Verify rather than trust
Check it yourself. Nothing here asks you to take our word.
A system whose thesis is that self-attested state is ungovernable cannot ask you to accept its own self-attestation. So this page is the list of things a skeptic can run — including the checks that return an uncomfortable answer.
01 — The record itself
The corpus, schemas, and conformance workflow are public
The platform repository
The governed corpus, the atom schemas, the tools, and the workflow that gates them.
github.com/tecthulhu/republic ↗The founding enactment
DEC-0001 landed by signed merge under its own ratification machinery. The tag marks the commit where the substrate began governing itself.
tag: dec-0001-enacted ↗Evidence records
Every control invocation writes an evidence record. They are committed, not stashed in an expiring CI artifact — a row that lives only in an artifact is provenance with a countdown timer.
platform/acta/ ↗Conformance runs
The control suite runs on every change and uploads its evidence rows publicly, so you can check the claim rather than trust the badge.
Actions history ↗02 — The load-bearing claim
“A red suite blocks merge” is evidenced separately, because it is two different facts
A CI job that detects a violation is a control. A repository rule that refuses the merge when that control is red is enforcement. Most projects claim the second while demonstrating only the first.
CTRL-0009 captures enforcement from the live branch-protection setting on every run — never from a constant — and writes it to platform/acta/ as EVID-ctrl0009-*. You can re-run it, or skip the repository entirely and query the setting directly with no credential at all:
curl -s https://api.github.com/repos/tecthulhu/republic/rules/branches/mainNo token, no account, no clone. What it returns today: both conformance contexts required by name, direct push and force-push refused, and no bypass actor — administrators included.
python platform/suite/enforcement/run.pyOr re-run the control yourself, from a clone of the repository.
Two qualifications ride with that claim
Branches are not required to be up to date with the base before merging. A green check therefore describes what the pull request tested, rather than always describing what lands.
No approving review is required. The merge is the owner’s act alone.
Both are recorded in the evidence rather than omitted from the claim. They are printed here for the same reason: a verification page that lists only the checks that pass has become a marketing page.
03 — Re-runnable measurements
The standing queries run against any commit
Coverage is a query, not an audit. “Is every requirement bound to enforcement?” returns a number, and you can compute that number yourself against any point in the history.
python platform/tools/embedder.py --reportRun from the tools directory against any commit. The standing query for rules without passing evidence moved from 22 to 8 across the spawn-contract work — a figure the query will confirm or contradict, rather than one you have to accept.
Enforcement effectiveness
Measured separately from coverage, deliberately. A rule can be bound to a control that runs perfectly while measuring the wrong property. Coverage is computable; validity is graded.
Evidence completeness
Structural — no control can execute without leaving a mark. “Is this claim currently evidenced” resolves to the latest passing evidence against the current subject digest.
Dangling claims
A specification no rule binds is a dangling claim. It is a standing query with a launch-blocking target of zero, not a review item someone might notice.
04 — The uncomfortable part
The system is built so that not working would show
A system that can only display itself succeeding is not testable. This is the check worth applying to any governance product, including this one: can it emit evidence of its own failure?
Two declared postures composed into an attack
The spawn gate requires a story that is resolvable, not ratified — necessarily so, since requiring active law to spawn the work that earns it would be a circular gate. Separately, agent authorship was not yet bounded by any mandate check.
An external reviewer found that the two compose: an agent could author a weakening of its own in-flight acceptance criteria, and the gate would grade against the weakened set — reintroducing the closed loop the whitepaper’s §3 names, inside the governance plane itself.
The attack was findable only because both postures were declared. Resolvable disclosure recruited the adversary into the security analysis. The mitigation is ruled and time-boxed; the episode is published as evidence rather than confession.
The evidence machinery catches its own defects
Adversarial “rogue” fixtures caught two real verifier defects before they shipped.
The supervision proof was built to assert truncation, not merely survival — because a buffering proxy would otherwise have faked the entire capability. Adjacent capability is not the required capability: a chain in which every hop is verified except one never-walked step is an unverified chain.
The suite’s own source declines to fake what it cannot prove locally, noting in a code comment that a surrogate claiming supervision coverage “would be the exact overclaim” the specification forbids.
05 — Licensing
The repository is public for inspection, not yet licensed for reuse
The outbound license is itself a proposed decision atom. DEC-0002 names AGPL-3.0-only, granted personally by the owner, and stands at proposed with no LICENSE file committed. Until it is enacted, the deliberate holding posture is all rights reserved.
That is stated here rather than left to inference, because a repository that is public and unlicensed is easy to mistake for one that is open source — and the difference matters to anyone deciding whether to build on it.