Constitution & governance

Version 1.0. This document states what Constitut measures, how it measures it, and the limits it may never cross. It is the settled part of the protocol. The mathematical law that turns evidence into positions lives in scoring-rules/ and is versioned separately; this Constitution governs that law and everything built on it.

This document is amended only through the process in GOVERNANCE.md. The scoring rules are amended through the same process. Neither is changed to suit a result.

Article I — Purpose

Constitut computes and publishes rankings of consulting firms from public, archived evidence under a disclosed rule set. Its purpose is to be a record that anyone can inspect, re-run, and hold to account — not an authority anyone must trust.

Constitut has no owners, no officers, and no opinions beyond its formulas. Humans touch the system in exactly two bounded, published places: the review of objections, and the rare extraordinary admission of a firm of exceptional public contribution (GOVERNANCE.md §4–§5). Neither may set a position by hand — an admission only makes a firm eligible to be computed; every decision of either kind is published. All else is decided by the formulas.

Article II — What is measured

Every firm is scored on six parameters. Their definitions, accepted evidence, verification rules, tier-specific weights, and recency treatment are fixed in the scoring rules in force and summarized here:

A firm's score is computed as:

FIRM_SCORE = Σ ( Pᵢ × Wᵢ ) × C

where Wᵢ are the tier-specific weights and C is the confidence multiplier determined by how much archived evidence stands behind the score. The weights, the confidence bands, the evidence floor, the recency decay, and the tie-break order are all published values in the scoring rules — nothing in the computation is discretionary.

Article III — The evidence law

1. Every scored fact has a public source URL and an archived copy (Wayback or equivalent). No evidence contributes to a score without both.
2. Below the evidence floor named in the rules, a firm is not listed in a table at all. Absence is not a judgment; it is the absence of sufficient public record.
3. Evidence is drawn only from public sources. No confidential, leaked, or privately obtained material is ever used, under any circumstance.
4. Adverse public records are admitted only under P2, only from official public sources, recorded neutrally as a citation with no commentary, and only after the pre-publication notice workflow gives the firm the opportunity to object.

Article IV — The limits that may never cross

These limits are what let the protocol stand without asking anyone to take it on faith. They are amendable only by the heightened process in GOVERNANCE.md, and any amendment that would remove a limit rather than strengthen it is out of order.

1. Rank upward only. Inclusion is recognition. Constitut never publishes a "worst," never ranks downward, and never attaches negative commentary to a named firm. The only negative the protocol expresses is absence from a table.
2. No firm profiles. A record page (/record/{slug}) contains only Constitut-computed data — ranks held, awards, standing, corrections, evidence counts. It carries no logo, description, contact, link, or any firm-supplied content. Constitut describes firms only in the facts of its own record.
3. Quiet authority. The protocol never praises itself. It carries no superlatives, no marketing language, and no self-claims as headlines. Every claim it makes is a verifiable record linked to its proof. Its standing is meant to come from behaving like an institution, not from saying it is one.
4. Determinism. Research and drafting of evidence may involve judgment; the computation of scores and standings may not. Only code computes positions, and the same inputs must always produce byte-identical published outputs. This is asserted on every run.
5. Truth only. No statement in the genesis record, the governance text, the methodology, or any published page asserts anything that cannot be proven from this repository, a committed key, or a public record. No claim of legal domicile is made unless and until a real steward entity is formed and can be verified.
6. The record is the source of truth. Rankings, rules, evidence, registry, and corrections live as plain-text files under version control. No runtime database is ever the source of a published position.

Article V — Positions are opinion

Positions are the protocol's opinion, computed from the disclosed methodology applied to publicly available evidence. This line accompanies every table. A firm that identifies an error in the evidence, the computation, or the standing behind its position may object; the process and service standard are published in ../objections/policy.md, and every objection and its resolution is recorded in public.

Article VI — Time and standing

Positions change on a fixed cadence: monthly signals update scores and evidence without moving ranks; quarterly snapshots are the official movement events; each January an annual edition is frozen and tagged as a citable vintage. Standing — how long a firm has held a rank, how long it has been present in a table, and the records of a table — is derived by code from ranking history and is never stored as opinion or authored by hand.

Article VII — Governance and continuity

The authority to sign snapshots, awards, and rule changes passes through a published key chain, rotated on a fixed schedule so that records signed in one year remain verifiable in the next. Rules and this Constitution change only through the amendment process in GOVERNANCE.md. Because the institution is a public history and a set of published keys, it can be continued by anyone: if its stewards are ever captured or vanish, the honest fork inherits the record and the legitimacy that follows it.

This Constitution takes effect with the genesis record (GENESIS.md) and is binding on every snapshot signed thereafter. Its version is named in every page footer alongside the scoring-rules version in force.

Version 1.0. This document states how the protocol changes, who may change it, how the signing authority passes from one period to the next, and how the institution survives the people who steward it. It is written to be legible to a reader in a decade and to bind the stewards as much as anyone.

The Constitution states the limits that may never cross. This document states the process by which everything else moves.

1. Stewards and the founding period

A steward performs the one continuing human function the protocol permits: the review of objections (../objections/policy.md). Stewards do not compute, adjust, or override positions — that is the exclusive work of the engine.

• Terms. Steward seats are held for fixed terms of four quarters (one year), renewable once consecutively, after which a seat must rotate to a different holder for at least one term. Terms and their holders (by pseudonymous steward identity) are recorded in this repository.
• Founding-steward period. Until an objection-review panel of at least three stewards is seated, the protocol operates under a single founding steward, and this is stated plainly wherever governance is described. The founding steward holds the same limits as any steward and no more: no authority to alter a computed position, only to resolve objections against the published evidence standard.
• Recusal. A steward with any relationship to a firm under objection recuses from that objection, and the recusal is noted in the corrections log.

No steward is named as an individual. The protocol's independence rests on its process and its record, not on the identity or reputation of any person.

2. Amending the rules and the Constitution

The scoring rules (scoring-rules/) and this governance framework are amended through a single, deliberate process. The Constitution's Article IV limits are amendable only under the heightened rule in §2.4.

1. Proposal. An amendment is published as a written proposal with its rationale, committed to the repository. Anyone may submit one; a steward tables it for consideration.
2. Notice. The proposal stands for one full quarter of public notice before it may take effect. The record is public throughout; objections to the proposal are received like any other.
3. Approval. After the notice period, the amendment requires the approval of a supermajority (two-thirds) of seated stewards. During the founding-steward period, an amendment additionally requires explicit Boss sign-off, and this dependency is stated honestly until the panel is seated.
4. Heightened limits. An amendment that touches an Article IV limit is in order only if it strengthens that limit. No process may remove the rank-upward-only rule, the no-profiles rule, the evidence law, the determinism requirement, the truth-only rule, or the record-as-source-of-truth rule.
5. Effect. Approved rule changes take effect only at the next quarter boundary, never mid-cycle and never retroactively. Past snapshots remain computed under the rules that were in force when they were signed. Every change is recorded in scoring-rules/CHANGELOG.md with its date, rationale, and commit.

The rules are versioned. The version in force is named in every snapshot, award record, and page footer, so any published position can be traced to the exact law that produced it.

3. The signing key chain and its rotation

Every snapshot, award record, and rule change is signed with the protocol's Ed25519 publishing key. Only the public key is committed to the repository (../registry/keys/); private keys never enter it.

• Genesis keys. The designers' keys sign the genesis record and the first Constitution. Once the publishing key is established, the genesis keys are publicly retired. Their retirement is itself a signed, committed act.
• Annual rotation ceremony. Once per year, the current publishing key signs its successor, and the signature is committed to the key chain (../registry/keys/). Because each key vouches for the next, a certificate signed in one year remains verifiable years later by following the chain. No key is ever silently replaced; a break in the chain is itself evidence.
• Verification. Because public keys live in the repository, any third party — a firm, a journalist, a court — can verify a signature offline, without trusting the website or the stewards.

4. Objection review

Objection review is the protocol's only discretionary act, and it is bounded by the published policy and the evidence standard. Stewards resolve each objection to one of four classes — upheld, partially upheld, rejected, out of scope — and every resolution, with the change it produced and the commit that recorded it, is published in ../objections/log.yaml. Rule objections are not resolved as corrections; they enter the amendment process in §2.

5. Extraordinary admission

The evidence floor keeps the tables honest: a firm with too little public record is not listed. But the floor is a measure of quantity, and quantity occasionally misses a firm of rare quality — a boutique whose single landmark, independently cited contribution to a category outweighs a larger firm's routine output. This section provides, narrowly and in public, for that case. It is the second and only other place — besides objection review — where humans touch the system, and it is bounded so that it never becomes a way to hand-set a position.

What it is. On a documented request, the steward committee may admit a firm that falls below the evidence floor of a specific table, on the strength of an extraordinary, independently verified public contribution to that category. The admission grants the firm a floor exception for that table and period; it does not set the firm's rank or score. The engine still computes the firm's position from its evidence under the published rules — the admission only makes the firm eligible to be computed.

What it is not. It is not a placement, a score adjustment, an award, or a penalty. It cannot move a firm up or down by hand, cannot lift a firm above what its evidence computes, and — consistent with the rank-upward-only rule — is never used to exclude or diminish any firm. It does not manufacture standing: tenure and records accrue only from periods actually ranked, never retroactively.

The bar. Admission is rare and exceptional. It requires a contribution that is (a) genuinely extraordinary for the category, (b) independently recognized — cited, adopted, or relied upon by parties other than the firm — and (c) supported, like all evidence, by a public source and an archived copy. Ordinary marketing, self-published material, or simple under-coverage does not qualify; the remedy for under-coverage is more evidence, not an exception.

The process. An admission follows the amendment-grade process of §2: a published request stating the firm, the table, and the extraordinary contribution with its sources; committee review against this bar; and approval by a supermajority (two-thirds) of seated stewards. During the founding-steward period it additionally requires explicit Boss sign-off, stated honestly until the panel is seated. Recusal rules (§1) apply. The decision — for or against — is signed with the publishing key and recorded in registry/admissions/ and the public record.

Its effect in the engine. An approved admission is a signed, committed input to the scoring engine, exactly like evidence. Because it is an input and not a hand edit of the output, determinism holds: anyone who clones the repository, with the same evidence and the same admission records, computes the same table byte-for-byte. An admitted firm below the floor is listed at the lowest confidence band and is marked in the ledger and on its record page as admitted, with a link to the signed decision, so the exception is never hidden. The admission may be objected to like any other record, and lapses if the firm later clears the floor on its own evidence.

This provision widens the human touch-point from one place to two — objection review and extraordinary admission — and no further. Both are public, both are bounded by this document, and neither may set a position: positions remain computed, not chosen.

6. The taxonomy registry

The category taxonomy (categories.yaml) is a versioned registry. Each category carries a permanent three-letter code that never changes once published, because codes anchor URLs and registry IDs forever. Display names may evolve; codes may not. A new category, tier, or geography enters only through the §2 amendment process with a published rationale, and its addition must break nothing already published — no redirects, no renumbering, no orphaned certificates.

7. Forkability and continuity

The deepest guarantee of independence is that the institution can be continued without its stewards. The entire record — rankings, rules, evidence, registry, corrections, and the public key chain — is a plain-text Git history. If the stewards are ever captured, coerced, or lost, anyone may fork the repository with the full record intact, and legitimacy follows the honest fork. No steward, including the founders, can quietly retire or rewrite the record.

This guarantee is reinforced by permanence anchors: every quarterly snapshot is Git-tagged, pinned to IPFS, and submitted to the Internet Archive. The dataset cannot be silently withdrawn, even by the protocol itself.

8. Funding

The strongest endurance model for an unattended institution is to need almost nothing, and the architecture is built to cost almost nothing. If contributions are ever accepted, they are capped, their sources published to the smallest unit in a public ledger, and never acknowledged with placement, naming, ranking effect, or thanks. No contribution buys standing, and none ever will.

9. Records of governance

Steward terms, key-rotation ceremonies, amendments, and recusals are all recorded as committed, dated entries in this repository. Governance leaves the same kind of public trail the rankings do: it is auditable by construction.

This framework takes effect with the genesis record and binds every steward and every snapshot signed thereafter. It is amended only by its own §2 process.