Session C13 — Target-scope re-baseline (2026-08-19)

Branch: docs/target-scope-rebaseline · Spec: roadmap C13 (owner review 2026-08-19 of the C12 pages)

Goal and outcome

Two documents, two clean roles. product/scope.md stops carrying implementation truth entirely — status table, evidence column and all 37 per-item status blockquotes are gone, and it reads as what it is: a description of Gemini Notebook’s functionality. product/target-scope.md is now the single place status lives, re-baselined against the v1 target instead of against the research catalog. After the re-baseline all 18 rows are ✅: every row’s target is what its roadmap session committed to, and every one of those commitments shipped. The page reads like a delivered v1 because it is one. Docs-only; no product code touched.

`product/scope.md` — removal, not redecoration

Status table dropped entirely (not reduced to an id + title index). With Status and Evidence gone the remainder would have duplicated the section headings verbatim while preserving the visual shape of the removed machinery — an invitation to re-populate it. The headings and the docs page’s own structure already provide the id → item index.
All 37 per-item status blockquotes removed, including the ❌/⏳ ones and their “Deferral (2026-08-18)” lines. Adopted research text is byte-identical: every removed line in git diff product/scope.md is a blockquote, a table row, or the status section’s own header/legend (verified by diffing with those patterns excluded — the residue is only the “## Implementation status” heading and its legend paragraph).
Kept: the header role/framing blockquotes (updated where they described the now-removed machinery), the §8 annotation (the one place the reading-order clustering is explained), and NF-15’s C12 pointer line to in-numbers.md (a cost-transparency pointer, not a status).
Added: a dated C13 addition to the role blockquote recording that this document no longer tracks status and where status now lives — future readers meet the removal as a decision, not a gap.
The §7 non-functional blockquote was rewritten: it used to explain the badge-tracking exception for NF-15. It now just names where the non-functional record lives (security register, feasibility register, architecture views, plus target-scope for quotas and guards).

Badge-only facts, relocated before deletion

Everything else the badges said was already recorded in target-scope.md, the roadmap cut list, or a handover. These were the facts that lived only in a badge, and where each went:

Fact	Relocated to
Auth: no SSO/OAuth, account deletion or plan association	SF-01 row’s target statement (stated as the target’s boundary)
Auth was built without an MVP marker, as the prerequisite for user-scoping	already in target-scope’s “What Marginalia v1 is”; not duplicated
Source viewer is dialog-only / not URL-addressable	CF-04 row’s known-limitation cell, linked to the A5 handover’s open items
SF-02: no “shared” notebooks in v1	SF-02 row’s target statement
SF-10: audio output language is German/English	SF-10 row’s target statement
SF-11: the full guard + quota numbers (20 MB, 200k words, 50 sources, 20 artifacts, 1 concurrent; 20 notebooks, 50 messages/day, 10 overviews/day)	SF-11 row’s target statement — previously target-scope pointed at the SF-11 badge for these
NF-15’s open cost controls (token budgets, model routing, caching, storage quotas)	target-scope’s operational-envelope paragraph
SF-08: signed-URL artifact access, no share/public links	SF-08 row (target + limitation)
SF-09: in-process by decision, D-2 stage 1	SF-09 row’s target statement

Nothing was destroyed. Facts already on the cut list (CF-02 formats, CF-04 Source Guide, CF-08 server-side stop/regenerate, CF-09, CF-10 note→source, SF-09 queued jobs, two-speaker audio, sharing, analytics) stayed there; SF-14’s “never scheduled, conscious gap” verdict stays in the C10 handover.

`product/target-scope.md` — re-baselined against the target

Semantics stated in the intro: status measures delivery against the v1 target defined in each row; the research catalog is descriptive and carries no status at all. Narrower-than-research scope is part of the target, stated in the target column, and does not degrade status.
New “What v1 targets” column giving each commitment in the session record’s words, with the roadmap session named. The research id moved into the capability cell as an inline link — it is traceability, not a first-class column. All 16 anchors re-verified against the rebuilt scope page’s actual heading ids.
Seven rows renamed so the name is the target’s, not the catalog’s: Notebook library / home screen → Notebook library · Source processing and knowledge indexing → Source processing and retrieval index · Source selection / context control → Source selection for retrieval · Conversation state → Chat history (persistent, multi-turn) · Notes → Notes, including saved chat answers · Artifact download and management → Audio artifact management · Usage limits and quotas → Usage guards and per-user quotas.
Ten statuses changed 🔶 → ✅ (SF-01, CF-02, CF-04, CF-08, CF-10, CF-12, SF-08, SF-09, SF-10, SF-11 — every 🔶 C12 carried), each because the limitation that produced the 🔶 sits outside the target: it is either on the roadmap cut list or was never part of the session’s goal. Reasoning per row is in the PR body.
Zero “badge” cells. Every known-limitation cell now states the limitation in a clause and links the real record (A3/A4/A5/A6/D2 handovers, SEC-7 in the security register).
Operational envelope no longer points at the deleted SF-11 badge for the numbers; it points at the row above, plus in-numbers and SEC-7/-10.

Two ambiguous targets — readings and why

Audio-transcript persistence (CF-12). Reading taken: outside D2’s target → ✅ with the gap stated. D2’s roadmap goal is “script generation from selected sources → TTS → artifact in Storage; Studio tile + async job status + player” — it never names persisting the script, and the D2 handover files the transcript under open questions / next sessions, calling it “a small follow-up”. The competing reading is that NF-11 lists transcript availability for audio, so an accessibility requirement implies it; rejected because NF sections are research requirements and v1 never committed to WCAG 2.2 AA. The gap is still visible: it is the row’s known limitation.
Request-rate limiting (SF-11). Reading taken: outside A6’s target → ✅ with the qualification stated. A6’s goal is “simple per-user quotas (NF-15 minimum)”; rate limiting appears in the record only as SEC-7, an accepted risk whose hardening trigger is “before public exposure”, conditional on the closed signup circle. The competing reading — that abuse limits are one capability, so an unlimited request rate is a hole inside the target — is rejected on the session record’s wording, but SEC-7 is linked from the row.

A third judgement worth flagging to the reviewer: A6 documented that regenerations are not counted against the daily audio quota. That limitation does sit inside the target (the quota is the commitment), so it was the strongest 🔶 candidate on the page. It stays ✅ because the targeted quotas shipped and are enforced, and the uncounted path remains bounded by the 1-concurrent and 20-per-notebook caps, so the cost-control purpose holds. It is stated in the row’s limitation cell either way. If the foreman disagrees, SF-11 is the one row to flip.

Requirement-9 repoints

Inside target-scope.md (the only live document that referenced the removed machinery): the operational envelope’s “concrete numbers live in the SF-11 status badge” link, the header’s “tracked with the same machinery … the research document’s badges” framing, and the exclusions section’s “with per-item status”.
Swept, nothing to repoint: product/in-numbers.md, product/security.md, product/faq.md, product/architecture/*.md. None of them links a status badge or the status table. security.md’s scope-doc anchors (NF-07…NF-17) and SEC-7’s SF-11 reference still resolve — headings are unchanged. architecture/index.md and logical.md cite scope.md as the ideal model, which is still true of a research catalog.
Flagged, not edited (apps/ is out of this session’s boundary):** two docs-app strings now describe the research catalog wrongly. apps/docs/src/pages/product/scope.astro:13 — description says “with per-item implementation status”, which is now false. And apps/docs/src/pages/index.astro:10 — the Start-page card still says “sorted into build phases. Phase 1 — the grounded-chat core loop — is the current target”, which C12 already demoted. Both are one-line string swaps for whichever session may touch apps/docs.

Verified

apps/docs bun run build from the worktree: 55 pages before this handover existed, 56 with it, exit 0 both times. Both rebuilt pages inspected as rendered HTML and as screenshots.
Greps over source and built HTML of both pages: zero bare-word “badge”/“badges”; zero ⏳ (and zero ✅/🔶/❌) in scope.md; zero implementation-status blockquotes in scope.md; zero “cluster n”; every remaining “§” sits inside link text.
Link integrity on the rebuilt target page: all 16 scope.md#… anchors match real heading ids in the rebuilt scope page; all five handover links resolve to built /sessions/… pages; roadmap.md → /roadmap/ and feasibility.md → /decisions/ resolve.
Repo root from the worktree: bun test 154 pass, 0 fail (the known local flake did not occur); bunx varlock run -- bun run build exit 0.
bun install in the fresh worktree left bun.lock unchanged.

Hot files

None. No new dependencies. bun.lock, root package.json, root AGENTS.md untouched — AGENTS.md’s product-docs block already describes scope.md as “descriptive; its groupings carry no plan meaning”, which stays true and now also matches the absence of status.

Open items / next sessions

The two flagged apps/docs strings above (foreman call: a one-line pointer session, or fold into the next docs-app touch).
The C12 handover’s note that “the target table mirrors the research table’s statuses by definition” is superseded by this session. Per correct-the-record, a dated correction blockquote was appended to that handover’s open-items entry (annotated, not rewritten); the update discipline is now single-file — target-scope.md only.
After merge: foreman dispatches deploy-static-sites.