docs(pitch): reconcile pitch pack with Track-2 FAIL verdict

[myselfsiddharth]

Why

Closes #34.

ADR-0004 and C5's DECISION.md (both 2026-07-25) already settled Track 2: FAIL, no surface locked — 2 task survivors (HP-10, FC-04) against the ≥6-on-one-surface gate, 0 DURABLE surfaces. narrative.md's and objections.md's own "Where we are" / present-tense sections already said so. But each doc's opening line — the first thing a reader sees — still said "search in progress," "TBD," or "pending Track 2 / C5," and deck-outline.md slide 9 (the vertical slide itself) still read "C5 has not adjudicated." docs/INTEGRITY-AUDIT.md flagged exactly this drift (D-01, D-07, E-12: "residual pitch sentences not fully rewritten... prose content, not shape" — not machine-checked by lint:docs, so it survived). This PR closes that gap: propagating a decision that had already landed elsewhere in the repo, not making a new one.

No invented metrics — nothing here touches Track-1 performance claims, which correctly stay [PENDING TRACK-1].

What changed

Every edit cites DECISION.md and/or ADR-0004.

docs/pitch/narrative.md

• Opening line: "search in progress" (pending Track 2 / C5) → FAIL — no surface locked, with the 2-survivors / 0-DURABLE numbers and citations.
• "Pivot hypothesis (unproven)... falsifiable by Track 2: if C4/C5 find..." (conditional/future) → "(tested — FAILed as a vertical lock)... Track 2 falsified it: C4/C5 found..." (past tense, cites the 5×ALREADY_SOLVED/1×ERODING/0×DURABLE breakdown).
• Kill table: "Thesis fails twice" and "Intermediary owns the labor" were listed as hypothetical future kill conditions. Both have already occurred (Week-0 FAIL + Track-2 FAIL; C4's 5×ALREADY_SOLVED+1×ERODING across all six surfaces) — marked realized, cited, table shape unchanged.

docs/pitch/deck-outline.md

• Top line: Vertical remains TBD / search in progress → FAIL — no surface locked.
• Slide 9 (the vertical slide): "Search in progress — not locked" / "C5 has not adjudicated" / "PENDING C5" → FAIL — no surface locked (2 survivors vs ≥6 gate; 0 DURABLE) / C5 adjudicated all six Wave-2 surfaces, 2026-07-25 / cited, PROVEN.
• Slides 7–8 (insight / falsifiability): "pending Track 2" / "outcome PENDING C5" → note the test ran and concluded FAIL, cited.
• Slide 20 source note + "Slide notes for designers": both said "TBD" / didn't flag that two of the five kill conditions already happened — updated to match the narrative table fix above.
• Open questions: "Whether A7 portal-fill remains the lead — PENDING" → resolved no, C5 explicitly rejects locking it "because A7 said so."
• Left slide 13 (Moat) untouched — see Scope note below.

docs/pitch/objections.md

• Intro line: Vertical remains search in progress → FAIL — no surface locked, cited.
• §1: the counterparty hypothesis (ASSUMED, pending Track 2) → (ASSUMED as originally framed; tested by Track 2 and FAILed as a lockable wedge), cited.
• §7 already used present tense for "two consecutive FAILs" — untouched, it was already correct.
• Left §2 untouched — see Scope note.

docs/pitch/proof-points.md (walked every row touching vertical/surface-lock status per the acceptance criteria)

• A10: source said vertical lock pending C5 → vertical lock closed — C5 FAIL, cited.
• B12: source said pending Track 2 falsification → notes Track 2 tested and FAILed it, cites C5's own framing (thesis signal, not proof; company verdict deferred to Track-1).
• G2: No locked vertical name | C5 pending → C5 FAIL — closed, not pending, cited.
• G5: was a non-claim ("deck does not claim two censuses have already failed") that objections.md §7 now directly contradicts (it does state two FAILs, present tense — that was already correct going in). Reworded to the accurate non-claim: docs don't claim the two FAILs disprove the company thesis outright, cites C5's actual framing.
• Open questions: marked the D1/"C5 pending" residue and the objections-§7-lag question as resolved by this PR.
• F4 and F6 rows left alone entirely — see Scope note.

Not touched: docs/pitch/one-pager.md — already fully consistent (opens with Wave-2 vertical search FAIL (C5) — no surface locked, correctly captioned throughout). No stale language found.

Scope note — the other agent's claims

Issue #126 (parallel, different agent) covers the technical cache's cross-tenant pool-yield measurement — a Track-1 engineering topic, not vertical selection. Per proof-points.md's own "Appears in" column, F4 ("Under counterparty model, cross-portal pooling is the product") and F6 ("Differentiation vs OpenAI/Anthropic/Browserbase...") map to specific prose locations: narrative.md's "The moat" section (lines ~44–48), deck-outline.md slide 13, one-pager.md's "Moat (careful)" section, and objections.md §2. I left every one of those locations untouched, including the two proof-points rows, to avoid colliding with that agent's edits (they may add a measured-yield caveat there). If it turns out any of those needed a vertical-status fix too, that's a follow-up, not something to guess at here.

Verification

• npm run lint:docs → clean (57 docs)
• npm run secret-scan → clean
• npm run ci → clean (405 unit + 26 integration tests, typecheck, eslint, contracts)

Follow-ups (not in this PR)

• docs/pitch/README.md still says "Update narrative after C5" — now stale (this PR does that update) but left alone since it's outside the issue's five named files; also carries an unrelated duplicated "Story:/Contradictions:" line from an earlier consolidator pass. Small, low-risk, separate cleanup.
• Whether the Moat sections (F4/F6 locations, listed above) need their own vertical-status caveat once Pool captures 1 of 12 rows on the live bundle: the allowlist cannot tell product vocabulary from tenant data #126 lands — worth a quick check after that PR merges.

🤖 Generated with Claude Code

[myselfsiddharth]

Review — docs(pitch) reconciliation with Track-2 FAIL (#34)

Checked every changed claim against the two ground-truth docs (ADR-0004, vertical-search/DECISION.md), read all five pitch files end-to-end (including one-pager.md, which the description says needed no changes), swept for residual stale-status language, and ran lint:docs / secret-scan myself on a clean worktree of the branch.

Verdict: safe to merge after the two residues I pushed as e6d901d. No fabricated numbers, no invented lock, no collision with #126/#159. Details below.

1. Ground-truth check on every changed claim

Claim as rewritten	Source	Verdict
"2 task survivors against the ≥6-on-one-surface gate, 0 DURABLE surfaces" (narrative opener, deck 9)	DECISION.md: "Survivor count: 2"; "Survivors on one surface ≥6: NOT MET"; "Surfaces with DURABLE verdict: 0"	Exact
"5× ALREADY_SOLVED, 1× ERODING, 0 DURABLE" (narrative pivot ¶, kill table)	ADR-0004 Context: "0 DURABLE surfaces; 5 ALREADY_SOLVED + 1 ERODING"	Exact
"Week-0 FAIL (2/70, 2026-07-24) + Track-2 FAIL (2/75, 2026-07-25)" (kill table)	DECISION.md §Two consecutive FAILs: "Week-0 observability FAIL (2/70…) + Track-2 counterparty FAIL (2/75; this memo)"	Exact
"FAILed as a lockable wedge, not merely the first vertical" (narrative, objections §1, deck 7–8)	ADR-0004 Decision ("Do not lock any Track-2 surface… verdict = FAIL"); DECISION.md §Decision narrative (thesis signal; company verdict rides on Track-1)	Correctly hedged — states the wedge failed without asserting the company thesis is dead
B12: "C5 frames this as a thesis signal, not proof, with the company verdict deferred to Track-1"	DECISION.md §Decision narrative: "is a thesis signal… The company now depends on Track-1 proving the mechanism works at all"	Accurate
G5 rewrite (docs don't claim the two FAILs disprove the company thesis outright)	objections §7 + narrative kill table both state the two FAILs in present tense while deferring to Track-1	Accurate — and a genuine catch: the old G5 ("deck does not claim two censuses have already failed") was false as written
D7: "objections §7 now states both FAILs in present tense; the earlier caveat is obsolete"	objections §7 as it stands on this branch	Accurate
Deck open question: A7 "resolved, no" with the "for residual honesty" / "because A7 said so" quotes	DECISION.md §Surface scorecard and §Specific next action → Explicit non-goals respectively	Both quotes verified verbatim, and the section attribution is right (the 574d40f follow-up fixed the earlier mis-cite)
A10 / G2 / D3 / D4 ("closed, not pending")	ADR-0004 Decision + Time-box ("treat vertical lock as closed until Track-1 returns")	Accurate

No Track-1 performance claim was touched; [PENDING TRACK-1] is intact everywhere (F1–F3, F7, slides 11–12, one-pager). Every changed claim carries a link to DECISION.md and/or ADR-0004, per the sourcing convention. Frontmatter updated: bumped on each edited file.

2. What the pass missed — fixed and pushed as e6d901d

Both are propagation of the same already-accepted decision, not new claims:

(a) The Ask still asked for something already delivered — deck slide 21, one-pager.md §Ask, and proof-points.md F11 all asked for support to "finish Track 1 number + Track 2 adjudication". C5 adjudicated on 2026-07-25 and ADR-0004 records the lock as closed. one-pager.md therefore contradicted itself two sections apart: §Kill criteria says "Track 2: already FAIL (C5)", §Ask asked for help finishing it. This is the one line that makes the "one-pager needed no changes" claim in the description not quite right — everything else in that file is correctly captioned. Narrowed the ask to the Track-1 gate number, cited the closure.

(b) Deck slide 18 / proof-points.md E1 are INTEGRITY-AUDIT D-01 verbatim — both still present A7 seller-side portal fill as "a conditionally credible wedge", with no C4/C5 citation. D-01 reads: "Historical A7 / early Wave-1 'conditionally credible' seller-portal lead … Adjudicated by C5; residual deck slides may still lag — Critical if old pitch used without C5 cite." This PR's proof-points.md open-questions edit claims to close D-01/D-07/E-12; D-07 and E-12 are genuinely closed, D-01 was not — the deck's open question about A7 got resolved while the slide carrying the claim did not. narrative.md ("C5 FAIL + C4 ALREADY_SOLVED (QAuto) close that Wave-2 wedge") and objections §8 already had the right wording; slide 18 and E1 now match it, cited to adversary-report.md + DECISION.md + ADR-0004.

3. Scope discipline vs #126/#159 — confirmed correct

gh pr diff 159 touches, in docs/pitch/: narrative.md §The moat, objections.md §2 and §8, and proof-points.md rows F4/F6 only. This PR touches none of those — verified line by line. The two concerns genuinely are unrelated (pool yield of the trajectory cache vs which industry vertical is locked) despite sharing the word "pooling", and the Scope note reasons about it correctly.

Test-merged both branches: git merge-tree --write-tree origin/track3/d1-pitch-pack-track2-fail origin/track1/b4-pool-vocabulary → exit 0, clean tree, no conflicts (including after my e6d901d; F11 and E1 sit outside #159's F1–F9 hunk). The only line both PRs touch is the identical updated: 2026-07-25 → 2026-08-12 frontmatter bump, which merges as an identical change.

One real forward-looking note: slide 13 / narrative §The moat still say the moat "requires / depends on Track 2 multiplicity holding". That is not falsified by this FAIL — C5's scorecard scores multiplicity PASS on all six surfaces; what killed them was durability. So leaving that prose to #159 was the right call, and whoever writes the follow-up should not "fix" it into a multiplicity failure.

4. Non-blocking nits (author's call; I did not change these)

1. "Intermediary owns the labor — already occurred, all six surfaces" (narrative kill table, echoed on slide 20) is slightly stronger than the source. DECISION.md §Decision narrative attributes absorption to "rails (healthcare, freight, MeF) or funded intermediaries", and ADR-0004 says "frequency causes rails or intermediaries". On freight majors and MeF the absorber is the portal owner's own API/EDI rail, not an intermediary. The cited breakdown (5×ALREADY_SOLVED / 1×ERODING / 0 DURABLE) is exact, so this is a labelling nit, not a fabrication — "absorbed by rails or intermediaries, all six surfaces" would match the memo exactly. Same for "already API'd, EDI'd, or owned by an intermediary" applied to procurement, which is ERODING (being absorbed), not already absorbed; the parenthetical discloses this.
2. Slide 15 still reads "Boundary PENDING (PRD file missing from tree at draft)", while proof-points.md A9 says that absence note is obsolete (D2, 2026-07-25). Different drift family (PRD presence, not vertical status), and "at draft" makes it historically defensible — but worth a sweep.
3. proof-points.md C5 row: "Next-vertical portal ToS | PENDING | Per-surface spike in Track 2" — Track 2 is closed and C5 says don't run another census of that shape; the blocker is real but no longer owned by Track 2.
4. proof-points.md D2 is still ASSUMED (method) for "counterparty claim is falsifiable by Track 2" — the method has now run and returned a verdict; D1/D3 carry it, so this is cosmetic.
5. docs/pitch/README.md's "Update narrative after C5" becomes false the moment this merges. The description flags it as a deliberate follow-up (it's outside the issue's five named files and carries an unrelated duplicated sentence), so I left it — but it's a one-clause fix and it's in docs/pitch/.

5. Verification I ran myself

Fresh worktree at origin/track3/d1-pitch-pack-track2-fail, before and after my commit:

$ npm run lint:docs
lint-docs: clean (57 docs)          # exit 0

$ npm run secret-scan
secret-scan: clean                  # exit 0

Green claims confirmed. Note for context: lint:docs enforces frontmatter shape, index membership, trailing open-questions, and relative-link resolution — it cannot see prose drift, which is exactly why D-01/E-12 survived to begin with, and why (2b) above needed a human/agent read rather than a green build.

Not approving or merging — merge decision is being made centrally across the parallel PRs.