++ b/docs/plans/acervo-vod-sob-demanda-impl.md
@@ -0,0 +1,165 @@
+# Impl: Acervo de falas: player resolve o trecho sob demanda e YouTube vira default quando existe
+
+Status: em execução
+Atualizado em: 2026-09-14
+Issue: #999
+Intenção: docs/plans/acervo-vod-sob-demanda.md
+Appetite restante: herdado (~1,5–2 dias eng; sem corte proposto — a fatia cabe porque não há schema, persistência nem IFrame API)
+
+## Leitura da intenção
+
+- **Outcome:** com o CDN da Câmara vazio, abrir uma fala ainda toca (YouTube quando existir; VOD do trecho exato resolvido no clique quando só houver VOD) e "Baixar" nunca entrega arquivo inexistente — o MP4 é resolvido e verificado na Câmara no clique.
+- **O que NÃO negociar:** render/RSC não chama a Câmara (resolução só no clique); `excerptTMs` verbatim (nunca vizinho); link do banco é cache (pode ser ignorado; nunca requisito); sem S3/espelho/transcode; sem segundo player/editor; sem migration/collection/`Consent` novos; crédito "Fonte: Câmara dos Deputados · CC BY 4.0" mantido; gate `canReadSpeechCatalog` vale para a resolução; URLs públicas do acervo intocadas; "Baixar" direto do card da lista sai; `PRONTO` com hash morto vira estado honesto, nunca arquivo morto.
+- **O que reavaliar (hipóteses da intenção):**
+  - Os helpers de `scripts/lib/camara*.mjs` não são importáveis de `src/` (`.mjs`, `node:crypto`, pilot em node puro) → decisão 1 extrai as regras puras para `src/lib/speechVod.ts`.
+  - `youtubeUrl` é `event.urlRegistro` do open data (formato variável) → parser defensivo novo (`parseYoutubeVideoId`), não um `slice` de querystring.
+  - A intenção **recomenda** IFrame API para o seek na transcrição, mas admite a degradação C (recarregar o iframe com `start=`) se não couber no appetite. Este plano **escolhe a degradação C** (decisão 4) e registra o gatilho de revisitação — divergência de engenharia permitida pela própria intenção, a destacar no gate.
+  - O draft usa "Não foi possível carregar o vídeo deste trecho" (falha) e a intenção usa "vídeo indisponível neste momento" (quadrante nenhum). O plano preserva as duas: quadrante sem oferta = indisponível sem "Tentar novamente"; falha/GERANDO após oferta = cena 4 com "Tentar novamente".
+
+## Abordagem recomendada
+
+```mermaid
+flowchart LR
+  A[Detalhe da fala] --> B{youtubeVideoId?}
+  B -- sim --> Y[iframe youtube-nocookie<br/>start = offset + t]
+  B -- não --> C{VOD armazenado +<br/>eventId/audioId/excerptTMs?}
+  C -- não --> N[Bloco indisponível<br/>+ Abrir fonte]
+  C -- sim --> P[Capa só VOD<br/>'Assistir o trecho']
+  Y --> D[Baixar vídeo MP4]
+  P --> R[POST resolver-vod<br/>só no clique]
+  D --> R
+  R --> G{Câmara video-sob-demanda}
+  G -- PRONTO --> V[probe Range 206 video/mp4<br/>playback e download]
+  V -- verificado --> S[video nativo no trecho 0-based<br/>+ download na aba aberta]
+  V -- falha --> F[Estado honesto<br/>+ Tentar novamente]
+  G -- GERANDO --> E[Pendência visível<br/>+ Tentar novamente]
+  G -- INDISPONIVEL/erro --> F
+```
+
+**Opções consideradas (arquitetura):** A) helpers puros em `src/lib/speechVod.ts` (dono único) + resolver server-only na vertical + rota JSON; B) duplicar helpers em `src/` e deixar `scripts/` intocado; C) `src/` importar `scripts/lib/camaraSpeeches.mjs` em runtime.
+**Recomendação:** A — o app fica com o dono das regras de contrato (URL de trecho, parse do estado, offset BRT), os scripts re-exportam via tsx (precedente `camara:import`) e a resolução interativa nasce server-only com política própria de latência.
+**Rejeitadas:** B porque é o twin que a engineering-standards proíbe (drift no contrato `trecho`/`estado`); C porque inverte a fronteira (runtime de produção dependendo de dev tooling), arrasta `node:crypto` para o bundle e não é o dono natural das regras.
+
+### Decisões de engenharia
+
+**1. Onde vive a resolução VOD (extração dos helpers).**
+Opções: A) extrair as regras puras de VOD (`CAMARA_USER_AGENT`, `buildVodUrl`, `parseVodStatus`, `parseDurationToSeconds`) para `src/lib/speechVod.ts`; `scripts/lib/camaraSpeeches.mjs` passa a importar/re-exportar de lá (define só o que é de script) e `camara:pilot` ganha o mesmo loader tsx do `camara:import`; B) duplicar em `src/lib/speechVod.ts` e congelar os scripts; C) `src/utilities/speech/` importar `scripts/lib/camaraSpeeches.mjs` por caminho relativo.
+Recomendação: **A** — dono único, sem tocar no contrato dos scripts; o pilot é o único ponto de atrito e migra para `cross-env NODE_OPTIONS="--no-deprecation --import=tsx/esm" node scripts/pilot-camara-speeches.mjs` (o `camara:import` já roda assim; nenhum módulo do pilot importa `server-only`, então não precisa do seed-loader).
+Rejeitadas: B porque duplica o contrato mais sensível do item (qualquer mudança de `estado`/`trecho` passaria a ter dois donos); C porque o Docker copia `scripts/`, mas produção passaria a depender de tooling de dev, o alias `@/*` não cobre `scripts/` e `node:crypto` entraria no bundle do servidor.
+Nota: a política HTTP **não** é unificada — `probeLink`/`getJson` dos scripts têm contrato de relatório em lote (retry 3×/2s, nunca lança); o resolver do app tem contrato interativo (bounded, mapeia falha para estado honesto). Fork de política deliberado, com gatilho de revisitação: se um terceiro call-site interativo surgir, extrair `probeLink` para um módulo compartilhado.
+
+**2. Rota de resolução.**
+Opções: A) `POST` JSON em `src/app/(campaign)/campanha/(app)/comunicacao/acervo/resolver-vod/route.ts` via `campaignJsonMutationRoute`, corpo `{ speechId }` (segmento estático convive com `acervo/[id]`; o wrapper não entrega params de rota, então o id vai no corpo, como `municipios/political-trend`); B) server action + `useActionState` no player; C) rota sob `campanha/api/**`.
+Recomendação: **A** — a convenção do repo (teste `codebaseConventions` recusa `POST` fora do wrapper) e a proteção estrutural de mesma origem já existem; o `(app)/comunicacao` está no manifesto de e2e; a rota é fine-grained (`dynamic = 'force-dynamic'`), autentica por cookie de campanha e responde JSON com estados discriminados.
+Rejeitadas: B porque o ladder de actions devolve `CampaignFormErrorState` (shape de formulário) para um resultado rico (URLs verificadas), e a fronteira de CSRF do wrapper é o mecanismo canônico para mutação JSON sob `/campanha`; C porque fica fora do prefixo do manifesto e exigiria entrada de allowlist no guard de convenções.
+
+**3. Player default com ambos (4 quadrantes sem segundo player).**
+Opções: A) um único `SpeechDetailPlayer` client com máquina de estados e **uma** superfície por vez (`youtube` quando `youtubeVideoId`; `prompt`/`native` só quando não há YouTube; `unavailable` quando não há nenhum caminho); B) dois componentes irmãos (YouTube + VOD) com um orquestrador; C) manter `<video>` e iframe soltos no page/actions server.
+Recomendação: **A** — YouTube é default quando existe (decisão de produto sinalizada na intenção), o MP4 do "ambos" não troca a superfície (só o `Baixar` resolve), e o "só VOD" mostra capa → resolvendo → `<video>` nativo (0-based, como hoje). Baixar/Abrir fonte passam para dentro do player porque o pending e a URL verificada são acoplados ao estado dele (`sourceUrl` entra por prop); a página segue RSC.
+Rejeitadas: B porque multiplica arquivos e reintroduz o orquestrador que o player único elimina; C porque pending/seleção de URL são interativos (client boundary).
+
+**4. Seek na transcrição com YouTube.**
+Opções: A) IFrame API (`enablejsapi`, `postMessage`/`seekTo`); B) recarregar o iframe com `start = offset + segment.startSeconds` (degradação C da intenção); C) link `?t=` abrindo o YouTube; D) desabilitar o clique.
+Recomendação: **B** — determinística, sem script de terceiro, o contrato vive no `src` do iframe (asserção de e2e por HTTP), e o clique continua posicionando no orador certo; o reload interrompe a reprodução, degradação explicitamente aceitável pela intenção.
+Rejeitadas: A porque o plumbing de script/mensagens não cabe com folga no appetite e não tem como ser provado no e2e browserless atual; C porque perde o posicionamento na página (regressão do C154); D porque a intenção marca como regressão.
+Com offset (`youtubeOffsetSeconds`) nulo (dados ausentes), a transcrição no YouTube não vira affordance de clique (sem "clique para posicionar") — melhor inerte que pular para outro orador. Gatilho de revisitação: se a assessoria relatar o "pisca/volta ao início" do reload, promover para A em item próprio (o view model já carrega o offset; nenhum dado muda).
+
+**5. Persistência do link resolvido.**
+Opções: A) não persistir (estado só no client); B) persistir best-effort quando o ator tem `canUpdateSpeech` (coordinator/candidate), engolindo falha; C) persistir com `overrideAccess: true`.
+Recomendação: **A** — o link é cache declaradamente não confiável; a resolução por clique é o fluxo desenhado; nenhuma escrita nova, nenhum teste de write path, nenhum risco de ouroboros (link morto regravado).
+Rejeitadas: B porque amplia superfície de escrita e cria semântica "cache às vezes" sem ganho de produto medido; C porque amplia a escrita (proibido pela intenção).
+Gatilho de revisitação: latência p50 medida da Câmara tornar o clique repetidamente penoso → item próprio com `canUpdateSpeech` e invalidação explícita.
+
+**6. "Baixar" do card da lista.**
+Opções: A) remover `downloadUrl` do `SpeechListItemViewModel`/`SpeechResultCard` (e do `speechListSelect`), preservando `sourceUrl`; B) manter o card baixando o link cru; C) trocar o botão do card por link para o detalhe.
+Recomendação: **A** — a intenção manda sair (link efêmero direto); `sourceUrl` continua no card; o detalhe já é o dono do "Ver fala/Assistir no trecho"; menos select na lista.
+Rejeitadas: B porque entrega arquivo possivelmente morto; C porque o card já tem "Assistir no trecho" como CTA primário.
+
+**7. Como o estado `GERANDO` é tratado.**
+Opções: A) uma chamada `getJson` limitada (2 tentativas, timeout ~15s, backoff curto) que devolve `gerando` e a UI mostra pendência + "Tentar novamente" sem poll; B) poll curto no servidor (ex.: 3×5s) antes de responder; C) poll automático no client.
+Recomendação: **A** — o clique tem orçamento de latência; a regeneração observada vai de segundos a minutos, e o convite a tentar de novo é o desenho da intenção ("limite + convite, sem travar a página").
+Rejeitadas: B porque segura a request por 15s mesmo quando a Câmara está presa (o `resolveVod` de 200s dos scripts é inadequado aqui); C porque adiciona timers/rate e um retry escondido que a intenção não pede.
+
+**8. Verificação do MP4 (probe) e entrega do download.**
+Opções: A) `PRONTO` → probe em paralelo de `linkParaReproducao` e `linkParaDownload` (GET `Range: bytes=0-1023`, fallback HEAD em 405/501, ~10s, nunca lança; verificado = HTTP ok e content-type presente ≠ `text/html`); `<video>` só com playback verificado; "Baixar" só entrega download verificado; B) confiar no `PRONTO` sem probe; C) verificar só no download.
+Recomendação: **A** — "Baixar só entrega arquivo verificado" e o player nativo não aponta para hash morto (o `PRONTO` velho medido devolve 403/hash morto). Se playback falha mas download passa, não há `<video>` (estado honesto com retry) mas o download verificado continua ofertado. Na entrega, o clique em "Baixar" abre uma aba em branco **sincronamente** (evita popup blocker) e a aponta para `downloadUrl` só depois de verificado; falha → fecha a aba e mostra o estado honesto. No quadrante ambos o YouTube continua tocando: a falha vira aviso inline ao lado das ações, **não** troca a superfície.
+Rejeitadas: B porque viola o aceite (arquivo morto entregue); C porque deixaria o player nativo servir hash morto (draft cena 5 diz "verificado antes de tocar").
+Rejeitadas na entrega: `window.open(downloadUrl)` depois do await (popup blocker) e `location.assign` (perde a página da campanha/perde o retry).
+
+**9. Cache/estado no client.**
+Opções: A) resultado da resolução só em `useState` do player; reload volta ao default (YouTube no "ambos", capa no "só VOD"); B) `sessionStorage` com o link resolvido; C) revalidar/gravar no servidor.
+Recomendação: **A** — sem link morto cacheado, sem estado compartilhado entre abas/sessões; "Tentar novamente" limpa o erro e re-POSTa.
+Rejeitadas: B porque ressuscita link efêmero (o problema que o item resolve) e duplica a fonte da verdade; C porque colide com a decisão 5 (sem escrita) e com o render sem Câmara.
+
+### Componentes / mudanças
+
+- **`src/lib/speechVod.ts`** (novo, puro, client-safe, sem imports): `CAMARA_USER_AGENT`, `buildVodUrl`, `parseVodStatus`, `parseDurationToSeconds` (movidos, dono novo), `parseYoutubeVideoId` (watch/`youtu.be`/embed/live/shorts; id `[A-Za-z0-9_-]{6,20}`; domínios YouTube), `excerptOffsetSeconds(excerptTMs, eventStartAt)` (epoch → hora BRT via `Intl` `America/Sao_Paulo` com `hour12:false`, menos o naive parseado como frame UTC; null se não parseável ou negativo; piso inteiro).
+- **`scripts/lib/camaraSpeeches.mjs`**: remove as definições movidas e as re-exporta de `../../src/lib/speechVod.ts`; `scripts/lib/camaraFetch.mjs` segue importando de `camaraSpeeches.mjs` (sem mudança).
+- **`package.json`**: `camara:pilot` ganha `cross-env NODE_OPTIONS="--no-deprecation --import=tsx/esm"` (espelha `camara:import`).
+- **`src/lib/schemas/speechVod.ts`** (novo): `speechVodRequestSchema = { speechId: positiveRelationshipId }` + constantes de mensagem (`SPEECH_VOD_INELIGIBLE_MESSAGE`, `SPEECH_VOD_FORBIDDEN_MESSAGE`, `SPEECH_VOD_NOT_FOUND_MESSAGE`, `SPEECH_VOD_GENERIC_ERROR_MESSAGE`) — exigidas pelo guard de convenções (sem literais em `safeMessages`/`throw new Error` sob `(campaign)`).
+- **`src/utilities/speech/speechVodResolver.ts`** (novo, `server-only`): `resolveSpeechVod({ eventId, audioId, excerptTms })` — `buildVodUrl` + uma chamada `getJson`-like bounded (2×, timeout 15s, UA da Câmara) + probe A em paralelo; devolve `{ state: 'pronto', playbackUrl, downloadUrl }` (URLs nulas quando o probe falha), `{ state: 'gerando' }`, `{ state: 'indisponivel' }`; lança só em falha de transporte/parse (o route colapsa no genérico).
+- **`src/app/(campaign)/campanha/actions/speech.ts`** (novo, ladder): `resolveSpeechVodForActor({ speechId })` — `getCampaignActionContext`, fast-fail `canReadSpeechCatalog(actor.role)`, `payload.find` da fala com `overrideAccess: false` e `select: { eventId, audioId, excerptTMs, vodPlaybackUrl, vodDownloadUrl }`; sem os campos/`vod` armazenado lança `SPEECH_VOD_INELIGIBLE_MESSAGE`; chama o resolver.
+- **`src/app/(campaign)/campanha/(app)/comunicacao/acervo/resolver-vod/route.ts`** (novo) + **`types.ts`**: `export const dynamic = 'force-dynamic'`; `POST = campaignJsonMutationRoute({ bodySchema, safeMessages, genericMessage }, handler)`; resposta `{ status: 'success', state, playbackUrl?, downloadUrl? }`.
+- **`src/utilities/speech/speechViewModels.ts`**: lista perde `downloadUrl`; detalhe perde `vodDownloadUrl` e ganha `youtubeVideoId`, `youtubeOffsetSeconds`, `vodResolvable` (vod armazenado **e** `eventId`/`audioId`/`excerptTMs`), calculados com os helpers puros.
+- **`src/utilities/speech/speechPageData.ts`**: `speechListSelect` perde `vodDownloadUrl`; `speechDetailSelect` ganha `vodDownloadUrl`, `eventId`, `audioId`, `excerptTMs`, `eventStartAt` (nenhum campo novo no banco — todos já existem).
+- **`src/components/campaign/speech/SpeechDetailPlayer.tsx`**: reescrito (client) com props `{ speechId, youtubeVideoId, youtubeOffsetSeconds, vodResolvable, segments, initialSeconds, sourceUrl }` e estados `idle | resolving | generating | ready | unavailable`; ações dentro do player.
+- **`src/app/(campaign)/campanha/(app)/comunicacao/acervo/[id]/page.tsx`**: passa as props novas e remove o bloco `<a>` de MP4 (o "Abrir fonte" passa ao player); asides/transcrição oficial intocados.
+- **`src/components/campaign/speech/SpeechResultCard.tsx`**: remove o `ActionLink` de download (e o import `DownloadIcon` morto).
+- **Migration:** nenhuma — sem schema change, sem `payload-types` novo, `push: false` intocado.
+- **Access / Consent:** nenhum `Consent`; leitura da fala com `overrideAccess: false` + gate `canReadSpeechCatalog` na action; nenhuma escrita; sem S3.
+- **UI:** Impeccable B — reusar `Button`/`Badge`, `aria-busy` + live region polite no pending (campanha-action-feedback), copy do draft; shape→craft→critique→polish no detalhe.
+
+### Dados → forma
+
+**Não — não se aplica (feedback de ação).** Nenhum KPI/gráfico novo: "resolvendo/gerando/indisponível" é o estado da ação no player (draft cenas 3–4), não dado de acervo; a única forma nova é a superfície do player, já coberta pela direção de UI acima.
+
+## Fases verificáveis
+
+1. **Fase 1 — Tracer puro + extração (dono único).** `src/lib/speechVod.ts` + re-exports em `scripts/lib/camaraSpeeches.mjs` + loader tsx do pilot.
+   - Testes: `tests/unit/speechVod.unit.spec.ts` (node) — `buildVodUrl`/`parseVodStatus`/`parseDurationToSeconds` preservados, `excerptOffsetSeconds` com o literal da intenção (`1786473834650` + 15:00 → 2634), o fixture do import (`1675801808560` + `2023-02-07T15:00` → 9008), caso DST antigo, negativo/não-parseável → null; `parseYoutubeVideoId` (watch/`youtu.be`/embed/lixo). `tests/unit/camaraSpeeches.unit.spec.ts` segue verde **pela re-exportação** (prova que a extração preservou o contrato).
+   - Verificação: `pnpm test:unit`, `pnpm typecheck`, `node --import=tsx/esm scripts/pilot-camara-speeches.mjs --help`.
+2. **Fase 2 — Server (view fields + resolver + rota).** Selects/view model, resolver server-only, schema/mensagens, action e rota.
+   - Testes: `tests/unit/speechVodResolver.unit.spec.ts` (fetch mockado: PRONTO+206 → URLs verificadas; PRONTO+403 → sem URL; GERANDO; INDISPONIVEL; timeout lança); `tests/int/speechAcervo.int.spec.ts` estendido (detalhe expõe `youtubeVideoId`, `youtubeOffsetSeconds` calculado, `vodResolvable` true/false; lista sem `downloadUrl`).
+   - Verificação: `pnpm test:unit`, `pnpm test:int`, `pnpm typecheck`, `pnpm build` (convivência `acervo/resolver-vod` × `acervo/[id]`).
+3. **Fase 3 — Player.** Máquina de estados, iframe com `start=` e capa só VOD, `<video>` nativo pós-PRONTO, aviso inline de falha no quadrante ambos.
+   - Testes: `tests/unit/speechDetailPlayer.unit.spec.tsx` (jsdom + testing-library, fetch mockado) — só VOD: capa → clique → pending/`aria-busy` → `<video src>` no trecho; falha → bloco honesto + "Tentar novamente" re-POSTa; ambos: iframe com `start=offset+t`, Baixar pendente/desabilitado e depois entrega, transcrição atualiza o `src`; sem YouTube sem VOD → bloco sem retry.
+   - Verificação: `pnpm test:unit`.
+4. **Fase 4 — Card/lista.** Remoção do download do card/view model e do select da lista.
+   - Testes: unit/componente do card + int da lista.
+   - Verificação: `pnpm test:unit`, `pnpm test:int`, `pnpm exec knip` (import órfão do `DownloadIcon` morto).
+5. **Fase 5 — E2E renegociado + gates.** `tests/e2e/campaignSpeechAcervo.e2e.spec.ts`:
+   - fixtures novas com `youtubeUrl` + `eventStartAt`/`eventId`/`audioId`/`excerptTMs` (ambos e só VOD) e sem vídeo;
+   - ambos: `youtube-nocookie.com/embed/<id>` com `start=<offset>` e "Baixar vídeo (MP4)" e `data-start-seconds`, sem `<video>` no load;
+   - só VOD: capa "Assistir o trecho", sem iframe/`<video>` no load, "Baixar vídeo (MP4)" presente (e não contém o link `vod.camara.leg.br` na lista do card);
+   - sem vídeo: "Vídeo indisponível neste momento" + "Abrir fonte", sem "Tentar novamente"; falha pós-clique coberta no unit da fase 3;
+   - rota: POST sem sessão → 401; advisor → mensagem de acesso (400); origem estranha → 403; fala inelegível → 400 (sem rede).
+   - Verificação: `pnpm gate:fast`, `pnpm test:e2e:affected`, `pnpm push`; changelog `docs/changelog/2026-09-14-c162-vod-sob-demanda.md`.
+
+## Rabbit holes / Não escopo (engenharia)
+
+- IFrame API do YouTube (seek sem reload) — degradação C aceita; gatilho na decisão 4.
+- Persistir/aquecer o link resolvido, re-verificação em background/cron, "gerar novamente" manual, regenerar os 924 em lote.
+- Espelhar/transcodificar MP4 (S3/Garage), download em lote, editor/corte.
+- Poll de `GERANDO` (servidor ou client) e qualquer chamada à Câmara no render/RSC.
+- Segundo player, redesign da lista/detalhe além da saída do "Baixar" do card, "Abrir no YouTube" no lugar do MP4.
+- Migration/collection/`Consent` novos; unificar a política de `probeLink` dos scripts com a do app (fork deliberado, decisão 1).
+
+## Riscos e mitigação
+
+- **`youtubeUrl` fora do formato esperado** (`urlRegistro` varia) → `parseYoutubeVideoId` defensivo com unit tests; null cai no caminho só VOD/indisponível (nunca iframe quebrado).
+- **Latência/anti-bot da Câmara** (400 "Acesso via bot", timeouts) → UA compartilhado, 2 tentativas com timeout curto, estado honesto com "Tentar novamente"; nenhuma página depende da Câmara para renderizar.
+- **Migração do pilot para tsx** (script dormente) → smoke `--help` + unit tests da re-exportação; rollback é uma linha de `package.json`.
+- **E2E browserless não clica** → interação (resolvendo → `<video>`, falha, retry, seek) provada no unit jsdom; e2e prova contrato HTTP, gates e o HTML inicial.
+- **Colisão de rota** `acervo/resolver-vod` × `acervo/[id]` → ids são numéricos e `pnpm build` valida; e2e cobre o POST.
+- **Manifesto de e2e**: `(app)/comunicacao`, `src/components/campaign/speech`, `src/utilities/speech` e `src/lib/speech*` já mapeiam para `campaignSpeechAcervo`; `src/lib/schemas/speechVod.ts` cai no prefixo `src/lib/speech`.
+- **Knip `exports:error`**: re-exports dos scripts cobertos pelas entradas `scripts/*.mjs`; símbolos novos usados por action/route/player; rodar `pnpm exec knip` na fase 4.
+
+## Aceite de engenharia
+
+- [ ] Aceite de produto da intenção ainda coberto (quadrantes, `excerptTMs` verbatim, Baixar verificado, estado honesto, crédito CC BY, gate do acervo).
+- [ ] Invariantes AGENTS/engineering-standards (dono único, sem twin, sem migration/collection/Consent, sem escrita nova, `overrideAccess: false`, client boundary, zero warnings, dead code).
+- [ ] Testes de domínio previstos (unit/resolver/player/int/e2e) onde o access/write path e os contratos de URL mudam.
+
+## Self-score (decision-quality, gate ≥4)
+
+5/5 — (1) todas as decisões caras (extração, rota, player único, persistência, verificação) têm opções + rejeitadas explícitas; (2) cabe no appetite herdado porque corta IFrame API, persistência e poll; (3) rabbit holes nomeados com corte; (4) depth check: reusa `campaignJsonMutationRoute`, ladder de actions, `getCampaignActionContext`, `speechPageData`/`speechViewModels` e o padrão de feedback de ação — nada de shell/abstração paralela; (5) o outcome da intenção permanece intacto, com a degradação 4-C explicitamente permitida e sinalizada ao gate humano.
diff --git a/package.json b/package.json
index d807ef80..b45b75ef 100644
--- a/package.json
+++ b/package.json
@@ -81,7 +81,7 @@
     "test:unit": "cross-env NODE_OPTIONS=--no-deprecation DATABASE_URL=postgresql://invalid:invalid@127.0.0.1:1/unit_tests_must_not_connect vitest run --config ./vitest.unit.config.mts",
     "images:resize": "node scripts/resize-images.mjs",
     "images:fill": "node scripts/fill-image.mjs",
-    "camara:pilot": "node scripts/pilot-camara-speeches.mjs",
+    "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",
     "typecheck": "tsc --noEmit",
     "typecheck:tsc6": "tsc6 --noEmit",
diff --git a/scripts/lib/camaraSpeeches.mjs b/scripts/lib/camaraSpeeches.mjs
index f22b530e..b5b591c9 100644
--- a/scripts/lib/camaraSpeeches.mjs
+++ b/scripts/lib/camaraSpeeches.mjs
@@ -11,13 +11,18 @@
  */
 import { createHash } from 'node:crypto'
 
+// The VOD/YouTube contract moved to its production owner (C162) and is
+// re-exported here so the C152/C153 scripts keep the same import surface.
+export {
+  buildVodUrl,
+  CAMARA_USER_AGENT,
+  parseDurationToSeconds,
+  parseVodStatus,
+} from '../../src/lib/speechVod.ts'
+
 /** Câmara's open-data deputy id for Jorge Solla (used by the pilot). */
 export const SOLLA_DEPUTY_ID = 178857
 
-/** Browser-like UA — the trecho/evento pages answer 400 "Acesso via bot" to curl. */
-export const CAMARA_USER_AGENT =
-  'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/125.0.0.0 Safari/537.36'
-
 /** Deep Infra OpenAI-compatible transcription endpoint (returns segments with `verbose_json`). */
 export const DEEPINFRA_TRANSCRIBE_URL = 'https://api.deepinfra.com/v1/openai/audio/transcriptions'
 
@@ -139,40 +144,6 @@ export const legislatureForDate = (iso) => {
   return null
 }
 
-/**
- * Parses the Câmara duration spellings into seconds: `0h04'03"` (excerpt card)
- * and `0:04:07` / `4:07` (VOD status). Null when unparseable.
- *
- * @param {unknown} value
- * @returns {number | null}
- */
-export const parseDurationToSeconds = (value) => {
-  const raw = String(value ?? '').trim()
-  if (raw === '') return null
-
-  const card = /^(\d+)h(\d{1,2})'(\d{1,2})"?$/.exec(raw)
-  if (card) {
-    const minutes = Number(card[2])
-    const seconds = Number(card[3])
-    if (minutes > 59 || seconds > 59) return null
-    return Number(card[1]) * 3600 + minutes * 60 + seconds
-  }
-
-  const parts = raw.split(':')
-  if (!parts.every((part) => /^\d{1,2}$/.test(part))) return null
-  if (parts.length === 3) {
-    const [hours, minutes, seconds] = parts.map(Number)
-    if (minutes > 59 || seconds > 59) return null
-    return hours * 3600 + minutes * 60 + seconds
-  }
-  if (parts.length === 2) {
-    const [minutes, seconds] = parts.map(Number)
-    if (seconds > 59) return null
-    return minutes * 60 + seconds
-  }
-  return null
-}
-
 const integerOrNull = (value) => {
   const n = Number(value)
   return Number.isInteger(n) && n > 0 ? n : null
@@ -222,38 +193,6 @@ export const parseEventExcerpts = (html, eventId) => {
   return excerpts
 }
 
-/** The async VOD-generation status URL for one excerpt. */
-export const buildVodUrl = (eventId, audioId, tMs) =>
-  `https://www.camara.leg.br/evento-legislativo/${eventId}/video-sob-demanda?idAudio=${audioId}&trecho=${tMs}`
-
-/**
- * Normalizes the `video-sob-demanda` JSON (`{ estado, video }`). `state` is
- * `GERANDO` while the server transcodes, `PRONTO` when the MP4 exists,
- * `INDISPONIVEL` when there is none.
- *
- * @param {unknown} json
- * @returns {{ state: string, video: { title: string | null, subtitle: string | null, duration: string | null, clock: string | null, downloadUrl: string | null, playbackUrl: string | null } | null }}
- */
-export const parseVodStatus = (json) => {
-  const data = json && typeof json === 'object' ? json : {}
-  const estado = typeof data.estado === 'string' ? data.estado : 'DESCONHECIDO'
-  const raw = data.video && typeof data.video === 'object' ? data.video : null
-  const pick = (key) => (typeof raw?.[key] === 'string' ? raw[key] : null)
-  return {
-    state: estado,
-    video: raw
-      ? {
-          title: pick('titulo'),
-          subtitle: pick('subtitulo'),
-          duration: pick('duracao'),
-          clock: pick('horario'),
-          downloadUrl: pick('linkParaDownload'),
-          playbackUrl: pick('linkParaReproducao'),
-        }
-      : null,
-  }
-}
-
 /**
  * Events of the speech's date, nearest-first by start time. The open-data
  * `uriEvento` comes empty, so the event is inferred from the date/time.
diff --git a/src/app/(campaign)/campanha/(app)/comunicacao/acervo/[id]/page.tsx b/src/app/(campaign)/campanha/(app)/comunicacao/acervo/[id]/page.tsx
index f7a813f3..050d27ea 100644
--- a/src/app/(campaign)/campanha/(app)/comunicacao/acervo/[id]/page.tsx
+++ b/src/app/(campaign)/campanha/(app)/comunicacao/acervo/[id]/page.tsx
@@ -1,5 +1,5 @@
 import config from '@payload-config'
-import { ArrowLeftIcon, DownloadIcon, ExternalLinkIcon } from 'lucide-react'
+import { ArrowLeftIcon } from 'lucide-react'
 import Link from 'next/link'
 import { notFound } from 'next/navigation'
 import { getPayload } from 'payload'
@@ -103,31 +103,16 @@ export default async function SpeechDetailPage({ params, searchParams }: SpeechD
 
             <div className="mt-4">
               <SpeechDetailPlayer
-                vodPlaybackUrl={view.vodPlaybackUrl}
+                speechId={view.id}
+                youtubeVideoId={view.youtubeVideoId}
+                youtubeOffsetSeconds={view.youtubeOffsetSeconds}
+                vodResolvable={view.vodResolvable}
                 segments={view.segments}
                 initialSeconds={initialSeconds}
+                sourceUrl={sourceUrl}
               />
             </div>
 
-            <div className="mt-3 flex flex-wrap items-center gap-2">
-              {view.vodDownloadUrl ? (
-                <Button asChild className="min-h-10">
-                  <a href={view.vodDownloadUrl} target="_blank" rel="noreferrer">
-                    <DownloadIcon data-icon="inline-start" aria-hidden="true" />
-                    Baixar vídeo (MP4)
-                  </a>
-                </Button>
-              ) : null}
-              {sourceUrl ? (
-                <Button asChild variant="outline" className="min-h-10">
-                  <a href={sourceUrl} target="_blank" rel="noreferrer">
-                    <ExternalLinkIcon data-icon="inline-start" aria-hidden="true" />
-                    Abrir fonte
-                  </a>
-                </Button>
-              ) : null}
-            </div>
-
             {view.officialTranscript ? (
               <details className="mt-5 rounded-lg border p-3">
                 <summary className="cursor-pointer text-sm font-medium">
diff --git a/src/app/(campaign)/campanha/(app)/comunicacao/acervo/resolver-vod/route.ts b/src/app/(campaign)/campanha/(app)/comunicacao/acervo/resolver-vod/route.ts
new file mode 100644
index 00000000..cdbe9f11
--- /dev/null
+++ b/src/app/(campaign)/campanha/(app)/comunicacao/acervo/resolver-vod/route.ts
@@ -0,0 +1,34 @@
+import { NextResponse } from 'next/server'
+
+import { resolveSpeechVodForActor } from '@/app/(campaign)/campanha/actions/speech'
+import {
+  SPEECH_VOD_FORBIDDEN_MESSAGE,
+  SPEECH_VOD_GENERIC_ERROR_MESSAGE,
+  SPEECH_VOD_INELIGIBLE_MESSAGE,
+  SPEECH_VOD_NOT_FOUND_MESSAGE,
+  speechVodRequestSchema,
+} from '@/lib/schemas/speechVod'
+import { campaignJsonMutationRoute } from '@/utilities/campaignJsonMutationRoute'
+
+import type { SpeechVodResolveResponse } from './types'
+
+export type { SpeechVodResolveResponse } from './types'
+
+export const dynamic = 'force-dynamic'
+
+export const POST = campaignJsonMutationRoute(
+  {
+    bodySchema: speechVodRequestSchema,
+    safeMessages: [
+      SPEECH_VOD_FORBIDDEN_MESSAGE,
+      SPEECH_VOD_INELIGIBLE_MESSAGE,
+      SPEECH_VOD_NOT_FOUND_MESSAGE,
+    ],
+    genericMessage: SPEECH_VOD_GENERIC_ERROR_MESSAGE,
+  },
+  async ({ speechId }) => {
+    const resolution = await resolveSpeechVodForActor({ speechId })
+
+    return NextResponse.json<SpeechVodResolveResponse>({ status: 'success', resolution })
+  },
+)
diff --git a/src/app/(campaign)/campanha/(app)/comunicacao/acervo/resolver-vod/types.ts b/src/app/(campaign)/campanha/(app)/comunicacao/acervo/resolver-vod/types.ts
new file mode 100644
index 00000000..f9e5974f
--- /dev/null
+++ b/src/app/(campaign)/campanha/(app)/comunicacao/acervo/resolver-vod/types.ts
@@ -0,0 +1,6 @@
+import type { SpeechVodResolution } from '@/lib/speechVod'
+
+/** Wire contract of `POST /campanha/comunicacao/acervo/resolver-vod` (C162). */
+export type SpeechVodResolveResponse =
+  | { status: 'success'; resolution: SpeechVodResolution }
+  | { status: 'error'; message: string }
diff --git a/src/app/(campaign)/campanha/actions/speech.ts b/src/app/(campaign)/campanha/actions/speech.ts
new file mode 100644
index 00000000..6e04c29e
--- /dev/null
+++ b/src/app/(campaign)/campanha/actions/speech.ts
@@ -0,0 +1,56 @@
+'use server'
+
+import { canReadSpeechCatalog } from '@/lib/campaignRoles'
+import {
+  SPEECH_VOD_FORBIDDEN_MESSAGE,
+  SPEECH_VOD_INELIGIBLE_MESSAGE,
+  SPEECH_VOD_NOT_FOUND_MESSAGE,
+  speechVodRequestSchema,
+} from '@/lib/schemas/speechVod'
+import type { SpeechVodResolution } from '@/lib/speechVod'
+import { getCampaignActionContext } from '@/utilities/campaignActionContext'
+import { resolveSpeechVod } from '@/utilities/speech/speechVodResolver'
+
+/**
+ * C162 — resolves the exact excerpt MP4 for one speech through the Câmara
+ * `video-sob-demanda` endpoint. The catalog gate is checked BEFORE the read so
+ * a denied actor gets the domain message (the read itself runs with
+ * `overrideAccess: false` and would only yield "not found"). The stored VOD
+ * link is an eligibility signal, never a URL handed to the client: the
+ * resolution owns the truth, and the response only carries probed links.
+ */
+export const resolveSpeechVodForActor = async (input: {
+  speechId: number
+}): Promise<SpeechVodResolution> => {
+  const { speechId } = speechVodRequestSchema.parse(input)
+  const { payload, actor } = await getCampaignActionContext()
+
+  if (!canReadSpeechCatalog(actor.role)) throw new Error(SPEECH_VOD_FORBIDDEN_MESSAGE)
+
+  const result = await payload.find({
+    collection: 'speech',
+    where: { id: { equals: speechId } },
+    depth: 0,
+    limit: 1,
+    pagination: false,
+    select: {
+      eventId: true,
+      audioId: true,
+      excerptTMs: true,
+      vodPlaybackUrl: true,
+      vodDownloadUrl: true,
+    },
+    user: actor,
+    overrideAccess: false,
+  })
+  const speech = result.docs[0]
+  if (!speech) throw new Error(SPEECH_VOD_NOT_FOUND_MESSAGE)
+
+  const hasStoredVod = Boolean(speech.vodPlaybackUrl || speech.vodDownloadUrl)
+  const { eventId, audioId, excerptTMs } = speech
+  if (!hasStoredVod || !eventId || !audioId || !excerptTMs) {
+    throw new Error(SPEECH_VOD_INELIGIBLE_MESSAGE)
+  }
+
+  return resolveSpeechVod({ eventId, audioId, excerptTms: excerptTMs })
+}
diff --git a/src/components/campaign/speech/SpeechDetailPlayer.tsx b/src/components/campaign/speech/SpeechDetailPlayer.tsx
index a651ac46..9b226a58 100644
--- a/src/components/campaign/speech/SpeechDetailPlayer.tsx
+++ b/src/components/campaign/speech/SpeechDetailPlayer.tsx
@@ -1,29 +1,108 @@
 'use client'
 
-import { FilmIcon } from 'lucide-react'
-import { useEffect, useRef, useState } from 'react'
+import { DownloadIcon, ExternalLinkIcon, FilmIcon, PlayIcon } from 'lucide-react'
+import { useEffect, useMemo, useRef, useState, type ReactNode } from 'react'
 
+import type { SpeechVodResolveResponse } from '@/app/(campaign)/campanha/(app)/comunicacao/acervo/resolver-vod/types'
 import { SpeechHighlightParts } from '@/components/campaign/speech/SpeechHighlightParts'
+import { Button } from '@/components/ui/button'
+import { Spinner } from '@/components/ui/Spinner'
+import { postCampaignJson } from '@/lib/campaignJsonRequest'
+import type { SpeechVodResolution } from '@/lib/speechVod'
 import { cn } from '@/lib/utils'
 import type { SpeechDetailSegmentViewModel } from '@/utilities/speech/speechViewModels'
 
-/**
- * C154 — native `<video>` plus the clickable ASR transcript. The VOD is the
- * Câmara excerpt MP4 and the segment `startSeconds` are relative to it, so a
- * click is a plain `currentTime` seek. No autoplay (browsers block it): the
- * user presses play and the video starts at the requested fragment.
- */
-export const SpeechDetailPlayer = ({
-  vodPlaybackUrl,
-  segments,
-  initialSeconds,
-}: {
-  vodPlaybackUrl: string | null
+const RESOLVE_ENDPOINT = '/campanha/comunicacao/acervo/resolver-vod'
+
+type ResolutionState =
+  | { kind: 'idle' }
+  | { kind: 'resolving' }
+  | { kind: 'generating' }
+  | { kind: 'resolved'; playbackUrl: string | null; downloadUrl: string | null }
+  | { kind: 'failed'; message: string | null }
+
+const toResolutionState = (resolution: SpeechVodResolution): ResolutionState =>
+  resolution.state === 'pronto'
+    ? {
+        kind: 'resolved',
+        playbackUrl: resolution.playbackUrl,
+        downloadUrl: resolution.downloadUrl,
+      }
+    : resolution.state === 'gerando'
+      ? { kind: 'generating' }
+      : { kind: 'failed', message: null }
+
+type SpeechDetailPlayerProps = {
+  speechId: number
+  /** C162 — YouTube default when the session link parses; null falls back to the VOD. */
+  youtubeVideoId: string | null
+  /** Seconds from the session start to the excerpt; null makes the YouTube seek inert. */
+  youtubeOffsetSeconds: number | null
+  /** Stored VOD + excerpt coordinates: the Câmara may be asked on click. */
+  vodResolvable: boolean
   segments: readonly SpeechDetailSegmentViewModel[]
   initialSeconds: number | null
-}) => {
+  sourceUrl: string | null
+}
+
+const StatusPanel = ({
+  title,
+  detail,
+  busy = false,
+  children,
+}: {
+  title: string
+  detail: string
+  busy?: boolean
+  children?: ReactNode
+}) => (
+  <div
+    role="status"
+    aria-live="polite"
+    className="flex aspect-video w-full flex-col items-center justify-center gap-2 rounded-lg border bg-muted/40 px-6 text-center text-muted-foreground"
+  >
+    {busy ? (
+      <Spinner
+        className="size-6 text-muted-foreground"
+        aria-label="Resolvendo o trecho na Câmara"
+      />
+    ) : (
+      <FilmIcon className="size-8" aria-hidden="true" />
+    )}
+    <p className="text-sm font-medium text-foreground/90">{title}</p>
+    <p className="max-w-md text-xs">{detail}</p>
+    {children}
+  </div>
+)
+
+/**
+ * C162 — one player for the four acervo quadrants: YouTube default when the
+ * session link exists (no Câmara call on render), the stored VOD otherwise
+ * resolved on click (`video-sob-demanda`, exact `excerptTMs`, links probed
+ * before use), and honest states when neither path has a playable file.
+ * Transcript clicks seek the current surface: native seconds on the MP4,
+ * session-offset seconds on the YouTube embed.
+ */
+export const SpeechDetailPlayer = ({
+  speechId,
+  youtubeVideoId,
+  youtubeOffsetSeconds,
+  vodResolvable,
+  segments,
+  initialSeconds,
+  sourceUrl,
+}: SpeechDetailPlayerProps) => {
   const videoRef = useRef<HTMLVideoElement>(null)
+  const [resolution, setResolution] = useState<ResolutionState>({ kind: 'idle' })
   const [activeStart, setActiveStart] = useState<number | null>(null)
+  const [youtubeStart, setYoutubeStart] = useState<number | null>(() =>
+    youtubeVideoId && youtubeOffsetSeconds !== null && initialSeconds !== null
+      ? youtubeOffsetSeconds + initialSeconds
+      : null,
+  )
+
+  const resolving = resolution.kind === 'resolving'
+  const playbackUrl = resolution.kind === 'resolved' ? resolution.playbackUrl : null
 
   useEffect(() => {
     const video = videoRef.current
@@ -38,9 +117,50 @@ export const SpeechDetailPlayer = ({
     }
     video.addEventListener('loadedmetadata', seek, { once: true })
     return () => video.removeEventListener('loadedmetadata', seek)
-  }, [initialSeconds, vodPlaybackUrl])
+  }, [initialSeconds, playbackUrl])
+
+  const requestResolution = async (deliverDownload: boolean) => {
+    // Opened synchronously inside the click gesture so the popup blocker sees
+    // a user-initiated tab; pointed at the verified URL only after the probe.
+    const pendingTab = deliverDownload ? window.open('', '_blank') : null
+    setResolution({ kind: 'resolving' })
+
+    try {
+      const { ok, payload } = await postCampaignJson<SpeechVodResolveResponse>(RESOLVE_ENDPOINT, {
+        speechId,
+      })
+      if (!ok || payload.status !== 'success') {
+        pendingTab?.close()
+        setResolution({
+          kind: 'failed',
+          message: payload.status === 'error' ? payload.message : null,
+        })
+        return
+      }
+
+      const next = toResolutionState(payload.resolution)
+      setResolution(next)
+      if (!deliverDownload) return
+
+      if (next.kind === 'resolved' && next.downloadUrl) {
+        if (pendingTab) pendingTab.location.href = next.downloadUrl
+        else window.open(next.downloadUrl, '_blank')
+        return
+      }
+      pendingTab?.close()
+    } catch {
+      pendingTab?.close()
+      setResolution({ kind: 'failed', message: null })
+    }
+  }
 
   const seekTo = (seconds: number) => {
+    if (youtubeVideoId) {
+      if (youtubeOffsetSeconds === null) return
+      setYoutubeStart(youtubeOffsetSeconds + seconds)
+      setActiveStart(seconds)
+      return
+    }
     const video = videoRef.current
     if (!video) return
     video.currentTime = seconds
@@ -57,30 +177,150 @@ export const SpeechDetailPlayer = ({
     setActiveStart(active?.startSeconds ?? null)
   }
 
+  const youtubeSrc = useMemo(() => {
+    const params = new URLSearchParams({ playsinline: '1', rel: '0' })
+    if (youtubeStart !== null && youtubeStart > 0)
+      params.set('start', String(Math.floor(youtubeStart)))
+    return `https://www.youtube-nocookie.com/embed/${youtubeVideoId}?${params.toString()}`
+  }, [youtubeVideoId, youtubeStart])
+
+  const seekable = youtubeVideoId ? youtubeOffsetSeconds !== null : Boolean(playbackUrl)
+
+  const downloadAvailable = vodResolvable
+
+  const media = youtubeVideoId ? (
+    <iframe
+      src={youtubeSrc}
+      title="Vídeo da sessão no YouTube"
+      loading="lazy"
+      allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share; compute-pressure"
+      allowFullScreen
+      className="aspect-video w-full rounded-lg border bg-black"
+    />
+  ) : playbackUrl ? (
+    <video
+      ref={videoRef}
+      controls
+      preload="metadata"
+      src={playbackUrl}
+      onTimeUpdate={onTimeUpdate}
+      className="aspect-video w-full rounded-lg border bg-black"
+    >
+      <track kind="captions" />
+    </video>
+  ) : resolution.kind === 'resolving' ? (
+    <StatusPanel
+      busy
+      title="Resolvendo o trecho na Câmara…"
+      detail="Isso pode levar alguns segundos. Se demorar, tente novamente em instantes — a página não trava."
+    />
+  ) : resolution.kind === 'generating' ? (
+    <StatusPanel
+      title="A Câmara está gerando o trecho deste vídeo."
+      detail="A geração pode levar alguns instantes. Tente novamente em um momento."
+    >
+      <Button
+        variant="outline"
+        className="mt-1 min-h-10"
+        onClick={() => void requestResolution(false)}
+      >
+        Tentar novamente
+      </Button>
+    </StatusPanel>
+  ) : resolution.kind === 'failed' ||
+    (resolution.kind === 'resolved' && !resolution.playbackUrl) ? (
+    <StatusPanel
+      title="Não foi possível carregar o vídeo deste trecho."
+      detail={
+        resolution.kind === 'failed' && resolution.message
+          ? resolution.message
+          : 'A Câmara não entregou o arquivo agora. A transcrição e a fonte oficial continuam disponíveis.'
+      }
+    >
+      <Button
+        variant="outline"
+        className="mt-1 min-h-10"
+        onClick={() => void requestResolution(false)}
+      >
+        Tentar novamente
+      </Button>
+    </StatusPanel>
+  ) : vodResolvable ? (
+    <StatusPanel
+      title="O trecho deste vídeo é gerado pela Câmara dos Deputados."
+      detail="Clique para resolver o arquivo exato desta fala — ele é verificado antes de tocar."
+    >
+      <Button className="mt-1 min-h-10" onClick={() => void requestResolution(false)}>
+        <PlayIcon data-icon="inline-start" aria-hidden="true" />
+        Assistir o trecho
+      </Button>
+    </StatusPanel>
+  ) : (
+    <StatusPanel
+      title="Vídeo indisponível neste momento."
+      detail="Esta fala não tem vídeo no YouTube nem trecho gerado pela Câmara disponível agora."
+    />
+  )
+
+  const inlineNotice = youtubeVideoId ? (
+    resolution.kind === 'generating' ? (
+      <p className="text-xs text-muted-foreground" role="status" aria-live="polite">
+        A Câmara está gerando o trecho deste vídeo. Tente novamente em um momento.
+      </p>
+    ) : resolution.kind === 'failed' ? (
+      <p className="text-xs text-destructive" role="status" aria-live="polite">
+        {resolution.message ??
+          'Não foi possível resolver o arquivo deste trecho na Câmara. Tente novamente.'}
+      </p>
+    ) : resolution.kind === 'resolved' && !resolution.downloadUrl ? (
+      <p className="text-xs text-muted-foreground" role="status" aria-live="polite">
+        O arquivo deste trecho não pôde ser verificado na Câmara agora.
+      </p>
+    ) : null
+  ) : null
+
   return (
-    <div data-slot="speech-player">
-      {vodPlaybackUrl ? (
-        <video
-          ref={videoRef}
-          controls
-          preload="metadata"
-          src={vodPlaybackUrl}
-          onTimeUpdate={onTimeUpdate}
-          className="aspect-video w-full rounded-lg border bg-black"
-        >
-          <track kind="captions" />
-        </video>
-      ) : (
-        <div className="flex aspect-video w-full flex-col items-center justify-center gap-2 rounded-lg border bg-muted/40 text-muted-foreground">
-          <FilmIcon className="size-8" aria-hidden="true" />
-          <p className="text-sm">VOD indisponível para esta fala.</p>
-        </div>
-      )}
+    <div data-slot="speech-player" aria-busy={resolving || undefined}>
+      {media}
+
+      <div className="mt-3 flex flex-wrap items-center gap-2">
+        {downloadAvailable ? (
+          <Button
+            className="min-h-10"
+            disabled={resolving}
+            aria-busy={resolving || undefined}
+            onClick={() => {
+              if (resolution.kind === 'resolved' && resolution.downloadUrl) {
+                window.open(resolution.downloadUrl, '_blank')
+                return
+              }
+              void requestResolution(true)
+            }}
+          >
+            {resolving ? (
+              <Spinner data-icon="inline-start" aria-label="Resolvendo o trecho" />
+            ) : (
+              <DownloadIcon data-icon="inline-start" aria-hidden="true" />
+            )}
+            Baixar vídeo (MP4)
+          </Button>
+        ) : null}
+        {sourceUrl ? (
+          <Button asChild variant="outline" className="min-h-10">
+            <a href={sourceUrl} target="_blank" rel="noreferrer">
+              <ExternalLinkIcon data-icon="inline-start" aria-hidden="true" />
+              Abrir fonte
+            </a>
+          </Button>
+        ) : null}
+      </div>
+
+      {inlineNotice}
 
       {segments.length > 0 ? (
         <section className="mt-5" aria-label="Transcrição">
           <p className="text-xs tracking-wide text-muted-foreground uppercase">
-            Transcrição (clique para posicionar)
+            {seekable ? 'Transcrição (clique para posicionar)' : 'Transcrição'}
           </p>
           <ol className="mt-2 space-y-0.5">
             {segments.map((segment) => {
@@ -91,8 +331,11 @@ export const SpeechDetailPlayer = ({
                     type="button"
                     data-start-seconds={segment.startSeconds}
                     onClick={() => seekTo(segment.startSeconds)}
+                    disabled={!seekable}
                     className={cn(
-                      'flex w-full gap-3 rounded-lg px-2 py-1.5 text-left transition-colors hover:bg-muted focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring',
+                      'flex w-full gap-3 rounded-lg px-2 py-1.5 text-left transition-colors',
+                      seekable && 'hover:bg-muted',
+                      'focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring',
                       active && 'bg-muted',
                     )}
                   >
diff --git a/src/components/campaign/speech/SpeechResultCard.tsx b/src/components/campaign/speech/SpeechResultCard.tsx
index 078fd94a..12ac6661 100644
--- a/src/components/campaign/speech/SpeechResultCard.tsx
+++ b/src/components/campaign/speech/SpeechResultCard.tsx
@@ -1,4 +1,4 @@
-import { DownloadIcon, ExternalLinkIcon, PlayIcon } from 'lucide-react'
+import { ExternalLinkIcon, PlayIcon } from 'lucide-react'
 import Link from 'next/link'
 import type { ReactNode } from 'react'
 
@@ -116,12 +116,6 @@ export const SpeechResultCard = ({ speech }: { speech: SpeechListItemViewModel }
             {speech.matchKind === 'segment' ? 'Assistir no trecho' : 'Ver fala'}
           </Link>
         </Button>
-        {speech.downloadUrl ? (
-          <ActionLink href={speech.downloadUrl} variant="outline">
-            <DownloadIcon data-icon="inline-start" aria-hidden="true" />
-            Baixar
-          </ActionLink>
-        ) : null}
         {speech.sourceUrl ? (
           <ActionLink href={speech.sourceUrl} variant="ghost">
             <ExternalLinkIcon data-icon="inline-start" aria-hidden="true" />
diff --git a/src/lib/schemas/speechVod.ts b/src/lib/schemas/speechVod.ts
new file mode 100644
index 00000000..991ce43f
--- /dev/null
+++ b/src/lib/schemas/speechVod.ts
@@ -0,0 +1,28 @@
+import { z } from 'zod'
+
+import { positiveRelationshipId } from '@/lib/schemas/primitives'
+
+/** The detail page asks the server to resolve one speech's excerpt VOD. */
+export const speechVodRequestSchema = z.object({
+  speechId: positiveRelationshipId,
+})
+
+/**
+ * Safe messages of the resolution route — shared constants on purpose:
+ * `mapCampaignFormActionError` matches by exact string, so a literal at one
+ * end only would silently collapse a real refusal into the generic error.
+ */
+
+/** The catalog gate (`canReadSpeechCatalog`) refused the actor. */
+export const SPEECH_VOD_FORBIDDEN_MESSAGE = 'Você não tem acesso ao acervo de falas.'
+
+/** The speech does not exist (or the actor cannot read it). */
+export const SPEECH_VOD_NOT_FOUND_MESSAGE = 'Fala não encontrada.'
+
+/** No stored VOD or missing `eventId`/`audioId`/`excerptTMs` — never call the Câmara. */
+export const SPEECH_VOD_INELIGIBLE_MESSAGE =
+  'Esta fala não tem trecho de vídeo para resolver na Câmara.'
+
+/** Transport/parse failure at the Câmara (or anything unmapped). */
+export const SPEECH_VOD_GENERIC_ERROR_MESSAGE =
+  'Não foi possível resolver o trecho. Verifique seu acesso e tente novamente.'
diff --git a/src/lib/speechVod.ts b/src/lib/speechVod.ts
new file mode 100644
index 00000000..b303f45e
--- /dev/null
+++ b/src/lib/speechVod.ts
@@ -0,0 +1,193 @@
+/**
+ * Pure VOD rules shared by the Câmara speech scripts (C152 pilot/C153 import)
+ * and the acervo player (C162): the `video-sob-demanda` URL/status contract,
+ * the crawl UA, the session-offset math and the YouTube video id.
+ *
+ * No I/O here — the scripts and the server resolver own the fetch/timeout
+ * policy, so the fragile rules stay unit-testable without network.
+ */
+
+/** Browser-like UA — the trecho/evento pages answer 400 "Acesso via bot" to curl. */
+export const CAMARA_USER_AGENT =
+  'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/125.0.0.0 Safari/537.36'
+
+/** The async VOD-generation status URL for one excerpt. */
+export const buildVodUrl = (eventId: number, audioId: number, tMs: number): string =>
+  `https://www.camara.leg.br/evento-legislativo/${eventId}/video-sob-demanda?idAudio=${audioId}&trecho=${tMs}`
+
+export type CamaraVodVideo = {
+  title: string | null
+  subtitle: string | null
+  duration: string | null
+  clock: string | null
+  downloadUrl: string | null
+  playbackUrl: string | null
+}
+
+export type CamaraVodStatus = {
+  state: string
+  video: CamaraVodVideo | null
+}
+
+/**
+ * What the acervo player can do after asking the Câmara about one excerpt:
+ * `pronto` carries only URLs that passed the media probe (either can still be
+ * null when its own probe failed), `gerando` means the server is transcoding,
+ * `indisponivel` means there is nothing usable.
+ */
+export type SpeechVodResolution =
+  | { state: 'pronto'; playbackUrl: string | null; downloadUrl: string | null }
+  | { state: 'gerando' }
+  | { state: 'indisponivel' }
+
+/**
+ * Normalizes the `video-sob-demanda` JSON (`{ estado, video }`). `state` is
+ * `GERANDO` while the server transcodes, `PRONTO` when the MP4 exists,
+ * `INDISPONIVEL` when there is none.
+ */
+export const parseVodStatus = (json: unknown): CamaraVodStatus => {
+  const data = json && typeof json === 'object' ? (json as Record<string, unknown>) : {}
+  const estado = typeof data.estado === 'string' ? data.estado : 'DESCONHECIDO'
+  const raw =
+    data.video && typeof data.video === 'object' ? (data.video as Record<string, unknown>) : null
+  const pick = (key: string): string | null =>
+    typeof raw?.[key] === 'string' ? (raw[key] as string) : null
+  return {
+    state: estado,
+    video: raw
+      ? {
+          title: pick('titulo'),
+          subtitle: pick('subtitulo'),
+          duration: pick('duracao'),
+          clock: pick('horario'),
+          downloadUrl: pick('linkParaDownload'),
+          playbackUrl: pick('linkParaReproducao'),
+        }
+      : null,
+  }
+}
+
+/**
+ * Parses the Câmara duration spellings into seconds: `0h04'03"` (excerpt card)
+ * and `0:04:07` / `4:07` (VOD status). Null when unparseable.
+ */
+export const parseDurationToSeconds = (value: unknown): number | null => {
+  const raw = String(value ?? '').trim()
+  if (raw === '') return null
+
+  const card = /^(\d+)h(\d{1,2})'(\d{1,2})"?$/.exec(raw)
+  if (card) {
+    const minutes = Number(card[2])
+    const seconds = Number(card[3])
+    if (minutes > 59 || seconds > 59) return null
+    return Number(card[1]) * 3600 + minutes * 60 + seconds
+  }
+
+  const parts = raw.split(':')
+  if (!parts.every((part) => /^\d{1,2}$/.test(part))) return null
+  if (parts.length === 3) {
+    const [hours, minutes, seconds] = parts.map(Number)
+    if (minutes > 59 || seconds > 59) return null
+    return hours * 3600 + minutes * 60 + seconds
+  }
+  if (parts.length === 2) {
+    const [minutes, seconds] = parts.map(Number)
+    if (seconds > 59) return null
+    return minutes * 60 + seconds
+  }
+  return null
+}
+
+const YOUTUBE_VIDEO_ID = /^[A-Za-z0-9_-]{6,20}$/
+const YOUTUBE_HOSTS = new Set(['youtube.com', 'youtube-nocookie.com', 'youtu.be'])
+const YOUTUBE_ID_PATH_PREFIXES = new Set(['embed', 'live', 'shorts', 'v'])
+
+/**
+ * YouTube video id of an `urlRegistro` link, or null when the URL is not a
+ * recognized YouTube form (`watch?v=`, `youtu.be/<id>`, `embed`/`live`/
+ * `shorts`/`v`) or the id does not look like one. Defensive on purpose: the
+ * open-data field varies, and a null simply falls back to the VOD path.
+ */
+export const parseYoutubeVideoId = (raw: unknown): string | null => {
+  const value = String(raw ?? '').trim()
+  if (value === '') return null
+
+  let url: URL
+  try {
+    url = new URL(value)
+  } catch {
+    return null
+  }
+  if (url.protocol !== 'https:' && url.protocol !== 'http:') return null
+
+  const host = url.hostname.toLowerCase().replace(/^(www|m|music)\./, '')
+  if (!YOUTUBE_HOSTS.has(host)) return null
+
+  const segments = url.pathname.split('/').filter(Boolean)
+  let candidate: string | null = null
+  if (host === 'youtu.be') {
+    candidate = segments[0] ?? null
+  } else if (segments[0] === 'watch') {
+    candidate = url.searchParams.get('v')
+  } else if (segments[0] && YOUTUBE_ID_PATH_PREFIXES.has(segments[0])) {
+    candidate = segments[1] ?? null
+  }
+
+  return candidate && YOUTUBE_VIDEO_ID.test(candidate) ? candidate : null
+}
+
+const SAO_PAULO_TIME_ZONE = 'America/Sao_Paulo'
+const NAIVE_DATETIME = /^(\d{4})-(\d{2})-(\d{2})[T ](\d{2}):(\d{2})(?::(\d{2}))?/
+const brtWallClock = new Intl.DateTimeFormat('en-CA', {
+  timeZone: SAO_PAULO_TIME_ZONE,
+  year: 'numeric',
+  month: '2-digit',
+  day: '2-digit',
+  hour: '2-digit',
+  minute: '2-digit',
+  second: '2-digit',
+  hour12: false,
+})
+
+/** BRT wall-clock of an epoch instant, as seconds since epoch-treated-as-UTC. */
+const wallClockSeconds = (epochMs: number): number | null => {
+  const parts = brtWallClock.formatToParts(new Date(epochMs))
+  const get = (type: Intl.DateTimeFormatPartTypes): number =>
+    Number(parts.find((part) => part.type === type)?.value)
+  const [year, month, day, hour, minute, second] = [
+    get('year'),
+    get('month'),
+    get('day'),
+    get('hour'),
+    get('minute'),
+    get('second'),
+  ]
+  if (![year, month, day, hour, minute, second].every(Number.isFinite)) return null
+  return Date.UTC(year, month - 1, day, hour % 24, minute, second) / 1000
+}
+
+/**
+ * Seconds from the session start (`eventStartAt`, naive Câmara wall clock) to
+ * the excerpt (`excerptTMs`, epoch ms). Both sides are read as BRT wall clock,
+ * so a session crossing midnight still measures the right distance. Null when
+ * either side is unparseable or the excerpt precedes the session start.
+ */
+export const excerptOffsetSeconds = (excerptTMs: unknown, eventStartAt: unknown): number | null => {
+  const epochMs = Number(excerptTMs)
+  if (!Number.isFinite(epochMs) || epochMs <= 0) return null
+
+  const match = NAIVE_DATETIME.exec(String(eventStartAt ?? '').trim())
+  if (!match) return null
+  const [, year, month, day, hour, minute, second = '0'] = match
+  const [y, mo, d, h, mi, s] = [year, month, day, hour, minute, second].map(Number)
+  if (mo < 1 || mo > 12 || d < 1 || d > 31 || h > 23 || mi > 59 || s > 59) {
+    return null
+  }
+  const startSeconds = Date.UTC(y, mo - 1, d, h, mi, s) / 1000
+
+  const excerptSeconds = wallClockSeconds(epochMs)
+  if (excerptSeconds === null) return null
+
+  const offset = Math.floor(excerptSeconds - startSeconds)
+  return offset >= 0 ? offset : null
+}
diff --git a/src/utilities/speech/speechPageData.ts b/src/utilities/speech/speechPageData.ts
index 6b0e8599..f8790573 100644
--- a/src/utilities/speech/speechPageData.ts
+++ b/src/utilities/speech/speechPageData.ts
@@ -37,7 +37,6 @@ const speechListSelect = {
   scopes: true,
   mentionedMunicipalities: true,
   presidingOfficer: true,
-  vodDownloadUrl: true,
   officialTextUrl: true,
   youtubeUrl: true,
 } as const
@@ -46,6 +45,11 @@ const speechDetailSelect = {
   ...speechListSelect,
   phase: true,
   vodPlaybackUrl: true,
+  vodDownloadUrl: true,
+  eventId: true,
+  audioId: true,
+  excerptTMs: true,
+  eventStartAt: true,
 } as const
 
 const segmentSelect = {
diff --git a/src/utilities/speech/speechViewModels.ts b/src/utilities/speech/speechViewModels.ts
index baf05269..c351eea9 100644
--- a/src/utilities/speech/speechViewModels.ts
+++ b/src/utilities/speech/speechViewModels.ts
@@ -15,6 +15,7 @@ import {
   type SpeechHighlightPart,
 } from '@/lib/speechHighlight'
 import { normalizeForSearch } from '@/lib/speechSearch'
+import { excerptOffsetSeconds, parseYoutubeVideoId } from '@/lib/speechVod'
 import type { Municipality } from '@/payload-types'
 import { speechScopeLabels, speechTopicLabels } from '@/utilities/speech/speechListUrl'
 
@@ -33,7 +34,6 @@ export type SpeechListRecord = {
   scopes?: SpeechScope[] | null
   mentionedMunicipalities?: (number | Municipality)[] | null
   presidingOfficer?: string | null
-  vodDownloadUrl?: string | null
   officialTextUrl?: string | null
   youtubeUrl?: string | null
 }
@@ -53,7 +53,6 @@ export type SpeechListItemViewModel = {
   keywords: string[]
   municipalities: { id: number; name: string }[]
   watchHref: string
-  downloadUrl: string | null
   sourceUrl: string | null
 }
 
@@ -80,8 +79,12 @@ export type SpeechDetailViewModel = {
   scopes: { value: SpeechScope; label: string }[]
   municipalities: { id: number; name: string }[]
   segments: SpeechDetailSegmentViewModel[]
-  vodPlaybackUrl: string | null
-  vodDownloadUrl: string | null
+  /** YouTube default source (C162): id when the session link parses, else null. */
+  youtubeVideoId: string | null
+  /** Session offset in seconds (excerpt epoch − session start), null when unknowable. */
+  youtubeOffsetSeconds: number | null
+  /** Stored VOD + excerpt coordinates — the Câmara may be asked on click. */
+  vodResolvable: boolean
 }
 
 const pad = (value: number): string => String(value).padStart(2, '0')
@@ -215,7 +218,6 @@ export const toSpeechListItemViewModel = ({
     keywords: speech.keywords ?? [],
     municipalities: municipalityViewModels(speech, municipalityLabels),
     watchHref: buildWatchHref(speech.id, matchedSegment, q),
-    downloadUrl: speech.vodDownloadUrl ?? null,
     sourceUrl: speech.officialTextUrl ?? speech.youtubeUrl ?? null,
   }
 }
@@ -223,6 +225,11 @@ export const toSpeechListItemViewModel = ({
 export type SpeechDetailRecord = SpeechListRecord & {
   phase?: string | null
   vodPlaybackUrl?: string | null
+  vodDownloadUrl?: string | null
+  eventId?: number | null
+  audioId?: number | null
+  excerptTMs?: number | null
+  eventStartAt?: string | null
 }
 
 export const toSpeechDetailViewModel = ({
@@ -235,27 +242,39 @@ export const toSpeechDetailViewModel = ({
   segments: readonly SpeechSegmentRecord[]
   query?: string
   municipalityLabels: ReadonlyMap<number, string>
-}): SpeechDetailViewModel => ({
-  id: speech.id,
-  speechAtLabel: formatSpeechAt(speech.speechAt),
-  type: speech.type ?? null,
-  phase: speech.phase ?? null,
-  durationLabel: formatSpeechDuration(speech.durationSeconds),
-  presidingOfficer: speech.presidingOfficer ?? null,
-  summary: speech.summary ?? null,
-  officialTranscript: speech.officialTranscript ?? null,
-  officialTextUrl: speech.officialTextUrl ?? null,
-  youtubeUrl: speech.youtubeUrl ?? null,
-  keywords: speech.keywords ?? [],
-  topics: topicViewModels(speech),
-  scopes: scopeViewModels(speech),
-  municipalities: municipalityViewModels(speech, municipalityLabels),
-  segments: segments.map((segment) => ({
-    startSeconds: segment.startSeconds,
-    endSeconds: segment.endSeconds,
-    startLabel: formatSpeechClock(segment.startSeconds),
-    parts: splitHighlightedParts(segment.text, query ?? ''),
-  })),
-  vodPlaybackUrl: speech.vodPlaybackUrl ?? null,
-  vodDownloadUrl: speech.vodDownloadUrl ?? null,
-})
+}): SpeechDetailViewModel => {
+  // The stored VOD link is a cache and never a URL handed to the client; it
+  // only signals that this record had a generated excerpt to re-resolve.
+  const vodResolvable = Boolean(
+    (speech.vodPlaybackUrl || speech.vodDownloadUrl) &&
+    speech.eventId &&
+    speech.audioId &&
+    speech.excerptTMs,
+  )
+
+  return {
+    id: speech.id,
+    speechAtLabel: formatSpeechAt(speech.speechAt),
+    type: speech.type ?? null,
+    phase: speech.phase ?? null,
+    durationLabel: formatSpeechDuration(speech.durationSeconds),
+    presidingOfficer: speech.presidingOfficer ?? null,
+    summary: speech.summary ?? null,
+    officialTranscript: speech.officialTranscript ?? null,
+    officialTextUrl: speech.officialTextUrl ?? null,
+    youtubeUrl: speech.youtubeUrl ?? null,
+    keywords: speech.keywords ?? [],
+    topics: topicViewModels(speech),
+    scopes: scopeViewModels(speech),
+    municipalities: municipalityViewModels(speech, municipalityLabels),
+    segments: segments.map((segment) => ({
+      startSeconds: segment.startSeconds,
+      endSeconds: segment.endSeconds,
+      startLabel: formatSpeechClock(segment.startSeconds),
+      parts: splitHighlightedParts(segment.text, query ?? ''),
+    })),
+    youtubeVideoId: parseYoutubeVideoId(speech.youtubeUrl),
+    youtubeOffsetSeconds: excerptOffsetSeconds(speech.excerptTMs, speech.eventStartAt),
+    vodResolvable,
+  }
+}
diff --git a/src/utilities/speech/speechVodResolver.ts b/src/utilities/speech/speechVodResolver.ts
new file mode 100644
index 00000000..1c1e7450
--- /dev/null
+++ b/src/utilities/speech/speechVodResolver.ts
@@ -0,0 +1,102 @@
+import 'server-only'
+
+import {
+  buildVodUrl,
+  CAMARA_USER_AGENT,
+  parseVodStatus,
+  type SpeechVodResolution,
+} from '@/lib/speechVod'
+
+/**
+ * C162 — interactive VOD resolution for the acervo detail. Deliberately NOT a
+ * reuse of `scripts/lib/camaraFetch.mjs` (whose `resolveVod` polls for ~200s
+ * and whose `probeLink` reports instead of deciding): the player's contract is
+ * a bounded click, so this module owns its own short retry, timeout and the
+ * media verification that decides what may reach the browser.
+ */
+
+const VOD_STATUS_ATTEMPTS = 2
+const VOD_STATUS_TIMEOUT_MS = 15_000
+const VOD_STATUS_RETRY_DELAY_MS = 1_000
+const MEDIA_PROBE_TIMEOUT_MS = 10_000
+
+const sleep = (ms: number) => new Promise((resolve) => setTimeout(resolve, ms))
+
+/** One bounded fetch of the `video-sob-demanda` JSON; throws after the retries. */
+const fetchVodStatus = async (url: string): Promise<unknown> => {
+  let lastError: unknown
+  for (let attempt = 1; attempt <= VOD_STATUS_ATTEMPTS; attempt += 1) {
+    try {
+      const response = await fetch(url, {
+        headers: { 'User-Agent': CAMARA_USER_AGENT, Accept: 'application/json' },
+        signal: AbortSignal.timeout(VOD_STATUS_TIMEOUT_MS),
+      })
+      if (!response.ok) throw new Error(`HTTP ${response.status} em ${url}`)
+      return await response.json()
+    } catch (error) {
+      lastError = error
+      if (attempt < VOD_STATUS_ATTEMPTS) await sleep(VOD_STATUS_RETRY_DELAY_MS)
+    }
+  }
+  throw lastError
+}
+
+type MediaProbe = { verified: boolean; retryWithHead: boolean }
+
+/**
+ * A media link counts as verified when the ranged GET answers 2xx with a real
+ * content type — a `text/html` body is the CDN's error page wearing a 200.
+ */
+const probeMediaOnce = async (url: string, method: 'GET' | 'HEAD'): Promise<MediaProbe> => {
+  const response = await fetch(url, {
+    method,
+    headers: {
+      'User-Agent': CAMARA_USER_AGENT,
+      ...(method === 'GET' ? { Range: 'bytes=0-1023' } : {}),
+    },
+    signal: AbortSignal.timeout(MEDIA_PROBE_TIMEOUT_MS),
+  })
+  const contentType = response.headers.get('content-type')
+  const verified = response.ok && contentType !== null && !contentType.includes('text/html')
+  await response.body?.cancel().catch(() => undefined)
+  return { verified, retryWithHead: response.status === 405 || response.status === 501 }
+}
+
+/** Never throws: a transport failure is just an unverified link. */
+const probeSpeechMedia = async (url: string | null): Promise<boolean> => {
+  if (!url) return false
+  try {
+    const first = await probeMediaOnce(url, 'GET')
+    if (first.verified) return true
+    if (first.retryWithHead) return (await probeMediaOnce(url, 'HEAD')).verified
+    return false
+  } catch {
+    return false
+  }
+}
+
+export const resolveSpeechVod = async ({
+  eventId,
+  audioId,
+  excerptTms,
+}: {
+  eventId: number
+  audioId: number
+  excerptTms: number
+}): Promise<SpeechVodResolution> => {
+  const status = parseVodStatus(await fetchVodStatus(buildVodUrl(eventId, audioId, excerptTms)))
+
+  if (status.state === 'GERANDO') return { state: 'gerando' }
+  if (status.state !== 'PRONTO' || !status.video) return { state: 'indisponivel' }
+
+  const [playback, download] = await Promise.all([
+    probeSpeechMedia(status.video.playbackUrl),
+    probeSpeechMedia(status.video.downloadUrl),
+  ])
+
+  return {
+    state: 'pronto',
+    playbackUrl: playback ? status.video.playbackUrl : null,
+    downloadUrl: download ? status.video.downloadUrl : null,
+  }
+}
diff --git a/tests/e2e/campaignSpeechAcervo.e2e.spec.ts b/tests/e2e/campaignSpeechAcervo.e2e.spec.ts
index a439e969..afceed17 100644
--- a/tests/e2e/campaignSpeechAcervo.e2e.spec.ts
+++ b/tests/e2e/campaignSpeechAcervo.e2e.spec.ts
@@ -1,16 +1,27 @@
+import { request as playwrightRequest } from '@playwright/test'
+
 import type { SpeechScope, SpeechTopic } from '@/lib/speechFacets'
 
 import type { CampaignE2EOwnership } from './fixtures/campaignE2EFixtures.js'
 import { assertCampaignRedirect, expect, rendered, test } from './fixtures/campaignHttpTest.js'
 
 /**
- * C154 — the communication vertical over real HTTP (browserless, OPS35/OPS87):
- * role gates, the acervo search rendering (highlight + excerpt), the filter
- * URL contract and the detail's player/transcript/actions. Playback itself is
- * browser territory and is not asserted here — the HTML carries the contract
- * (`<video src>`, `data-start-seconds`).
+ * C154/C162 — the communication vertical over real HTTP (browserless,
+ * OPS35/OPS87): role gates, the acervo search rendering (highlight + excerpt),
+ * the filter URL contract and the detail's player/transcript/actions. The
+ * player picks the surface from the row coordinates — YouTube default when the
+ * session link parses, stored-VOD capa resolved on click otherwise, honest
+ * unavailable state when neither exists — and the stored VOD link itself is
+ * never rendered (C162). Playback is browser territory and is not asserted
+ * here; the HTML carries the contract (`<video src>` only after a click,
+ * `data-start-seconds`, the iframe `start`).
  */
 
+const YOUTUBE_VIDEO_ID = 'lLhRDkSPw0A'
+const EXCERPT_TMS = 1786473834650
+const EXCERPT_SESSION = '2026-08-11T15:00'
+const EXCERPT_OFFSET_SECONDS = 2634
+
 const createSpeech = async (
   campaign: { fixtures: CampaignE2EOwnership },
   input: {
@@ -21,6 +32,8 @@ const createSpeech = async (
     keywords?: string[]
     durationSeconds?: number
     withVideo?: boolean
+    youtubeUrl?: string | null
+    eventStartAt?: string
   },
 ) => {
   const marker = input.marker
@@ -42,6 +55,11 @@ const createSpeech = async (
       searchText: `a retomada da saude ${marker} na bahia`,
       presidingOfficer: 'Pompeo de Mattos',
       officialTextUrl: 'https://camara.leg.br/discurso',
+      eventId: 67091,
+      audioId: 558641,
+      excerptTMs: EXCERPT_TMS,
+      eventStartAt: input.eventStartAt ?? EXCERPT_SESSION,
+      youtubeUrl: input.youtubeUrl ?? null,
       ...(input.withVideo === false
         ? {}
         : {
@@ -66,7 +84,7 @@ const createSpeech = async (
   return speech
 }
 
-test.describe('communication vertical (C154)', () => {
+test.describe('communication vertical (C154/C162)', () => {
   test('communicator lands on the acervo and sees only the vertical nav, with her scoped Sollinha', async ({
     campaign,
     campaignRequest,
@@ -111,12 +129,15 @@ test.describe('communication vertical (C154)', () => {
     expect(rendered(await advisorHome.text())).not.toContain('href="/campanha/comunicacao"')
   })
 
-  test('search renders the highlighted excerpt and the detail carries the player', async ({
+  test('search keeps the download off the card; the both-sources detail defaults to YouTube', async ({
     campaign,
     campaignRequest,
   }) => {
     const marker = campaign.fixtures.value('falateste')
-    const speech = await createSpeech(campaign, { marker })
+    const speech = await createSpeech(campaign, {
+      marker,
+      youtubeUrl: `https://www.youtube.com/watch?v=${YOUTUBE_VIDEO_ID}`,
+    })
 
     const user = await campaign.fixtures.createCampaignUser('communicator')
     const request = await campaignRequest(user, user.password)
@@ -128,16 +149,46 @@ test.describe('communication vertical (C154)', () => {
     expect(resultsHtml).toContain('<mark')
     expect(resultsHtml).toContain('Assistir no trecho')
     expect(resultsHtml).toContain('1 fala encontrada')
+    // C162 — the direct download left the card, and the stored ephemeral link
+    // never reaches any list HTML.
+    expect(resultsHtml).not.toContain('Baixar')
+    expect(resultsHtml).not.toContain('vod.camara.leg.br')
 
-    const detail = await request.get(`/campanha/comunicacao/acervo/${speech.id}?t=0&q=${marker}`)
+    const detail = await request.get(`/campanha/comunicacao/acervo/${speech.id}?t=43&q=${marker}`)
     expect(detail.status()).toBe(200)
     const detailHtml = rendered(await detail.text())
     expect(detailHtml).toContain('data-slot="speech-player"')
-    expect(detailHtml).toContain('<video')
+    // Default surface is the YouTube embed, positioned by the session offset
+    // (2634) plus the deep-linked transcript second (43).
+    expect(detailHtml).toContain(`youtube-nocookie.com/embed/${YOUTUBE_VIDEO_ID}`)
+    expect(detailHtml).toContain(`start=${EXCERPT_OFFSET_SECONDS + 43}`)
     expect(detailHtml).toContain('data-start-seconds="0"')
     expect(detailHtml).toContain('Baixar vídeo (MP4)')
     expect(detailHtml).toContain('Abrir fonte')
     expect(detailHtml).toContain('Fonte: Câmara dos Deputados')
+    expect(detailHtml).not.toContain('<video')
+    expect(detailHtml).not.toContain('vod.camara.leg.br')
+  })
+
+  test('a VOD-only speech offers the on-click resolution instead of a player', async ({
+    campaign,
+    campaignRequest,
+  }) => {
+    const marker = campaign.fixtures.value('sovod')
+    const speech = await createSpeech(campaign, { marker })
+
+    const user = await campaign.fixtures.createCampaignUser('communicator')
+    const request = await campaignRequest(user, user.password)
+
+    const detail = await request.get(`/campanha/comunicacao/acervo/${speech.id}`)
+    expect(detail.status()).toBe(200)
+    const html = rendered(await detail.text())
+    expect(html).toContain('Assistir o trecho')
+    expect(html).toContain('Baixar vídeo (MP4)')
+    expect(html).not.toContain('<video')
+    expect(html).not.toContain('<iframe')
+    // The stored link is an eligibility signal, never a rendered URL.
+    expect(html).not.toContain('vod.camara.leg.br')
   })
 
   test('facet filters narrow the results and the empty state suggests clearing', async ({
@@ -186,7 +237,7 @@ test.describe('communication vertical (C154)', () => {
     )
   })
 
-  test('a speech without VOD renders the unavailable block instead of a player', async ({
+  test('a speech without video renders the unavailable block instead of a player', async ({
     campaign,
     campaignRequest,
   }) => {
@@ -198,7 +249,66 @@ test.describe('communication vertical (C154)', () => {
 
     const detail = await request.get(`/campanha/comunicacao/acervo/${speech.id}`)
     const html = rendered(await detail.text())
-    expect(html).toContain('VOD indisponível')
+    expect(html).toContain('Vídeo indisponível neste momento')
+    expect(html).toContain('Abrir fonte')
+    // No stored VOD means nothing to resolve: no MP4 button and no retry.
     expect(html).not.toContain('Baixar vídeo (MP4)')
+    expect(html).not.toContain('Tentar novamente')
+  })
+
+  test.describe('POST /campanha/comunicacao/acervo/resolver-vod (C162)', () => {
+    const endpoint = '/campanha/comunicacao/acervo/resolver-vod'
+
+    test('rejects a request without a campaign session', async ({ campaign }) => {
+      const anonymous = await playwrightRequest.newContext({ baseURL: campaign.baseURL })
+      try {
+        const response = await anonymous.post(endpoint, { data: { speechId: 1 } })
+        expect(response.status()).toBe(401)
+      } finally {
+        await anonymous.dispose()
+      }
+    })
+
+    test('rejects a cross-origin request before any read', async ({
+      campaign,
+      campaignRequest,
+    }) => {
+      const marker = campaign.fixtures.value('origem')
+      const speech = await createSpeech(campaign, { marker })
+      const user = await campaign.fixtures.createCampaignUser('communicator')
+      const request = await campaignRequest(user, user.password)
+
+      const response = await request.post(endpoint, {
+        data: { speechId: speech.id },
+        headers: { Origin: 'https://evil.example' },
+      })
+      expect(response.status()).toBe(403)
+      expect(await response.text()).toContain('Requisição inválida.')
+    })
+
+    test('refuses an actor outside the catalog gate', async ({ campaign, campaignRequest }) => {
+      const marker = campaign.fixtures.value('assessor')
+      const speech = await createSpeech(campaign, { marker })
+      const advisor = await campaign.fixtures.createCampaignUser('advisor')
+      const request = await campaignRequest(advisor, advisor.password)
+
+      const response = await request.post(endpoint, { data: { speechId: speech.id } })
+      expect(response.status()).toBe(400)
+      expect(await response.text()).toContain('Você não tem acesso ao acervo de falas.')
+    })
+
+    test('refuses a speech without a stored VOD without calling the Câmara', async ({
+      campaign,
+      campaignRequest,
+    }) => {
+      const marker = campaign.fixtures.value('inelegivel')
+      const speech = await createSpeech(campaign, { marker, withVideo: false })
+      const user = await campaign.fixtures.createCampaignUser('communicator')
+      const request = await campaignRequest(user, user.password)
+
+      const response = await request.post(endpoint, { data: { speechId: speech.id } })
+      expect(response.status()).toBe(400)
+      expect(await response.text()).toContain('não tem trecho de vídeo')
+    })
   })
 })
diff --git a/tests/int/speechAcervo.int.spec.ts b/tests/int/speechAcervo.int.spec.ts
index 284bcf07..cc1bfd56 100644
--- a/tests/int/speechAcervo.int.spec.ts
+++ b/tests/int/speechAcervo.int.spec.ts
@@ -196,6 +196,38 @@ describe('speech acervo (C154)', () => {
     expect(view.presidingOfficer).toBe('Pompeo de Mattos')
   })
 
+  it('derives the YouTube default and the resoluble VOD from the detail row (C162)', async () => {
+    const withBoth = await createSpeech({
+      youtubeUrl: 'https://www.youtube.com/watch?v=lLhRDkSPw0A',
+      vodPlaybackUrl: 'https://vod.camara.leg.br/old.mp4',
+      vodDownloadUrl: 'https://vod.camara.leg.br/old.mp4',
+      excerptTMs: 1786473834650,
+      eventStartAt: '2026-08-11T15:00',
+    })
+    const onlyYoutube = await createSpeech({ youtubeUrl: 'https://youtu.be/lLhRDkSPw0A' })
+
+    const { communicator } = await createUsers()
+    const both = await loadSpeechDetailPageData(payload, communicator, withBoth)
+    expect(both.youtubeVideoId).toBe('lLhRDkSPw0A')
+    expect(both.youtubeOffsetSeconds).toBe(2634)
+    expect(both.vodResolvable).toBe(true)
+
+    const youtubeOnly = await loadSpeechDetailPageData(payload, communicator, onlyYoutube)
+    expect(youtubeOnly.youtubeVideoId).toBe('lLhRDkSPw0A')
+    expect(youtubeOnly.vodResolvable).toBe(false)
+  })
+
+  it('keeps the direct download link off the list view model (C162)', async () => {
+    const id = await createSpeech({ vodDownloadUrl: 'https://vod.camara.leg.br/old.mp4' })
+
+    const { communicator } = await createUsers()
+    const data = await loadSpeechAcervoPageData(payload, communicator, {})
+    const row = data.rows.find((item) => item.id === id)
+
+    expect(row).toBeDefined()
+    expect('downloadUrl' in (row as object)).toBe(false)
+  })
+
   it('allows communicator/coordinator/candidate and denies advisor/leader', async () => {
     const id = await createSpeech()
     const { communicator, coordinator, candidate, advisor, leader } = await createUsers()
diff --git a/tests/unit/speechDetailPlayer.unit.spec.tsx b/tests/unit/speechDetailPlayer.unit.spec.tsx
new file mode 100644
index 00000000..ff41fa9a
--- /dev/null
+++ b/tests/unit/speechDetailPlayer.unit.spec.tsx
@@ -0,0 +1,246 @@
+import { act, cleanup, fireEvent, render, screen, waitFor } from '@testing-library/react'
+import { afterEach, beforeAll, beforeEach, describe, expect, it, vi } from 'vitest'
+
+import { SpeechDetailPlayer } from '@/components/campaign/speech/SpeechDetailPlayer'
+import type { SpeechDetailSegmentViewModel } from '@/utilities/speech/speechViewModels'
+
+const PLAYBACK_URL = 'https://vod.camara.leg.br/trecho.mp4'
+const DOWNLOAD_URL = 'https://vod.camara.leg.br/trecho-download.mp4'
+
+const SEGMENTS: readonly SpeechDetailSegmentViewModel[] = [
+  {
+    startSeconds: 0,
+    endSeconds: 5,
+    startLabel: '00:00',
+    parts: [{ text: 'Primeiro trecho', highlighted: false }],
+  },
+  {
+    startSeconds: 43,
+    endSeconds: 50,
+    startLabel: '00:43',
+    parts: [{ text: 'O trecho sobre a saúde', highlighted: false }],
+  },
+]
+
+const pronto = (playbackUrl: string | null, downloadUrl: string | null) => ({
+  status: 'success' as const,
+  resolution: { state: 'pronto' as const, playbackUrl, downloadUrl },
+})
+const gerando = { status: 'success' as const, resolution: { state: 'gerando' as const } }
+const indisponivel = { status: 'success' as const, resolution: { state: 'indisponivel' as const } }
+
+const fetchMock = vi.fn()
+
+const respondWith = (payload: unknown, ok = true) => {
+  fetchMock.mockResolvedValue({ ok, json: async () => payload })
+}
+
+const deferred = <T,>() => {
+  let resolve!: (value: T) => void
+  const promise = new Promise<T>((promiseResolve) => {
+    resolve = promiseResolve
+  })
+  return { promise, resolve }
+}
+
+const renderPlayer = (props: Partial<Parameters<typeof SpeechDetailPlayer>[0]> = {}) =>
+  render(
+    <SpeechDetailPlayer
+      speechId={42}
+      youtubeVideoId={null}
+      youtubeOffsetSeconds={null}
+      vodResolvable
+      segments={SEGMENTS}
+      initialSeconds={null}
+      sourceUrl={null}
+      {...props}
+    />,
+  )
+
+const videoElement = () => document.querySelector('video')
+const iframeElement = () => document.querySelector('iframe')
+const playerRoot = () => document.querySelector('[data-slot="speech-player"]')
+
+beforeAll(() => {
+  // jsdom does not implement media playback; the seek contract is what matters.
+  Object.defineProperty(HTMLMediaElement.prototype, 'play', {
+    configurable: true,
+    value: () => Promise.resolve(),
+  })
+  Object.defineProperty(HTMLMediaElement.prototype, 'currentTime', {
+    configurable: true,
+    writable: true,
+    value: 0,
+  })
+})
+
+beforeEach(() => {
+  fetchMock.mockReset()
+  vi.stubGlobal('fetch', fetchMock)
+})
+
+afterEach(() => {
+  cleanup()
+  vi.unstubAllGlobals()
+  vi.restoreAllMocks()
+})
+
+describe('SpeechDetailPlayer — VOD-only quadrant', () => {
+  it('resolves on click, shows the pending state and plays the exact excerpt', async () => {
+    const pending = deferred<{ ok: boolean; json: () => Promise<unknown> }>()
+    fetchMock.mockReturnValue(pending.promise)
+
+    renderPlayer()
+
+    expect(screen.getByText(/O trecho deste vídeo é gerado pela Câmara/)).toBeDefined()
+    expect(videoElement()).toBeNull()
+    expect(iframeElement()).toBeNull()
+
+    fireEvent.click(screen.getByRole('button', { name: /assistir o trecho/i }))
+
+    await waitFor(() => {
+      expect(playerRoot()?.getAttribute('aria-busy')).toBe('true')
+    })
+    expect(screen.getByText(/Resolvendo o trecho na Câmara/)).toBeDefined()
+    const busyDownload = screen.getByRole('button', { name: /baixar vídeo \(mp4\)/i })
+    expect((busyDownload as HTMLButtonElement).disabled).toBe(true)
+
+    await act(async () => {
+      pending.resolve({ ok: true, json: async () => pronto(PLAYBACK_URL, DOWNLOAD_URL) })
+    })
+
+    await waitFor(() => {
+      expect(videoElement()?.getAttribute('src')).toBe(PLAYBACK_URL)
+    })
+    expect(playerRoot()?.getAttribute('aria-busy')).toBeNull()
+
+    fireEvent.click(document.querySelector('button[data-start-seconds="43"]')!)
+    expect(videoElement()?.currentTime).toBe(43)
+  })
+
+  it('shows the honest state after a failure and re-posts on retry', async () => {
+    respondWith(indisponivel)
+    renderPlayer()
+
+    fireEvent.click(screen.getByRole('button', { name: /assistir o trecho/i }))
+    expect(await screen.findByText('Não foi possível carregar o vídeo deste trecho.')).toBeDefined()
+
+    respondWith(pronto(PLAYBACK_URL, null))
+    fireEvent.click(screen.getByRole('button', { name: /tentar novamente/i }))
+
+    await waitFor(() => {
+      expect(videoElement()?.getAttribute('src')).toBe(PLAYBACK_URL)
+    })
+    expect(fetchMock).toHaveBeenCalledTimes(2)
+    expect(JSON.parse(String(fetchMock.mock.calls[0]?.[1]?.body))).toEqual({ speechId: 42 })
+  })
+
+  it('shows the generating state with a retry and no player', async () => {
+    respondWith(gerando)
+    renderPlayer()
+
+    fireEvent.click(screen.getByRole('button', { name: /assistir o trecho/i }))
+
+    expect(await screen.findByText(/A Câmara está gerando o trecho deste vídeo/)).toBeDefined()
+    expect(screen.getByRole('button', { name: /tentar novamente/i })).toBeDefined()
+    expect(videoElement()).toBeNull()
+  })
+
+  it('keeps the verified download offered when only the playback probe failed', async () => {
+    respondWith(pronto(null, DOWNLOAD_URL))
+    renderPlayer()
+
+    fireEvent.click(screen.getByRole('button', { name: /assistir o trecho/i }))
+
+    expect(await screen.findByText('Não foi possível carregar o vídeo deste trecho.')).toBeDefined()
+    expect(screen.getByRole('button', { name: /baixar vídeo \(mp4\)/i })).toBeDefined()
+    expect(videoElement()).toBeNull()
+  })
+})
+
+describe('SpeechDetailPlayer — YouTube quadrant', () => {
+  it('embeds the session at the offset and seeks the transcript by reloading start', async () => {
+    renderPlayer({
+      youtubeVideoId: 'lLhRDkSPw0A',
+      youtubeOffsetSeconds: 2634,
+      initialSeconds: 100,
+    })
+
+    expect(videoElement()).toBeNull()
+    expect(iframeElement()?.getAttribute('src')).toBe(
+      'https://www.youtube-nocookie.com/embed/lLhRDkSPw0A?playsinline=1&rel=0&start=2734',
+    )
+
+    fireEvent.click(document.querySelector('button[data-start-seconds="43"]')!)
+
+    await waitFor(() => {
+      expect(iframeElement()?.getAttribute('src')).toContain('start=2677')
+    })
+  })
+
+  it('leaves the transcript inert when the session offset is unknown', () => {
+    renderPlayer({ youtubeVideoId: 'lLhRDkSPw0A', youtubeOffsetSeconds: null })
+
+    expect(screen.queryByText(/clique para posicionar/i)).toBeNull()
+    const segment = document.querySelector('button[data-start-seconds="43"]') as HTMLButtonElement
+    expect(segment.disabled).toBe(true)
+  })
+
+  it('resolves the MP4 on download and opens the verified URL in a pre-opened tab', async () => {
+    const pending = deferred<{ ok: boolean; json: () => Promise<unknown> }>()
+    fetchMock.mockReturnValue(pending.promise)
+    const tab = { location: { href: '' }, close: vi.fn() }
+    const openSpy = vi.spyOn(window, 'open').mockReturnValue(tab as unknown as Window)
+
+    renderPlayer({
+      youtubeVideoId: 'lLhRDkSPw0A',
+      youtubeOffsetSeconds: 2634,
+      initialSeconds: null,
+    })
+
+    fireEvent.click(screen.getByRole('button', { name: /baixar vídeo \(mp4\)/i }))
+    expect(openSpy).toHaveBeenCalledWith('', '_blank')
+    expect(
+      (screen.getByRole('button', { name: /baixar vídeo \(mp4\)/i }) as HTMLButtonElement).disabled,
+    ).toBe(true)
+
+    await act(async () => {
+      pending.resolve({ ok: true, json: async () => pronto(PLAYBACK_URL, DOWNLOAD_URL) })
+    })
+
+    await waitFor(() => {
+      expect(tab.location.href).toBe(DOWNLOAD_URL)
+    })
+    expect(tab.close).not.toHaveBeenCalled()
+    expect(iframeElement()).not.toBeNull()
+  })
+
+  it('closes the tab and shows the honest notice when the resolution fails', async () => {
+    respondWith({ status: 'error', message: 'Não foi possível resolver o trecho.' }, false)
+    const tab = { location: { href: '' }, close: vi.fn() }
+    vi.spyOn(window, 'open').mockReturnValue(tab as unknown as Window)
+
+    renderPlayer({ youtubeVideoId: 'lLhRDkSPw0A', youtubeOffsetSeconds: 2634 })
+
+    fireEvent.click(screen.getByRole('button', { name: /baixar vídeo \(mp4\)/i }))
+
+    await waitFor(() => {
+      expect(tab.close).toHaveBeenCalled()
+    })
+    expect(await screen.findByText('Não foi possível resolver o trecho.')).toBeDefined()
+    expect(iframeElement()).not.toBeNull()
+  })
+})
+
+describe('SpeechDetailPlayer — no-video quadrant', () => {
+  it('renders the unavailable block without retry or download', () => {
+    renderPlayer({ vodResolvable: false, sourceUrl: 'https://imagem.camara.leg.br/diario.pdf' })
+
+    expect(screen.getByText('Vídeo indisponível neste momento.')).toBeDefined()
+    expect(screen.queryByRole('button', { name: /tentar novamente/i })).toBeNull()
+    expect(screen.queryByRole('button', { name: /baixar vídeo/i })).toBeNull()
+    expect(screen.getByRole('link', { name: /abrir fonte/i })).toBeDefined()
+    expect(videoElement()).toBeNull()
+    expect(iframeElement()).toBeNull()
+  })
+})
diff --git a/tests/unit/speechVod.unit.spec.ts b/tests/unit/speechVod.unit.spec.ts
new file mode 100644
index 00000000..b73c56d8
--- /dev/null
+++ b/tests/unit/speechVod.unit.spec.ts
@@ -0,0 +1,84 @@
+// @vitest-environment node
+
+import { describe, expect, it } from 'vitest'
+
+import { excerptOffsetSeconds, parseYoutubeVideoId } from '@/lib/speechVod'
+
+/**
+ * The VOD URL/status/duration contracts moved from `scripts/lib/camaraSpeeches.mjs`
+ * to `@/lib/speechVod` (C162) stay pinned by `camaraSpeeches.unit.spec.ts`, which
+ * imports them through the script's re-export — this file owns the new pure rules.
+ */
+
+describe('excerptOffsetSeconds', () => {
+  it('measures the offset between the excerpt and the session start (intent literal)', () => {
+    expect(excerptOffsetSeconds(1786473834650, '2026-08-11T15:00')).toBe(2634)
+  })
+
+  it('matches the C153 import fixture (import literal)', () => {
+    expect(excerptOffsetSeconds(1675801808560, '2023-02-07T15:00')).toBe(9008)
+  })
+
+  it('reads both sides as BRT wall clock across an old DST transition', () => {
+    // 2018-11-04 00:00 BRT: clocks move to -02:00; 14:00Z reads as 12:00 BRT.
+    expect(excerptOffsetSeconds(1541340000000, '2018-11-04T11:00')).toBe(3600)
+    // The day before the transition still reads at -03:00: 14:00Z = 11:00 BRT.
+    expect(excerptOffsetSeconds(1541253600000, '2018-11-03T10:00')).toBe(3600)
+  })
+
+  it('crosses midnight without losing the date', () => {
+    // 2024-05-15 02:30Z = 2024-05-14 23:30 BRT.
+    expect(excerptOffsetSeconds(1715740200000, '2024-05-14T23:00')).toBe(1800)
+  })
+
+  it('accepts the space spelling of the naive datetime', () => {
+    expect(excerptOffsetSeconds(1786473834650, '2026-08-11 15:00')).toBe(2634)
+  })
+
+  it('returns null when the excerpt precedes the session start', () => {
+    expect(excerptOffsetSeconds(1786473834650, '2026-08-11T16:00')).toBeNull()
+  })
+
+  it('returns null on unparseable or out-of-range input', () => {
+    expect(excerptOffsetSeconds(null, '2026-08-11T15:00')).toBeNull()
+    expect(excerptOffsetSeconds(0, '2026-08-11T15:00')).toBeNull()
+    expect(excerptOffsetSeconds(1786473834650, null)).toBeNull()
+    expect(excerptOffsetSeconds(1786473834650, 'sem data')).toBeNull()
+    expect(excerptOffsetSeconds(1786473834650, '2026-13-11T15:00')).toBeNull()
+    expect(excerptOffsetSeconds(1786473834650, '2026-08-11T25:00')).toBeNull()
+    expect(excerptOffsetSeconds(1786473834650, '2026-08-11T15:75')).toBeNull()
+  })
+})
+
+describe('parseYoutubeVideoId', () => {
+  const videoId = 'lLhRDkSPw0A'
+
+  it('parses the open-data watch link', () => {
+    expect(parseYoutubeVideoId(`https://www.youtube.com/watch?v=${videoId}`)).toBe(videoId)
+  })
+
+  it('parses the shortened, embed, live, shorts and mobile forms', () => {
+    expect(parseYoutubeVideoId(`https://youtu.be/${videoId}?t=30`)).toBe(videoId)
+    expect(parseYoutubeVideoId(`https://www.youtube.com/embed/${videoId}`)).toBe(videoId)
+    expect(parseYoutubeVideoId(`https://www.youtube-nocookie.com/embed/${videoId}`)).toBe(videoId)
+    expect(parseYoutubeVideoId(`https://www.youtube.com/live/${videoId}?feature=share`)).toBe(
+      videoId,
+    )
+    expect(parseYoutubeVideoId(`https://www.youtube.com/shorts/${videoId}`)).toBe(videoId)
+    expect(parseYoutubeVideoId(`https://m.youtube.com/watch?v=${videoId}`)).toBe(videoId)
+  })
+
+  it('rejects non-YouTube hosts, unparseable values and non-video links', () => {
+    expect(parseYoutubeVideoId(`https://example.com/watch?v=${videoId}`)).toBeNull()
+    expect(parseYoutubeVideoId(`youtube.com/watch?v=${videoId}`)).toBeNull()
+    expect(parseYoutubeVideoId('https://www.youtube.com/playlist?list=PL123')).toBeNull()
+    expect(parseYoutubeVideoId('https://www.youtube.com/')).toBeNull()
+    expect(parseYoutubeVideoId('')).toBeNull()
+    expect(parseYoutubeVideoId(null)).toBeNull()
+  })
+
+  it('rejects an id that does not look like a YouTube id', () => {
+    expect(parseYoutubeVideoId('https://www.youtube.com/watch?v=ab')).toBeNull()
+    expect(parseYoutubeVideoId('https://www.youtube.com/watch?v=../etc/passwd')).toBeNull()
+  })
+})
diff --git a/tests/unit/speechVodResolver.unit.spec.ts b/tests/unit/speechVodResolver.unit.spec.ts
new file mode 100644
index 00000000..fb94711a
--- /dev/null
+++ b/tests/unit/speechVodResolver.unit.spec.ts
@@ -0,0 +1,146 @@
+// @vitest-environment node
+
+import { afterEach, describe, expect, it, vi } from 'vitest'
+
+import { resolveSpeechVod } from '@/utilities/speech/speechVodResolver'
+
+const VOD_PLAYBACK = 'https://vod.camara.leg.br/trecho.mp4'
+const VOD_DOWNLOAD = 'https://vod.camara.leg.br/trecho-download.mp4'
+
+const jsonResponse = (body: unknown, status = 200): Response =>
+  new Response(JSON.stringify(body), {
+    status,
+    headers: { 'content-type': 'application/json' },
+  })
+
+const mediaResponse = (status: number, contentType: string | null): Response =>
+  new Response(status >= 200 && status < 300 ? 'video-bytes' : '', {
+    status,
+    ...(contentType ? { headers: { 'content-type': contentType } } : {}),
+  })
+
+type FetchCall = { url: string; method: string; range: string | null; userAgent: string | null }
+
+const stubFetch = (
+  handler: (url: string, init?: RequestInit) => Promise<Response>,
+): FetchCall[] => {
+  const calls: FetchCall[] = []
+  vi.stubGlobal(
+    'fetch',
+    vi.fn(async (input: RequestInfo | URL, init?: RequestInit) => {
+      const headers = (init?.headers ?? {}) as Record<string, string>
+      calls.push({
+        url: String(input),
+        method: (init?.method ?? 'GET').toUpperCase(),
+        range: headers.Range ?? null,
+        userAgent: headers['User-Agent'] ?? null,
+      })
+      return handler(String(input), init)
+    }),
+  )
+  return calls
+}
+
+const readyBody = {
+  estado: 'PRONTO',
+  video: {
+    titulo: 'Trecho',
+    linkParaReproducao: VOD_PLAYBACK,
+    linkParaDownload: VOD_DOWNLOAD,
+  },
+}
+
+afterEach(() => {
+  vi.unstubAllGlobals()
+})
+
+describe('resolveSpeechVod', () => {
+  it('returns both URLs only after the ranged probes verify them', async () => {
+    const calls = stubFetch(async (url) =>
+      url.includes('video-sob-demanda') ? jsonResponse(readyBody) : mediaResponse(206, 'video/mp4'),
+    )
+
+    await expect(
+      resolveSpeechVod({ eventId: 67091, audioId: 558641, excerptTms: 3 }),
+    ).resolves.toEqual({
+      state: 'pronto',
+      playbackUrl: VOD_PLAYBACK,
+      downloadUrl: VOD_DOWNLOAD,
+    })
+
+    const probes = calls.filter((call) => call.method === 'GET' && call.range === 'bytes=0-1023')
+    expect(probes).toHaveLength(2)
+    expect(probes.every((probe) => probe.userAgent?.includes('Mozilla'))).toBe(true)
+    expect(calls[0]?.url).toContain('trecho=3')
+  })
+
+  it('drops a playback link whose probe fails but keeps the verified download', async () => {
+    stubFetch(async (url) =>
+      url.includes('video-sob-demanda')
+        ? jsonResponse(readyBody)
+        : url === VOD_PLAYBACK
+          ? mediaResponse(403, null)
+          : mediaResponse(206, 'video/mp4'),
+    )
+
+    await expect(resolveSpeechVod({ eventId: 1, audioId: 2, excerptTms: 3 })).resolves.toEqual({
+      state: 'pronto',
+      playbackUrl: null,
+      downloadUrl: VOD_DOWNLOAD,
+    })
+  })
+
+  it('rejects a link that answers 200 with an HTML error page', async () => {
+    stubFetch(async (url) =>
+      url.includes('video-sob-demanda')
+        ? jsonResponse(readyBody)
+        : mediaResponse(200, 'text/html; charset=utf-8'),
+    )
+
+    await expect(resolveSpeechVod({ eventId: 1, audioId: 2, excerptTms: 3 })).resolves.toEqual({
+      state: 'pronto',
+      playbackUrl: null,
+      downloadUrl: null,
+    })
+  })
+
+  it('falls back to HEAD when the ranged GET is refused', async () => {
+    stubFetch(async (url, init) => {
+      if (url.includes('video-sob-demanda')) return jsonResponse(readyBody)
+      return (init?.method ?? 'GET').toUpperCase() === 'GET'
+        ? mediaResponse(405, null)
+        : mediaResponse(200, 'video/mp4')
+    })
+
+    await expect(resolveSpeechVod({ eventId: 1, audioId: 2, excerptTms: 3 })).resolves.toEqual({
+      state: 'pronto',
+      playbackUrl: VOD_PLAYBACK,
+      downloadUrl: VOD_DOWNLOAD,
+    })
+  })
+
+  it('maps GERANDO and INDISPONIVEL without probing anything', async () => {
+    const calls = stubFetch(async () => jsonResponse({ estado: 'GERANDO', video: null }))
+    await expect(resolveSpeechVod({ eventId: 1, audioId: 2, excerptTms: 3 })).resolves.toEqual({
+      state: 'gerando',
+    })
+    expect(calls).toHaveLength(1)
+
+    vi.unstubAllGlobals()
+    stubFetch(async () => jsonResponse({ estado: 'INDISPONIVEL', video: null }))
+    await expect(resolveSpeechVod({ eventId: 1, audioId: 2, excerptTms: 3 })).resolves.toEqual({
+      state: 'indisponivel',
+    })
+  })
+
+  it('throws after the bounded retries when the Câmara is unreachable', async () => {
+    const calls = stubFetch(async () => {
+      throw new Error('connect ETIMEDOUT')
+    })
+
+    await expect(resolveSpeechVod({ eventId: 1, audioId: 2, excerptTms: 3 })).rejects.toThrow(
+      'connect ETIMEDOUT',
+    )
+    expect(calls).toHaveLength(2)
+  })
+})