Public proof ledger

Every claim arrives with the way to defeat it.

This page is a bounded public rendering of the estate’s claim registry. Measurements, rulings, scope, exclusions, producers, and falsifiers remain distinct.

Composite evidence snapshot · 09 August 20264 pinned sources

Public claim registry: RMU_Yig_Tower/RMU_PARALLEL/CLAIMS.json
SHA-256 5d0c2ea05c07ec25…
Current mutable tower index: RMU_Yig_Tower/RMU_PARALLEL/INDEX.json
SHA-256 f62de3f4a74e44ed…
AD-1 submission manifest: Armada_AD_Submission/AUDIT_SUBMISSION_MANIFEST.json
SHA-256 5284e8b220ed65de…
Owner-relayed first determination record: RMU_Yig_Tower/RMU_PLAN_AMENDMENTS/AD1_FIRST_DETERMINATION_20260809T033603Z/RECORD.json
SHA-256 9a59f0018ed73d18…

Read this first

Standing boundary.

These limits control every favorable statement below.

The system has not been independently verified. Its results were produced by its author and checked by author-operated instruments; external review remains open.
The system is tamper-evident, not tamper-proof. Detection is not prevention, and an actor able to rewrite artifacts can also rewrite adjacent records.
No performance comparison is claimed. Speed is a design concern, not a measured advantage in this record.
No component has completed qualification. Clean-room rebuild, atomic publication, and an externally held signature remain unmet; the qualified composition refuses every request.
Conformance has been observed on one host. Cross-host agreement remains unmeasured.
The first Audit Director headline was relayed by the owner as PASS-WITH-QUALIFICATION for custody and governance only. Mathematics was excluded, writer/reviewer role overlap was disclosed, and the director’s written reasoning remains owed.

MEASUREMENT01 · engine.verification

Whole-engine verification

One command runs every layer’s tests and adversary suites, re-measures each layer’s identity from disk, and reports a single verdict.

Scope: The nine layers under RMU_PARALLEL/development_layers on the host where the command is run.
Evidence: RMU_PARALLEL/verify_engine.py and each layer’s BUILD_RESULT record
Producer: python verify_engine.py
Expected result: VERDICT: CLEAN
What defeats it: The command reports NOT CLEAN, a measured layer identity differs from its manifest, or any suite reports a nonzero failure count.

What this does not establish

It says nothing about hosts where it has not been run.
It establishes suite agreement and recorded identity, not mathematical correctness by itself.
It covers no artifact outside development_layers/.

MEASUREMENT02 · spectral.flagship

Exact spectral reproduction

The engine recomputes the Harper operator’s band structure over ten filed flux columns and lands on the recorded figures: 111 band edges and 54 bands on every column.

Scope: α = 1/q for q in {1,2,3,4,6} and {5,7,8,9,12}, phase zero, with exact integers and rationals and no floating point in a deciding path.
Evidence: LAYER_007_SPECTRAL/tests/battery.py
Producer: python development_layers/LAYER_007_SPECTRAL/tests/battery.py
Expected result: 0 failed
What defeats it: Any filed column produces a different edge or band count, or a transfer product’s determinant differs from exactly one.

What this does not establish

It covers those ten columns and no other flux.
Agreement with a prior record is not a proof that either result is mathematically correct.
The prior figures were produced by the same author, making this a reproduction rather than independent corroboration.
Current review found PSI[9] and X0_INTERVAL[8] unconstrained: mutating either moves the spectrum while the 172-test battery and 15 adversary cases remain green. The filed counts therefore do not presently validate those inputs.

RULING03 · exact.or.refuse

Exact or refuse

A computation emits a supported result or a typed refusal, and never an approximation presented as a result.

Scope: The design law applied across the live development layers; a refusal names what was refused and why.
Evidence: RMU_PARALLEL/SPINE_CHARTER.md and refusal classes in each layer
What defeats it: A deciding path returns a rounded, truncated, or approximated value without producing a typed refusal.

What this does not establish

A ruling is a design decision, not evidence that the implementation is defect-free.
It does not establish that every refusal is well worded; refusal quality is measured separately.

MEASUREMENT04 · refusal.quality

Refusals are audited

Every refusal site in the engine carries text long enough to act on, measured by a static pass across the refusal surface.

Scope: All refusal-class raise sites across the nine layers and their instruments.
Evidence: RMU_PARALLEL/instruments/refusal_audit.py
Producer: python instruments/refusal_audit.py
Expected result: 0 thin
What defeats it: The audit reports any refusal site below the filed static threshold.

What this does not establish

Length is a proxy for actionability, not actionability itself; sufficiently long but useless text can pass.
A remedy-keyword heuristic is reported but decides nothing.

MEASUREMENT05 · conformance.singlehost

Independent-route conformance

Sixteen pinned vectors spanning every layer agree between the engine and a separate computation route that shares no code with it.

Scope: The reference host, at the source digest recorded in the conformance record.
Evidence: LAYER_009_CONFORMANCE and its emitted conformance record
Producer: python development_layers/LAYER_009_CONFORMANCE/tests/battery.py
Expected result: 0 failed
What defeats it: Any vector diverges between the two routes or the kit reports a verdict other than CONFORMS.

What this does not establish

It has run on one host; agreement elsewhere is unmeasured.
Agreement between two routes is not proof of correctness.
It covers exactly the pinned vectors and no other input.

MEASUREMENT06 · audit.submission-integrity

Audit submission integrity

The first AD-1 submission envelope contains seven declared artifacts whose measured SHA-256 values and byte counts match its manifest.

Scope: The immutable files listed by Armada_AD_Submission/AUDIT_SUBMISSION_MANIFEST.json at the inspected generation.
Evidence: Armada_AD_Submission/AUDIT_SUBMISSION_MANIFEST.json and its seven listed artifacts
Producer: SHA-256 and byte-count comparison against the submission manifest
Expected result: 7/7 hashes match · 7/7 byte counts match
What defeats it: Any listed artifact is absent, its measured digest or byte count differs, or the scope is represented as covering mathematics.

What this does not establish

Manifest agreement establishes custody of the submitted bytes, not truth of their contents.
The submission covers custody and governance only; mathematics, novelty, and product qualification are excluded.
Claude prepared both the change report and review report because no Codex writer lane ran for the work under review.
The favorable Audit Director headline entered the estate through the owner’s relay; the director’s written statement of limitations is still owed.

RULING07 · failures.preserved

Failure remains inspectable

Failed work is retained with its reasons; a correction arrives as a new record naming what it supersedes.

Scope: The estate’s append-only discipline and supersession practice.
Evidence: Frozen failure, poisoned-branch, plateau, and no-standing entries retained in the candidate registry
What defeats it: A failed artifact is deleted rather than retained, or a correction silently edits a filed record in place.

What this does not establish

Retention does not guarantee every failure was recognized when it occurred.
It does not establish that every retained record is complete.

External standing remains open

A better challenger is useful.

Independent review is not implied by internal agreement. Reviewers are invited to reproduce, refute, narrow, or identify a missing falsification surface.

Enter as a reviewer