CONSTITUTION v1.0 — FROZEN 2026-07-18

Applies to founders, employees, contractors, reviewers, investors, advisors, and board members. Signature required before access to customer data or claim records. Every law here binds behavior or system design; none is decoration.

1. Purpose

We help organizations state the truth about their AI — discovered from their real systems, organized as evidence, and expressed so that the people relying on those statements are never misled. The immediate product is the AI Disclosure Package. The compounding asset is the customer-owned evidence registry. The product may change; the evidence standard may not.

2. Trust Precedence (Amendment 001)

Evidence standard > recipient protection > customer interests > growth > revenue > speed.

• Recipient protection is a constraint, not a commercial relationship: we never help a customer create a materially misleading impression.
• The customer decides whether to submit an answer. The customer never controls the evidence record, claim classification, or the rendering of uncertainty.
• We may decline delivery, revoke an issued artifact, or terminate an engagement where concealment or material misrepresentation is attempted.

3. Permanent Laws (no board, investor, or ownership vote can override; entrenched in contracts and charter to the maximum extent law allows)

P1. We never sell a conclusion. We charge for collection, organization, analysis, drafting, review, maintenance, and hosting — never for a predetermined or favorable outcome. P2. No numeric trust scores of organizations, ever. Evidence coverage and freshness are the only published measures. P3. No silent rewriting of history. All records are append-only; corrections are visible entries. P4. Customers own their data and records: full export in open formats, deletion on request, no training of shared models on customer content without explicit, informed, revocable, granular opt-in (default off). P5. No certification, assurance opinion, or attestation of work this entity was paid to prepare. That institution, if it ever exists, is structurally separate. P6. The claim-type system's integrity: no role — including the CEO — may override how uncertainty renders. P7. No sale or licensing of customer-derived intelligence. Aggregate benchmarks only with opt-in, k≥20, published methodology, retroactive exit right.

4. Strongly Protected Laws (change requires supermajority board + Trust Officer concurrence + 60-day advance notice to customers)

S1. The claims-language policy (§6 below and Blueprint §2.1 as amended). S2. Review independence: reviewers report outside sales; reviewer compensation has zero revenue linkage; sales roles have no write access to claim records. S3. Transparency-report scope and quarterly publication. S4. The benchmark rules under P7. S5. This tier's own membership list.

5. Customer Data Rights (contractual, every customer)

D1. Read-only, least-privilege, pre-listed access scopes. D2. Source code is never retained: facts and hashes extracted, code discarded; snippets only with per-snippet consent. D3. Never collected: credentials/secrets, production end-user personal data, anything outside declared scope; accidental ingestion is a disclosed incident. D4. Instant self-service revocation of any access. D5. Full export ≤30 days, deletion certified ≤35 days; the surviving issuance log (hashes, dates, types, approver roles — no confidential content) is disclosed in advance. D6. Access transparency: every human/system access to customer data is logged and customer-visible on request. D7. Legal and government requests are disclosed to the customer unless we are legally gagged; we challenge overbroad requests; counts are published.

6. Claims-Language Laws (Amendment 002)

C1. The only claim vocabulary is: Directly Observed · Customer Declared · Document Supported · Inferred · Human Reviewed · Third-Party Attested · Unsupported/Unknown. C2. "Verified" is not used — not as a badge, seal, product name, or marketing term. "Independently observed" only as a descriptive sentence with exact scope, method, source, date, and limitations. C3. Never used as verdicts, about customers or ourselves: certified, compliant, validated, safe, secure, trustworthy, approved, audit-ready. Replacements: "consistent with [requirement] based on [evidence] as of [date]"; "evidence coverage: N of M claims supported." C4. Marketing, sales, and support language is held to this policy identically to product language. Violations are trust incidents. C5. Every delivered artifact states scope, dates, limitations, and claim types; gap lists appear in the executive summary, never only in appendices.

7. Evidence Integrity Laws

E1. Every material statement in a deliverable carries a claim ID; the assembly process rejects unreferenced material statements. AI drafts prose; AI cannot introduce a fact. E2. Evidence is hashed on ingest; originals (or extracted-fact + hash where we decline custody) are stored append-only. E3. Claim types are provenance, not quality grades; types are never silently upgraded; staleness decays visibly toward Unknown. E4. Contradictions are recorded, rendered as conflicts, and resolved only by reviewer entry — never averaged, never picked silently. E5. "We do not know" is a complete, deliverable, successful answer.

8. Review Independence Laws

R1. No artifact ships reviewed only by its author — even if the second reviewer is an external contractor. R2. First approval of any claim, all Inferred claims, all declaration intake, all legally sensitive families (training-data rights, PII flows, warranties), and every externally delivered artifact require human review. R3. Reviewers recuse on ownership, equity, or authorship conflicts. R4. Customer pressure to soften a claim goes to review, never to accounts; outcomes are limited to (a) new evidence changes the claim or (b) the claim stands. A third pressure incident triggers off-boarding review.

9. Correction and Revocation Laws

X1. Material error: record corrected visibly, customer notified ≤5 business days with corrected artifact. X2. If a delivered artifact reached recipients, the customer must forward the correction; refusal → we mark the artifact "superseded" in the issuance log and note the refusal. X3. Corrections are free, forever, for everyone. Paying for faster correction of our errors is prohibited. X4. Revocation (concealment/misrepresentation discovered): issuance-log entry flips to revoked; engagement terminates; the decline log records why.

10. Prohibited Business Models

Payment for favorable findings, suppression, claim-type upgrades, faster error corrections, trust-page embellishment, placement or ranking; undisclosed referral economics; preferential claim treatment for investors or partners (their engagements receive enhanced review sampling); person-level employee surveillance features; sale of customer-derived intelligence; fear-based marketing.

11. The Company We Refuse to Become

Surveillance company · data broker · pay-to-play verifier · opaque scoring authority · compliance-theater vendor · fear marketer · lock-in business · uncertainty-hider · proof-fabricator · silent history-rewriter · customer-competitor · shadow regulator or market gatekeeper · certifier of self-prepared work · AI-washing accomplice · too-big-to-audit. (Operational detail: Blueprint §20 — temptation, early signal, policy, enforcement, breach action for each.)

12. Amendment Process

Written proposal → impact-on-trust analysis → approval at the affected tier (Permanent: cannot be amended to weaken; may be strengthened by unanimous board + Trust Officer + customer notice; Strongly Protected: per §4; Policies: executive, logged) → published changelog entry. This constitution has a visible diff history; a silently changed constitution is a breached one.

13. Oath (signed by every founder and employee)

I will not ship a claim without its evidence or its honest absence. I will defend "we do not know" as a success. I will not sell a conclusion, a score, or a softer sentence. I will keep the customer's record theirs and the recipient's reliance safe. I will correct errors visibly, quickly, and free, and never rewrite history. I will hold our marketing to the standard of our product. When trust and anything else conflict, I will follow the precedence in §2 — and if the honest version of this company cannot survive, I will close it rather than run the dishonest one under an honest flag.

14. Enforcement and Breach Consequences

• Any employee may refuse an instruction violating this constitution by citing it; retaliating against that refusal is a terminable offense by the instructor.
• Claims-language and evidence-integrity violations are trust incidents: logged, root-caused, reported in the transparency report.
• Fabrication, backdating, or silent record modification: termination of those involved, customer notification, public report entry.
• Constitutional breach at the executive level: the Trust Officer (or external trust advisor pre-hire) reports directly to the board; the whistleblower channel bypasses management.
• Repeated red trust metrics (Blueprint §13) freeze growth spending until green. Growth never buys down a trust deficit.

Frozen as v1.0 on 2026-07-18 incorporating Amendments 001–003. Diff history begins now.