Marketing rots. Ours is internally verified and publicly graded. Every material claim we make in public sits in this ledger with an honesty grade and a dated receipt — and the standing rule is when in doubt, claim down. A wording change on any page and its row here move together, in one reviewed change, or neither ships. Two limits on the rows themselves: a row states what a limitation means for a client, never the condition or coverage gap that would let someone exploit it — and a row never restates a figure the registry blocks.
Structural. Holds without anyone's good intentions — code, gates, or contracts make the claim true.
Read from a primary source — production database, signed contract, bank ledger — with the measurement date shown. A dated number, not a live counter.
Real, with a named gap. The claim holds on the main path; the edge that doesn't is stated, not hidden.
We say it, we believe it, the receipt isn't attached yet. The weakest grade we allow ourselves in public — and it says so.
A forward promise about how we will behave, with the mechanism that enforces it named. Graded on structure, not history.
A claim we used to make and stopped — because the receipt didn't reproduce, or a stricter rule caught it. Retirements stay listed. That's the point.
A positioning or identity claim the founder has ruled true on the record. Judgment, owned by a name — not a measurement.
| Claim, as made | Where | Grade | Receipt / the honest edge |
|---|---|---|---|
| Isolation, two-layer wording: engagement BUILT in isolated hardened cells (Layer A, real VMs); running product's data = enforced request scoping, graded separately (Layer B, PARTIAL) | system.html commitment 1 + isolation-cells card; claims.html row 6 | Layer A REAL / Layer B PARTIAL [REGISTR… | 2026-07-17 rewrite per method PR #45 ("the distinction that keeps us honest"): prior copy ("your code, credentials, AND DATA never share a blast radius"; "your product never shares a machine") implied physical per-tenant runtime isolation — the exact overclaim C-13 blocks. Now capped: MAY say built-in-isolated-cells; MUST NOT imply the live product runs in its own VM. History: 1C buyer-line rewording (2026-07-10) was itself superseded by this newer ruling — corrections are the methodology. Upgrade when Layer B's fast-path fix carries a prod receipt.ledger row 1 |
| "The developer supply chain is treated as hostile. Always." / "Known-malicious packages hard-fail our CI before they can ever install; untrusted code and dependencies are opened first in disposable intake cells, never on a machine that matters; and when something gets through anyway, the response is a documented playbook we've executed on real compromises." | system.html:107–108; echoed index.html:98 ("quarantined") | PARTIAL → ASPIRATIONAL [PR#6 — OPEN] | 2026-07-10 copy-truthing (ruling 1A): copy rewritten down to the graded reality — known-IOC hard-fail in CI (real), disposable intake cells (real practice), executed IR playbook (real). The retired sentence claimed quarantined-intake-as-pipeline + provenance, which have no code receipt. Coverage is being widened under a founder-approved change; grade moves when it lands. Flag RESOLVED.ledger row 2 |
| "Every governed action is logged and attributed to the engagement it served" | system.html:117 | HOLDS-WITH-WORDING-CHANGE [PR#2 — MERGED] | The engagement-level attribution wording IS the M2 fix, already applied in this seed. "Logged" LIVE (governed_actions, auth_events); attribution is org/engagement-level only — no operator identity on audit writes yet. Site issue #6 (REVERT-WHEN-SHIPPED) tracks upgrading to per-operator wording when the token overhaul threads userId into audit writes.ledger row 3 |
| "Safety fails closed." / "blocked before they run — structurally, by enforcement that works even when our own systems are down" | system.html:125–126 | ENFORCED (with documented edges) [PR#6 … | Strongest commitment: local catastrophic denylist blocks with zero network; loop actors fully fail-closed; 5 subagent-bypass layers wired; halting monitors booted. Edge: outside that floor, general checks open during an outage of our own systems, by documented design — footnoted at client altitude, without the operational specifics.ledger row 4 |
| Structural command safety card — catastrophic floor blocks with zero network dependency; "the one edge we'll name before you ask" | system.html structural-safety card | ENFORCED | 2026-07-17: "Fail closed, always." replaced with the edge-honest wording (general checks open during our own API outage, by documented design) — the row-4 footnote the audit asked for. Flag RESOLVED.ledger row 5 |
| "The fail-closed monitors halt and page on any unattributable or non-compliant run. They already have: an early credential-wiring issue was caught exactly as designed." | system.html:205 | ENFORCED [PR#6 — OPEN] | LoopComplianceMonitor genuinely halts (loop_policies.enabled=false); LoopRun.apiKeyId NULL = unattributable → halt + page (invariant I6). The "already have" anecdote matches the audit's credential-wiring catch.ledger row 6 |
| "When something goes wrong, the response is proven." / "Incidents are handled with a documented discipline, backed by live halt-and-page monitoring:" | system.html:134–135; meta description system.html:7 ("documented incident response") | PARTIAL → ASPIRATIONAL [PR#6 — OPEN] | 2026-07-10 copy-truthing (ruling 1B, per R2b/R3): "rehearsed" dropped until a drill cadence exists — "proven" rests on the real 2026-04-28 drill + executed real compromises; "documented… live halt-and-page" matches the audit's exact safe-to-claim wording. Meta description updated in the same PR (same claim, missed by the drop). Flag RESOLVED. UPDATE 2026-07-16: a monthly rollback-drill.yml cron now exists (method deploy-gate.md:111, incident-response.md:60) — the audit's "no scheduled drill" line is stale in the good direction; "rehearsed" may return once the cadence has cycles on record.ledger row 7 |
| "324 autonomous PR reviews … every one logged and acted on, measured from the production database" (stat tiles + prose) | system.html:216; index.html:165; proof.html:255; investors.html:127,129,271 | MEASURED — as of 2026-07-09 | Replaces the retired 491/95.7% (method PR #4, MERGED: no prod table or sum reproduces it). Receipt: operator-run read-only SQL against Railway prod 2026-07-09 — autonomous_pr_reviews = 324, 100% acted (method repo 01-method/gate-3-build.md:42). ROT WATCH: reviews accrue continuously on the 10-min tick, so this number is a dated measurement, not a live counter — re-run SELECT COUNT(*) FROM autonomous_pr_reviews; before quoting it anywhere new, and re-verify "10 repositories / 7 operators" (SELECT COUNT(DISTINCT "repoFullName") FROM autonomous_pr_reviews;) the same way. Resolves site issue #9.ledger row 8 |
| "Every acceptance criterion advances through a chain of delivery stages, each backed by a linked commit, deploy, or test result." | system.html:117 | ENFORCED [PR#6 — OPEN, D3 row] | AC evidence is structured with commit/deploy links and auto-recorded (platform schema + delivery-loop code, cited internally). The *four evidence tiers* language, if ever added to the site, is ASPIRATIONAL — no coded artifact.ledger row 9 |
| Founder monthly-revenue totals (registry C-10 class) — RETIRED FROM SITE | formerly founding-five.html hero; claims.html retired row | RETIRED — RESOLVED by Travis merging PR… | Conflict history: published at ASSERTED on explicit "send it"; upstream registry C-10 blocks revenue totals; Travis ruled by merging the fallback same day — hero now carries the registry-cleared measured claims (rows 19–20). Returns as MEASURED if/when a books receipt lands (registry C-10 pending-upgrade). The LinkedIn post is Travis's own surface and is not governed by this ledger.ledger row 10 |
| "We seal predictions before outcomes and publish the score — wins and misses" + PRD-001 = 7 hit / 4 partial / 2 miss, both misses published; PRD-002 sealed | ledger.html; claims.html row 11 | MEASURED | Method repo 05-predictions/ (append-only ledger, founder-scored). Public page carries NO client names — de-identified per issue #4 gate; scores and mechanism only.ledger row 11 |
| Founding Five vested structure: "half my fee only exists if you win; your outcome publishes on the ledger either way" + $7,500 entry / success fee = 10% of collections capped at $7,500 (terms revised 2026-07-23, see row 61) | founding-five.html term sheet; ledger.html rows FF-001…FF-005 | COMMITMENT | Forward promise graded on structure: ledger rows pre-reserved in public before any signature; success fee contractually conditional. Moves to CONTRACTED per-founder as SOWs exist.ledger row 12 |
| claims.html itself ("every claim on this site, graded") | claims.html | ENFORCED (by convention) | claims.html is the rendered public face of THIS file; the binding convention in the header (page edit + ledger row move together or neither merges) is what makes the page's own claim true. claims.html and CLAIMS.md must not drift — any PR touching one touches both.ledger row 13 |
| Deploy-gate specifics: SHA-match wait, one global concurrency lock, six smoke checks (two prod-fatal), halt-and-page on failure or cancellation, rollback = re-promote through the same gate | system.html "The deploy gate, specifically" card | ENFORCED / VERIFIED | Method repo 02-runbooks/deploy-gate.md:22-59, graded against the live promote workflow; verified 2026-07-16 session.ledger row 14 |
| "Zero failures were governance blocks — every failure was access, credentials, or infrastructure" | proof.html receipts lede | MEASURED — as of 2026-07-09 | Same prod measurement as the 324 stat: failure classification across pr_review_jobs (14/106) and loop_runs lanes shows no tier/verifier/ledger/breaker trips (method 01-method/gate-3-build.md:44-60). Client repo names in the source table stay internal.ledger row 15 |
| "It was never the AI" + three standing rules (ship-suppressed-first; auditor held stricter; a standard remembered has already failed) + "when we doubt, we say the doubt — then price the test" | system.html doctrine block; pricing.html Prove tier | COMMITMENT (founder-ruled) | Founder-confirmed judgment rules J6, J11, J10, J8, J7 (method repo 03-judgment/judgment-library.md, ruled entries). Behavioral doctrine graded on the ruled library, not on enforcement code; J11's suppressed-first shipping is additionally wired in practice (audit D7).ledger row 16 |
| Hero: "You built something real with Claude. We'll take it the rest of the way — and I stand behind it in production." (voice-split ruled via meaning lane, method PR #81 / C-01 updated) + subhead + CTA | index.html hero | FOUNDER-CONFIRMED [REGISTRY C-01/C-08] | Positioning ruled by Travis 2026-07-16/17 (method 05-market/positioning-and-pricing.md §0, "the source copy the site renders").ledger row 17 |
| "The last mile the platforms won't run" thesis (Anthropic makes the prototype easier; the production gap doesn't shrink) | index.html why-now | FOUNDER-CONFIRMED (thesis) [REGISTRY C-03] | Ruled positioning; the client's "leave the last mile to the little folks" quote stays OFF the site (client-quote tier, C-11).ledger row 18 |
| Proof strip: 12/12 delivered milestones ≤ $8,375 · 5/5 clients entered four-figure · 0/3 biggest cold asks closed | index.html stat row; pricing.html lede | MEASURED [REGISTRY C-04/C-05/C-06] | HubSpot, 20 deals (method 05-market/scoreboard-reconciliation.md §2). Framed as MILESTONES, never "12 clients" — the reconciliation explicitly forbids that framing. No client names or per-client dollars on-page.ledger row 19 |
| "One client expanded a self-built prototype into $23.5K of production milestones — from a single $7.5K proving milestone" | proof.html receipts lede (moved off homepage 2026-07-21, CMO ruling: ratios stay above the fold, dollar arcs live with the receipts) | MEASURED (anonymized) [REGISTRY C-07] | HubSpot-measured (reconciliation §1/§3). NOTE: supersedes a founder-flagged figure that is BLOCKED under registry C-09 and must never appear — including here.ledger row 20 |
| PRD-003 sealed row | ledger.html | MEASURED (ledger state) | Method 05-predictions/PREDICTION-003 merged 2026-07-17, sealed with PRD-001 calibration applied. De-identified on-page.ledger row 21 |
| Legacy scale claims softened to measured shapes: "10+ products / five industries" → "twelve delivered milestones across five clients" (hero-note, about ×2) and "20 deals on the ledger" (home stat) | index.html; about.html | MEASURED [REGISTRY] | 2026-07-17 conformance sweep: the 10+/five-industries framing had no registry receipt; replaced with deal-ledger shapes (20 deals, 12 milestones, 5 clients — reconciliation §2). Biographical claims on about.html (30 yrs, 800+ engineers, $300M, $28B, 99→99.99%) are resume-tier, outside registry scope — unchanged, flagged here for the record. "Hundreds of runs" (system.html) is supportable: 449 measured runs across lanes as of 2026-07-09.ledger row 22 |
| About story: "learned from clients" (radiologist / nurse / insurance agency owner, anonymized) + "Engagement 000 is us" + "no platform software until a design partner signs" | about.html prose | FOUNDER-APPROVED narrative | Learned-from-clients = founder's own public account (his LinkedIn surface), professions only, no names (C-11 respected). Engagement 000 = plan of record, founder-approved 2026-07-10 (method 04-dogfood/ENGAGEMENT-000-productlove.md:1,50,53,56); the no-build-before-signature ruling quoted directly.ledger row 23 |
| "Another, who built his prototype in Claude, has expanded it into $14K+ of delivered milestones, with a larger phase in progress" | proof.html receipts lede (moved off homepage 2026-07-21, same ruling) | CONTRACTED (anonymized) [REGISTRY C-09 ✅] | Registry row resolved 2026-07-17: the won amount supports the published "$14K+" framing; the expansion figure is C-09-blocked and is not restated. Framed "in progress," never "paid" (not collected, reconciliation §5). No client name (C-11).ledger row 24 |
| Operate tier concreteness: tokens refreshed before lapse, billing automation fail-closed, compliance calendar ahead of lead time, quarterly earning-its-keep reviews | pricing.html Operate tier | SERVICE DESCRIPTION (record-grounded) | Describes the defined service in method 02-runbooks/production-support.md (drafted from the recurring support record). Offer copy, not a market claim; the fail-closed billing posture matches commitment 4's grading.ledger row 25 |
| "What you own": code + system yours on payment; logic ownership + vertical exclusivity for strategic clients; retainer opt-in, never lock-in | pricing.html; consistent with founding-five.html "What you own" | CONTRACTED (pattern, 7 real SOWs) | Method 02-runbooks/contract-and-sow-posture.md — extracted verbatim-grounded from the real SOW corpus; IP-tiering is the documented default/strategic dial.ledger row 26 |
| "Validate the diagnosis, contest the cure, keep the standard" (the vet-case verdict form) | proof.html vet case | FOUNDER-CONFIRMED (J1 refined) | Judgment library v0.2 J1 refinement, marked "now correct for the website" in the amendment; row 16's judgment-principles class extends to it.ledger row 27 |
| Homepage Founding Five strip ("This section comes down when the seats are filled") | index.html pre-footer | COMMITMENT | Nav decision on record: FF stays out of the buyer top-nav (different audience); dedicated strip + footer link while seats are open; strip removal on fill is itself a commitment this ledger will track.ledger row 28 |
| Verbatim named testimonials: Adam Brochert, MD (HelpMeRad) + Reena Jadhav (HealCircle); anonymized case cards now name-linked | index.html quote grid; proof.html vet case | FOUNDER-CONFIRMED — permission on recor… | Ruled 2026-07-20 (method 05-market/site-cutover-2026-07-20.md); quotes migrated verbatim from the legacy site where they were already published with attribution. Resolves site issue #3; issue #4's gate satisfied for these two names + HelpMeRad/TrueTime/HealCircle ONLY.ledger row 29 |
| "Context is not the same as review" thesis (author and reviewer must be different processes) | proof.html incidents section | FOUNDER-CONFIRMED (thesis) | Founder-authored argument migrated from the legacy proof page (published under his name); an argument about mechanism, not a measurement, and framed as such.ledger row 30 |
| Documented incidents: #001 4-drifts/25-findings/~30s/0-shipped · #002 9-unauth-routes → 1 structural rule · #003 9+ fix(security) commits / 57 RBAC tests / 11 route files | proof.html incidents section | PLATFORM-RECORDED (self-reported; #003 … | Regraded 2026-07-21 per coherence-audit finding 9 (J10): #001/#002 are the platform recording itself; #003 carries repo-history receipts. #002's structural rule cross-cites the audit-graded ungated-routes invariant (alignment audit, commitment 1 receipts); #003 is grep-able in the platform repo history. #001 is the platform's own persona-review record.ledger row 31 |
| Named case depth: HelpMeRad (6wks MVP, HIPAA day one, training-example library (count PENDING-GRADE per C-16 class)), HealCircle (5wks to launch, 911 escalation, 0 PHI stored, 686 posts migrated), an edtech equity partnership (8 modules; name gated pending consent) | proof.html case cards | FOUNDER-CONFIRMED naming [C-17] + platf… | Names per 2026-07-20 permission ruling; depth figures migrated from legacy case studies (client-approved publications). Equity-model disclosure was already public on legacy site.ledger row 32 |
| Bio depth: Mindbloom co-founded, $160M+ raised, acquired (Welltok → Virgin Pulse); Microsoft/Nordstrom/Virtuoso arc; 99%→99.99% uptime; RV/travel story | about.html | FOUNDER-CONFIRMED (self-reported; publi… | Migrated from legacy about pages where published under the founder's name; funding/acquisition are public-record facts; the RV story is the founder's own narrative, framed as narrative.ledger row 33 |
| Investor-page figures (collected · contracted+pipeline · client and operator counts) — the numbers live on the noindex investors page and are not restated on this indexed one | investors.html (noindex, footer-linked) | FOUNDER-CONFIRMED w/ methodology note [… | Figures ruled on the record 2026-07-20; methodology footnote added on-page; contracted/pipeline figures dated as-of June–July 2026 with "current numbers on a call." Upgrades to MEASURED on Stripe/books export. Investors page hereby ENTERS this ledger's scope (was deferred pending issue #5).ledger row 34 |
| Platform page: four-layer architecture, component counts "mid-2026 figures" (73 tools / 9 hooks / personas), without-vs-with scenarios, "we asked the AI" feature list, "not for sale — comes with the work" | platform.html | PLATFORM RECORD (as-of dated) + FOUNDER… | Migrated from legacy product pages, reframed per Engagement-000 sequencing: the platform is presented as the practice's infrastructure and the Founding Five's toolkit, NOT a SaaS offer — no pricing, no Get Access. Counts date-stamped mid-2026 to prevent rot.ledger row 35 |
| Domain cutover config: CNAME=thecollectiveai.dev + canonical og/404/README URLs on the domain | CNAME (merged PR #35); all pages og tags; 404; README (this PR) | CONFIG (cutover ruled 2026-07-20) | Decision record: method 05-market/site-cutover-2026-07-20.md. Inert until Travis flips Cloudflare DNS. Honesty note: PR #35's commit message claimed these URL edits but its script had failed — only CNAME landed there; this PR carries the real edits.ledger row 36 |
| Hero tension line ("It works in the demo. You can't bet the business on it.") + hero cred strip (Three decades — Microsoft · Nordstrom · Virtuoso · Mindbloom acquired; per registry C-18, supersedes the earlier 25+ claim-down) | index.html hero | FOUNDER narrative + BIOGRAPHICAL (publi… | 2026-07-21 CMO adjustment (external critique adjudicated; adopted only receipt-compatible items). Tension line names the ICP's known reality, no manufactured claims; cred entries are the row-33 biographical set, no "Xbox" (appears in no receipt).ledger row 37 |
| Crawlability layer: robots.txt (all crawlers + AI welcomed), sitemap.xml (investors excluded — noindex), llms.txt (AI-readable summary + "cite the grade" instruction), canonical tags | robots.txt; sitemap.xml; llms.txt; all pages | CONFIG | 2026-07-21, founder-directed ("accessible to search engines and AI"). llms.txt instructs AI citers to carry grades with numbers — the claims discipline extended to machine readers. Railway mirror stays noindexed.ledger row 38 |
| Last-Mile Checklist page: 12 items, ungated, "complete enough to use without us"; hero secondary CTA now points to it | last-mile.html; index.html hero | CONTENT (runbook-derived) + COMMITMENT … | Items derived from the practice's own runbooks (production-support, deploy-gate, incident-response, contract posture) and the documented incidents; item 9 encodes the HRM-Bridge examples-are-policy lesson WITHOUT its ungraded stats (C-16 still pending). Ungated by design — the diagnostic CTA now has real fulfillment behind it.ledger row 39 |
| Story sharpening (external critique round 2, adjudicated): "buy one milestone" philosophy block + 3-step strip; hero chip broadened to "founders who've outgrown the prototype" (ruled H1 unchanged, keeps Claude specificity); outcome-language tier headings; metrics-first case openers | index.html; pricing.html; proof.html | FOUNDER-CONFIRMED positioning (J5 resta… | Milestone philosophy = C-04/C-06 in second person. Declined from same critique: hero-line swap (praised asset, no churn), hiding FF strip (live campaign), invented dollar-impact figures (no receipts — enter when graded). Case-opener metrics all previously published/receipted. UPDATE 2026-07-22: Prove-tier entry price subsequently ruled to a one-time fixed-scope milestone (see row 52); the milestone philosophy now matches the tier price wording.ledger row 40 |
| Homepage trim (~30%): shift/method/proof ledes tightened, supply-chain box condensed, pricing grid → 3 outcome mini-cards, ungraded "11/10" anecdote stat REMOVED | index.html | EDITORIAL + conformance | Critique round 2 adoption. The 11/10 stat was pre-registry, unreceipted, and cute — cut per the grading rule (charm is not a grade). No graded claim wording altered.ledger row 41 |
| Nav CTA contrast fix (specificity bug: .nav-links a slate overrode .btn white → slate-on-pine, sitewide) | styles.css (all 11 nav pages) | BUGFIX (accessibility) | Founder-reported 2026-07-21; one shared-stylesheet rule; hover/focus states covered.ledger row 42 |
| PROCESS FAILURE, recorded: claims.html drifted 28 rows behind this ledger across ~10 PRs, violating row 13's own no-drift rule — caught by founder question 2026-07-21 | claims.html (now generated); this ledger | RETIRED failure mode (structural fix) | J8 applied: the standard was remembered, so it failed. Fix: claims.html's table is now GENERATED from CLAIMS.md (scripts/render-claims.py) and a PR-blocking CI check (claims-sync-check.yml) fails any divergence. Full-surface drift audit run same day: sitemap/llms/canonicals/links CLEAN; investors footer variant fixed in same PR. The miss and the gate both stay on the public record.ledger row 43 |
| "Code is law" applied to this ledger's own conventions: registry-blocked strings CI-scanned on every public surface; page-edit PRs must move the ledger or declare claims-exempt on the record | .github/workflows/claims-guard.yml | ENFORCED (gate, not memory) | Founder's 2026-07-21 post ("Instructions are suggestions. Examples are training. Code is law.") applied to our own shop the day it was published: rows 13's sibling conventions were still prose — now they fail PRs. Pattern-list edits require a conscious same-PR act.ledger row 44 |
| COHERENCE AUDIT RESPONSE (method PR #68): BLOCKING hotfixes — ungranted client names anonymized on page + in this ledger; Operator Program retired per the written ruling; pre-pivot title/meta/og swapped to ruled positioning; unruled investor figures struck pending A4; sub-$7K corrected to $7.5K; zombie 10+ framings cleared; rows 31/32/33/34-adjacent regraded per J10 findings; bio claims capped at self-reported and years claimed down to 25+ | sitewide | ENFORCED (audit-driven) | Independent adversarial audit, method 02-runbooks/site-lane-coherence-audit-2026-07-21.md. Open rulings A1–A7 tracked there; A2 shipped with the audit's default wording pending Travis.ledger row 45 |
| Guard hardening: protected-name patterns removed from the serving repo's workflow; name enforcement moves to registry-derived checks (A6 plumbing) | .github/workflows/claims-guard.yml | PARTIAL (interim) | Finding 3/16: a denylist must never enumerate what it protects in a public-serving tree. Interim: non-sensitive figure patterns stay inline; name scanning parked on the METHOD_REGISTRY read secret (Travis mints).ledger row 46 |
| README-DEPLOY.md REMOVED from the served tree — it named a walled client (C-11) on a public URL since launch; content long superseded by issues #3–#8 | (deleted file) | ENFORCED (leak closed) | Caught by the hotfix PR's exit breach-scan, missed by both the guard and the independent audit — third mirror-class lesson this week: the scan surface must be the full tracked file list, not a narrower filter. Guard follow-up queued with the A6 claims.json work.ledger row 47 |
| Rulings A2/A4/A5/A7 applied (Travis, 2026-07-21): title/meta wording CONFIRMED; investor figures RESTORED as FOUNDER-CONFIRMED with as-of dating and the measured-subset note; bio bar confirmed (self-reported cap, 25+ years); FF seat count now lives in seats.json (Travis-updated) with a guard job enforcing the strip-removal promise | investors.html; seats.json; claims-guard.yml | FOUNDER-CONFIRMED (rulings on record) | Audit rulings letter, method site-lane-coherence-audit-2026-07-21.md. A1 remains OPEN — the edtech partner stays anonymous pending consent. A6 plumbing waits on CLAIMS_REGISTRY_TOKEN (site repo has no secrets yet).ledger row 48 |
| Mobile navigation (hamburger, sitewide — nav links were display:none on phones with no menu) + JSON-LD structured data (Organization/WebSite on index, Person on about, Service+3 offers on pricing) for SEO/AEO | styles.css; reveal.js; all nav pages; index/about/pricing head | CONFIG (accessibility + SEO/AEO) | Founder-reported mobile-nav gap 2026-07-21. Structured data mirrors on-page graded claims only (tier prices already public; bio entities per A5 bar). AEO layer now: llms.txt + JSON-LD + canonicals + sitemap — all inert to the public until the DNS flip.ledger row 49 |
| "Who you'll work with": Travis + Lindsey Harvey (SVP Partnerships; Loop & Tie founding GTM, first-customer-to-$70M+, named enterprise logos), with photos | about.html team section | SELF-PUBLISHED + FOUNDER-DIRECTED (rule… | Travis ruled her self-published LinkedIn (title, bio, figures — her claims, quoted as hers), his public welcome announcement, and the photo he supplied constitute sufficient basis; no further ask needed. Bio near-verbatim from her own public wording.ledger row 50 |
| FF page commitments sharpened (page-review edits E1–E5): weekly 90-min live block + daily async ("the block never goes dark"); 12-dimension measurement graded wk 1/12/23 with published delta; success fee explained as expansion-ladder trigger (Option A ruled — entry deals priced below $10K by design); "Principal of your own practice"; precedent sentence promoted to its own beat | founding-five.html | COMMITMENT (bounded, checkable) | Edits from the FF-lane page review, founder-forwarded 2026-07-21. HELD pending rulings: R2 Honest Exit row (needs the refund figure before it prints), R3 block day/time (copy stays day-agnostic). E6 verified: no "Quantum" on page (0 hits), no co-delivery latitude, honest print/filters/personas untouched. Cadence/measurement claims mirror enrollment Terms A/B/E held in the FF lane.ledger row 51 |
| PRD-004 sealed row (internal Interlock prediction — the ledger applied to our own operations; baseline captured, scores Oct 2026) | ledger.html | MEASURED (ledger state) | Method 05-predictions/PREDICTION-004-interlock.md, sealed 2026-07-21 by founder ruling. Internal-ops subject de-identified on-page; publishing our own operational bets is the Engagement-000 posture applied forward.ledger row 52 |
| CROSS-LANE CONSISTENCY FIX: meaning-lane PR #66 (ruled: hero voice-split, three-decades per C-18) merged past a FAILING ledger-coverage gate (non-required check) leaving ledger rows + about.html stale; aligned here | about.html (4 spots); rows 17/37 updated | ENFORCED (fix-forward) | The gate worked and was overridden by merge rights — second live proof the guard checks must be ruleset-REQUIRED (Travis, control-plane, now urgent). Rulings themselves honored: C-01 voice-split + C-18 are registry-backed (method #81).ledger row 53 |
| CONVERSION-PATH FIX: every "Book" CTA now opens the founder's Calendly (probed 200; month param omitted so the link never stales); mailto demoted to visible-text fallback (travis@, founder-published; unverified hello@ eliminated); FF page keeps its intentional mailto filter | all pages | ENFORCED (fix + lesson) | Founder-caught pre-flip: every CTA funneled to a bare mailto — invisible no-op without a mail client, address never provisioned-verified. Root cause: audits validated that links RESOLVE, never that conversions BEHAVE (my crawler explicitly skipped mailto). Lesson to checklist: behavioral test of every conversion action before any launch.ledger row 54 |
| Email architecture: sales@ = funnel fallback (routes to Lindsey), info@ = general, travis@/lindsey@ on their who-you'll-work-with cards; FF page keeps travis@ ("read by me" filter) | index contact; about cards; llms.txt; founding-five unchanged | CONFIG (founder-directed 2026-07-21) | All four addresses CONFIRMED provisioned by founder 2026-07-21 — the conversion path is fully live-backed (Calendly primary, provisioned email fallback); unverified hello@ eliminated. ADDENDUM 2026-08-04: FF page mailto CTAs replaced after founder-reported blank-tab failure (same class as row 54) — nav anchors to the bottom apply section; address is JS-assembled (never in static HTML, spam-scrape hardening), shown as selectable text + copy button, mailto demoted to secondary. The one-email filter is unchanged; the mechanism now behaves.ledger row 55 |
| Public status page linked (footer sitewide, market-standard placement; system.html cites it as the monitoring claim's living receipt) | all footers; system.html; llms.txt | ENFORCED (probed live before linking) | status.thecollectiveai.dev → UptimeRobot, HTTP 200 verified pre-ship per the row-54 lesson. A public monitor is the strongest form of the commitment-4 claims.ledger row 56 |
| LIVE: thecollectiveai.dev serves this site (cutover verified 2026-07-22 06:28 UTC — 16/16 paths 200, branded 404, Calendly CTAs, status link, github.io 301 chain; www pending cert issuance at verification time) | the domain itself | ENFORCED (verified live) | Full post-cutover pass in the site session. The legacy site and its retired claims left the public internet at the same moment. Crawl layer (robots/sitemap/llms.txt) now reachable by search and AI engines — the SEO/AEO clock started here.ledger row 57 |
| Hosting moved to Railway (collective-site-live serves thecollectiveai.dev) with a staged GitHub workflow: push main → review service → gated approval → live. Repo Dockerfile/Caddyfile; Pages deploy disabled | Dockerfile; Caddyfile; .github/workflows/railway-deploy.yml | CONFIG (post-incident) | 2026-07-22 launch-day outage post-mortem: grey-cloud cert gap (my error) → moved off Pages to Railway per founder. "merge=deploy" restored and GitHub-visible. Needs 2 Railway project tokens + a production environment reviewer (Travis).ledger row 58 |
| FF page clarity pass: category noun ("a six-month practice-building apprenticeship … not employment, not a franchise, not a promise of supplied clients"); total price $15,000 stated in prose above the term sheet; card 01 "my books" made literal (anonymized SOWs, recorded calls, pipeline history, pricing decisions, delivery artifacts); card 03 header "my name on the line" → "my review on the line"; new CURRICULUM row (24 weeks mapped end to end; weeks 1–6 built module-by-module citing the receipt each lesson came from) | founding-five.html hero, boot cards 01/03, term sheet | FOUNDER-CONFIRMED (rulings) + MEASURED … | Three-way external review of the FF launch copy, adjudicated 2026-07-22. Name-use ruled METAPHORICAL (Travis, 2026-07-22): no client-facing right to invoke Travis/The Collective on a Principal's engagement is offered, so the header now claims only what the card body always said — review before it goes out. Category noun qualified per the DOL-apprenticeship concern (the bare word implies paid employment). Total-price disclosure closes a no-fine-print gap: a buyer previously had to sum two rows. Curriculum claim graded to its true build state — W1–6 authored, receipts-cited and lint-clean in founding-five PR #17/#21; W7–24 outlined and source-mapped in the Phase 2 blueprint (PR #20), NOT authored — the row says "mapped end to end" and "weeks 1–6 are built" for exactly that reason, and must not be upgraded to "24 weeks built" until the modules exist. STILL HELD, unchanged: R2 Honest Exit refund figure; R3 block day/time. Two open rulings now gate further FF terms copy — the access boundary (weekly block, business-day async, one active opportunity in formal review, out-of-scope list; "the block never goes dark" survives only if the terms define it) and the success-fee trigger (published as *signature*; a move to *collected revenue* is a term change that moves row 12 and this page together).ledger row 59 |
| (a) PUBLIC-SURFACE HYGIENE on this ledger — exploit-enabling specifics generalized (the categories are recorded in the internal registry); registry-BLOCKED/RETIRED figures no longer restated by the rows that block them; investor financials off this INDEXED page (they stay on noindex investors.html). (b) FF page copy pass — "no fine print beneath" retired in both places; hero exclusions split from the proposition; grade badge given its own line; "an ownership" and "both" fixed; measurement and later-weeks rows made legible to a cold reader | CLAIMS.md rows 2/4/9/10/20/24/34/45/46/47; founding-five.html hero, $15K line, term-sheet header + curriculum/measurement rows | ENFORCED (hygiene) + EDITORIAL (copy) | STRAND NOTE, recorded per the verify-by-content rule: the hygiene pass was pushed to PR #87's branch but did NOT reach main when #87 merged 2026-07-23 21:50 — third strand of this class this week (heartbeat #77, curriculum #17, now this). The leaks were live on the public claims page until this PR. Structural fix remains "delete head branch on merge" + do not merge a branch with in-flight commits. "No fine print beneath" retired because it is an absolute the enrollment agreement cannot honor — necessary operating terms (payment, IP, confidentiality, public reporting, early exit, qualifying opportunity) are not fine print, but the phrase promised their absence. Fourth external review adjudicated 2026-07-23; two of its claim-precision suggestions REJECTED as claim-down errors: "five of five entered and expanded" is exactly supported (C-04/C-05 + 100% milestone-entry expansion), and the client did self-build the prototype, so that sentence stands. Two blockers remain OPEN and gate the rest of this page: the success-fee trigger (signed vs collected — "only exists if you win" is undefined until ruled, and row 12 moves with it) and the access boundary (card 03 + cadence row still promise unbounded review; narrowing a published promise is a term change, not an editorial one, so the meaning lane did not touch it).ledger row 60 |
| FF TERM CHANGE (two founder rulings): success fee restructured from "$7,500 due on the first $10K+ SOW" to 10% of collections during the term, capped at $7,500; access boundary stated on the page — one 90-minute block weekly for 24 weeks, async review answered on business days, one opportunity in formal review at a time; "the block never goes dark" retired from both places it appeared | founding-five.html success-fee row, cadence row, card 03, $15K prose; row 12 updated | COMMITMENT (bounded, checkable) | (1) The old flat-fee trigger created misaligned incentives at its boundary and punished exactly the expansion the method teaches; a capped percentage of collections removes both defects and defines "if you win" for the first time. The full analysis is preserved in the internal registry. (2) Access promises had drifted maximal ("every deal, every call, daily async, never goes dark") — the opposite of the sealed enrollment design (system is the product, defined cadence not on-demand, cadence may shift while outcomes hold, senior-operator co-delivery permitted). Two independent reviewers flagged that narrowing express page promises later would read as exactly the fine print this page denies. The bound is not a retreat: "one opportunity in formal review at a time" describes the rotating hot seat that has been in the curriculum since week 2. Still OPEN: the Honest Exit row — no early-exit/refund policy appears anywhere, and it is the second question a $7,500-upfront applicant asks; it prints when Travis sets the figure.ledger row 61 |
| FF ledger publication clarified as a TERM: outcomes publish by row number (FF-001…FF-005), not by name, unless the founder asks to be named | founding-five.html accountability row + card 03; ledger.html FF paragraph | COMMITMENT (founder-ruled at merge) | The reserved-rows-before-signature commitment is the strongest thing on the page and stays. The unqualified half — "your outcome publishes either way" — carried an unnamed cost: a reputation-sensitive applicant reads it as *I pay $15,000 and my failure becomes a public record with my name on it*, which lands hardest on exactly the target personas (the exec who won once; the domain expert in a small field). NOTE ON WHAT THIS IS: the prior ledger wording de-identified CLIENTS, and was silent on whether the FOUNDER is named — so this is not a restatement of existing policy, it is a term being set. Row-number-default protects the downside; opt-in naming preserves the upside for a founder who wants credit for a win. All three surfaces moved together per the binding convention.ledger row 62 |
| NEW PAGE think.html ("Think it through") — a guided prospect thinking session. Material claims on it, mapped: (a) HelpMeRad line "6 weeks to production, daily clinical use, zero engineers hired, client became investor" = the proof.html case strip verbatim-in-substance, rows 29/32; (b) HealCircle line "5 weeks to a live event, 686 posts migrated, zero PHI stored" = row 32 depth figures; (c) "clients start small, and every expansion has followed a delivered milestone" = rows 19/24 (C-04/C-05 milestone-entry + expansion-on-evidence); (d) privacy line, updated again with transcript recording (founder-directed): the page now states plainly that an ANONYMOUS record of questions and answers, guided clicks included, is kept and read to improve the page; no identity is asked or attached — BY-CONSTRUCTION the record carries only a random session id, never IP or identity fields (think-api/server.mjs), and the page, the ledger, and the model's own answer about data all say the same thing; a record links to a person only if they volunteer identifying details in free text; (e) the four "stress-test" transcripts are labeled on-page as simulated, written by us. The page states NO prices, NO new client names, NO new outcome figures. | think.html | INHERITED (rows 19/24/29/32) + BY-CONST… | Page added from the method lane's prospect-experience build (skeptic-reviewed; c-check-draft sweep clean on visible copy). CTA links to the standard Calendly working-session slot. LIVE LAYER SHIPPED in the follow-up PR: think-api/ (Node proxy, Railway private networking) + one Caddy route; the page reveals the ask-anything line only when the service answers its probe, so absence degrades to the guided page. GROUNDING: the model's context = system-prompt.md + method-context.md, the latter distilled ONLY from this site's graded public pages (Delivery Assurance commitments, last-mile checklist, rows 29/32 proof) so the generative line can only elaborate on already-graded material. GOVERNANCE OF THE GENERATIVE LINE: think-api/system-prompt.md — one question at a time, never quotes a price, qualifies out honestly, discloses it is an AI, declines off-topic/roleplay, proof limited to the two consented case lines (rows 29/32), ~12-turn cap, no persistence. CAVEAT, on the record: a generative line can phrase things this ledger has not graded; the prompt confines it to graded material, but that is governance, not construction — the transcripts tab shows prospects how the rules are meant to hold. ADDENDUM 2026-08-04: the record now also includes three interaction clicks (receipt-copy, stress-tests-tab view) within the stated "guided clicks included" scope, and an optional src campaign word carried on inbound links (self-chosen link tag, sanitized [a-z0-9-], never identity). ADDENDUM 2 (2026-08-04): the record now also counts page views (one per session) and stamps a coarse two-letter country code read from the CDN edge — the IP itself is never received into the record; the page's privacy note says the same in the same breath.ledger row 63 |
| Operator review view at /review.html (unlinked, noindex): reads the Think anonymous record behind separate, revocable, read-only operator access controls; identifies sessions never people; names appear only where a prospect typed one | review.html; think-api/server.mjs | CONFIG (internal tool) | Built 2026-08-04 for operator prospect-intelligence. Access model: key pasted once, held in that browser only, bearer header per request; no transcript bytes in page source. Analytics engine behavior-tested against the real event vocabulary (9 assertions). Extends row 63's transcript claim with the read side. ADDENDUM 2026-08-04: view enriched with derived session metadata (duration, depth, intent, self-ID quote, automated flag, src ranking); think-api can ping the operator's own Slack once per human session at its first meaningful moment (NOTIFY_WEBHOOK, unset=off, fail-closed; content limited to what the view already shows; internal/probe sessions never ping; 8 notifier + 10 enrichment assertions).ledger row 64 |
| HONEST EXIT term printed (closes the R2 ruling HELD since rows 51/59/61): full refund of the $7,500 entry through the Week-2 bailout checkpoint; exit there, refunded in full; after Week 2 the entry is committed; success fee unaffected (collections-only, unchanged) | founding-five.html term sheet (new row between Entry and Success fee) | COMMITMENT (bounded, checkable) | The term names what the curriculum already builds: module c001-live-w2 "Discovery Clinic — and the Honest Exit", isBailoutCheckpoint: true, "anyone without a schedulable prospect decides to push or exit clean" (founding-five repo curriculum/c001-modules.ts:484-498, verified on main 2026-08-04). Answers the second question every $7,500-upfront applicant asks, BEFORE the launch post instead of in its comment thread. Ruling mechanism per row 10 precedent: Travis rules by merging this PR. The enrollment agreement's refund clause must mirror this row's wording.ledger row 65 |
| PUBLIC-SURFACE HONESTY PASS (P0): four system.html claims brought down to the wording their own rows already ruled (supply-chain posture and its reach, fail-closed scope with its edge stated at client altitude, incident notification as a commitment); verification lexicon standardized to "internally verified and publicly graded" ("audited" now reserved for named third parties); a small number of internal-class fragments re-homed from this ledger to the internal registry, with history preserved there | system.html commitments 2/4/5; claims.html lede + header; rows 59/61/64/65 | ENFORCED (hygiene) + EDITORIAL | Target: a skeptical reader should find precision and accountability; no reader should find operational detail usable against us or our clients. Wording follows rows 2, 4/5, and 7 as already ruled; nothing was upgraded, three claims were claimed DOWN. History is never deleted, only re-homed.ledger row 66 |
Why publish this? Because a delivery practice that grades its clients' software on evidence should survive the same treatment. If someone selling you certainty won't grade their own claims, ask why.