Independent reproduction lane: content and resource semantics

25 September 2026. This reviewer did not author the funded-audit runtime, shared mathematical models, export builder or verifier. It previously drafted domain teaching cases and the positioning comparison, so it is not independent of the entire packet. This is an additional agent lane within the same orchestration; it is not external human review or a separate institutional approval.

Qualified content assent

Assent to including the funded-audit extension and the bounded release claims in a curated public research seed. No resource-accounting or interpretation blocker was found in the inspected ordinary demonstration envelope. Exact archive reproduction remains a separate post-freeze check; this document does not assert that a not-yet-frozen ZIP has passed.

The extension makes a useful concrete advance: the earlier coupled agents acted on an exogenous perceived audit probability, whereas this model reserves hours, funds inspections and completion rewards, executes selected participating tasks, and moves sanctioned collateral into a separate account. The synthesis and claim map correctly keep those engines separate rather than treating the extension as proof that the entire protocol’s institutions now exist.

Independent accounting reconstruction

The reviewer ran 3,060 configuration/seed cases spanning populations 2, 4, 8 and 16; every quota from zero to the population; all three declared role modes; concealed or disclosed selection; review/audit hour costs .1, .5 and 1; and five seeds. Heterogeneous effort and outside-option vectors induced participation and refusal. Of these cases, 2,160 executed and 900 correctly stopped before offering because positive-quota self-audit was unavailable.

Without calling the model’s utility or ledger helpers, the reviewer reconstructed rational expected utilities from the declared inputs, selected the best response with the documented tie order, and reconstructed person-hour debits and token transfers from actor outputs. Checks covered selected participating audit slots, unused refused slots, no executed self-audit, exact used/reserved/capacity ordering, reward funding, audit cost, escrow conservation, and stage denominators. All these checks passed. funded-accounting-sweep.json records dimensions and the inspected runtime source hash. This independently samples execution paths; it is not an exhaustive proof over every representable configuration or detection draw.

Resource and information distinctions that must survive publication

• A task decision unit is not necessarily a distinct human. Several units share one execution person. The model expressly assumes separable utility and does not solve that person’s portfolio decision or hidden common control.
• Utility effort is not booked human time. Both honest and shallow completions spend the same review slot; private effort differences do not create free capacity. The manuscript makes this separation explicit.
• Whole-program worst-case reservations are conservative. Shared roles can reserve twelve hours while executing ten, preserving the unconditioned quota lottery. A draw that happens to fit a smaller capacity does not justify silently rerolling or conditioning on that capacity.
• A selected refusal consumes no audit slot and no review time in this model. That is an explicit omission of real refusal/triage labor, not an empirical finding that refusal is free. The selected, executed and unused denominators make the distinction inspectable.
• The conditional participating-actor audit marginal needs hidden selection and independence from participation. Disclosure changes the actor’s information; all actions commit before outcomes in the main branch. The final evaluator trace is not a cryptographic concealment mechanism.
• Distinct person labels, escrow availability, automatic auditor effort, detection quality and legitimate loss collection are stipulated. The toy transfers cannot establish independent governance, actual wealth, consent or legal enforceability.

These restrictions are already present in the funded manuscript and synthesis. They are substantive premises, not peripheral disclaimers. The claim ledger’s bounded C09–C10 formulations are consistent with them.

Remaining numerical boundary

A separate extreme-input probe used the accepted finite configuration value actor_escrow=1e308. Conversion of the aggregate exact escrow to a displayed float raises OverflowError: integer division result too large for a float. No successful overdraw or fabricated balanced report results. This is outside the small teaching values used in the retained runs, but means finite scalar validation alone is not a total runtime guarantee for every finite configuration. A future generalized API should reject unsafe output ranges or serialize exact totals. This is not a blocker for the explicitly bounded research fixtures; do not advertise arbitrary finite-range robustness.

Artifact and publication claims

The core reproduction guide separates standard-library checks from optional Pandoc, Matplotlib and PDF dependencies. The verifier checks exact inventories, content hashes, selected local links and a limited set of disclosure patterns. The guide does not represent those checks as comprehensive privacy, security, rights or scientific validation. A hash binds a result to bytes, not to consent, truth or publication authority.

The collective report correctly leaves the final ZIP digest and clean-extraction result to an external release record, avoiding a self-referential claim inside the archive. This reviewer has not independently performed visual PDF inspection, tested every optional renderer, or established dissemination eligibility.

The readiness conclusion remains positive and conditional: this is suitable material for public criticism as a protocol seed, with accountable attribution, rights disposition and exact release decisions still supplied by the maintainer. It does not promise an operating scientific review, rights-handling or dispute service. The fresh archive check should report its digest and results separately.