── SPEC COMPONENT 01 ──────────────────────────────────────── TYPE: Core Function COMPONENT: YouTube Transcript Extraction DETAIL: Accept a YouTube URL, validate it, extract video ID, fetch the transcript/captions as the source text for summarization. CONFIDENCE: 90% BASIS: 2 signals (URL input + summary output both stated) BUILDER ACTION: In Bolt, build a server-side function (Node) using a free transcript scraper approach (e.g. `youtube-transcript` npm package — uses YouTube's public timedtext endpoint, no API key). Do NOT use YouTube Data API v3 for transcripts (it doesn't expose them) or paid transcript APIs. ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 02 ──────────────────────────────────────── TYPE: Core Function COMPONENT: Summarization Engine DETAIL: Convert transcript into (a) short summary, (b) key points list, (c) timestamps mapped to each key point. CONFIDENCE: 60% BASIS: 1 signal — "summary with key points and timestamps" stated, but the engine is unspecified BUILDER ACTION: FLAGGED AS ASSUMPTION (see Log A01). If "no paid APIs" is strict: implement extractive summarization client-side (TextRank via a JS library) — lower quality. If a free-tier LLM is acceptable (Gemini free tier, local model): server-side call with transcript chunking. Ask Monna's end-user which path before wiring. ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 03 ──────────────────────────────────────── TYPE: User Flow COMPONENT: Primary Flow: Paste → Process → Read DETAIL: User pastes URL → client validates format → server fetches transcript → engine produces summary + keyed timestamps → result rendered with clickable timestamps linking to `youtube.com/watch?v=ID&t=Xs`. CONFIDENCE: 85% BASIS: 2 signals (input and output both defined in RAW_IDEA) BUILDER ACTION: Build as single-page app in Bolt: input field + submit, loading state, results panel. Make each timestamp an anchor link opening the video at that second in a new tab. ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 04 ──────────────────────────────────────── TYPE: Data Model COMPONENT: Result Object DETAIL: `{ videoId, title, summary: string, keyPoints: [{ text, timestampSeconds, timestampLabel }] }`. No persistence layer specified — treat results as ephemeral, in-memory only. CONFIDENCE: 70% BASIS: 1 signal (output structure implied by RAW_IDEA) BUILDER ACTION: Define this as a TypeScript interface in Bolt. Do not add a database — out of scope (see Fence). ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 05 ──────────────────────────────────────── TYPE: User Roles COMPONENT: Single Anonymous User DETAIL: No auth, no accounts, no roles mentioned. One user type: anonymous visitor. CONFIDENCE: 80% BASIS: 1 signal (absence of any role language in RAW_IDEA) BUILDER ACTION: Skip all auth scaffolding in Bolt. No login UI. ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 06 ──────────────────────────────────────── TYPE: Edge Case COMPONENT: No Captions Available DETAIL: Video has captions disabled, is auto-caption-only in a non-supported language, or is age-restricted/private. Transcript fetch returns empty or error. CONFIDENCE: 95% BASIS: known platform behavior + transcript-dependency in Component 01 BUILDER ACTION: Server returns typed error `NO_TRANSCRIPT`; UI shows "This video has no available captions — summary can't be generated." Do not attempt audio transcription (would require paid/heavy API — violates CONSTRAINTS). ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 07 ──────────────────────────────────────── TYPE: Edge Case COMPONENT: Invalid / Non-Video URLs DETAIL: Playlist URLs, channel URLs, Shorts URLs, malformed strings, non-YouTube domains. CONFIDENCE: 90% BASIS: standard input validation requirement implied by "paste a URL" BUILDER ACTION: Client-side regex validation for `watch?v=`, `youtu.be/`, and `/shorts/` patterns. Shorts → extract ID and process normally. Playlists/channels → inline error "Paste a single video link." ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 08 ──────────────────────────────────────── TYPE: Edge Case COMPONENT: Long Videos (2hr+ transcripts) DETAIL: Transcripts can exceed processing limits (token limits if LLM path, performance if client-side TextRank). CONFIDENCE: 75% BASIS: inferred from Component 02 architecture BUILDER ACTION: Chunk transcript into ~10-minute segments, summarize per chunk, merge. Cap at a stated max (e.g. 3 hours) with a friendly error beyond it. ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 09 ──────────────────────────────────────── TYPE: Edge Case COMPONENT: Rate Limiting / Scraper Fragility DETAIL: Free transcript endpoints are unofficial; YouTube can throttle or change them. Concurrent requests from one deployment IP may get blocked. CONFIDENCE: 85% BASIS: known behavior of unofficial transcript libraries BUILDER ACTION: Add retry-with-backoff (max 2 retries), surface a `TEMPORARILY_UNAVAILABLE` error state, and add a 1-request-at-a-time client-side queue. Note in README that this is the fragile dependency of the build. ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 10 ──────────────────────────────────────── TYPE: Edge Case COMPONENT: Non-English Videos DETAIL: Transcript may be in any language; summary language behavior is undefined in RAW_IDEA. CONFIDENCE: 50% BASIS: inferred — no signal in input BUILDER ACTION: FLAGGED AS ASSUMPTION (see Log A03). Default: summarize in the transcript's language. Defer translation feature. ───────────────────────────────────────────────────────────── ── SPEC COMPONENT 11 ──────────────────────────────────────── TYPE: Scope Item COMPONENT: Timestamp Precision DETAIL: Timestamps come from caption segment start times — accurate to the segment (~2–7s granularity), not word-level. CONFIDENCE: 90% BASIS: caption data structure (known) BUILDER ACTION: Map each key point to the start time of the caption segment where its source content begins. Display as `MM:SS` (or `H:MM:SS` over 1hr). ───────────────────────────────────────────────────────────── ── ASSUMPTIONS LOG ────────────────────────────────────────── ASSUMPTION 01: Summarization engine unspecified → Assumed free-tier LLM or client-side extractive (decision gate, two paths specified) → Confidence: 60% ASSUMPTION 02: No persistence mentioned → Assumed ephemeral results, no history/saving → Confidence: 70% ASSUMPTION 03: Output language unspecified → Assumed same language as transcript → Confidence: 50% ASSUMPTION 04: "Key points" count unspecified → Assumed 5–8 points per video, scaling with length → Confidence: 55% ASSUMPTION 05: No auth mentioned → Assumed fully anonymous tool → Confidence: 80% ───────────────────────────────────────────────────────────── ── SCOPE FENCE ────────────────────────────────────────────── IN: URL validation / transcript extraction / summary + key points + clickable timestamps / no-caption + invalid-URL + long-video error states / single-page responsive UI OUT: audio transcription (Whisper etc.) / user accounts & saved history / playlist or channel batch processing / translation / browser extension / export (PDF/Notion) BUILDER NOTE: Bolt handles the frontend + Node server functions natively. The unofficial transcript dependency is the single point of failure — Bolt's WebContainer environment may also have CORS/network restrictions on the timedtext endpoint, so test transcript fetch FIRST before building UI. If it fails in Bolt's runtime, the fallback is deploying the fetch function to a free serverless target (Cloudflare Workers free tier — fits CONSTRAINTS). ───────────────────────────────────────────────────────────── ◆ SIGNAL SCREEN: 11 components produced. 3 below 70% threshold — all flagged as assumptions (Components 02, 10, and Assumption 04), not stated as fact. 1 critical decision gate: summarization engine path (A01) — that one needs an answer before the build starts, everything else can proceed.