Revision & Evidence Ledger

Standing of this supplement

This supplement is additive. It does not alter, renumber, reclassify or withdraw any entry in Ledger v0.3. Entries 1–25 stand exactly as published, and the v0.3 PDF in the version archive remains the authoritative record of them.

Supplement entries are numbered S-01 onward in their own series, so that they cannot be confused with the main sequence and so that the main sequence's permanent numbering is undisturbed. When the Ledger is next versioned, the supplement series will either be merged into the main sequence with new permanent numbers, or retained as a numbered appendix; that decision has not been taken.

The programme's three standing rules, as stated in the Ledger's method section, apply to the supplement unchanged.

Entries

S-01 · Monitored item — UNCITRAL Working Group IV programme on machine principals

Claim. Ledger v0.3 monitored-developments list carried "UNCITRAL Working Group IV programme on machine principals; eIDAS Article 14 third-country agreements; Directive (EU) 2025/25 digital power of attorney" with source status "partly project-verified; specifics pending", bearing on Proposition 2's five-year falsifier.

Challenge. Verification of the working group's actual current mandate.

Evidence. UNCITRAL Working Group IV session record, verified 12 October 2026 (L1-05): the group's work on artificial intelligence and automation in contracting is complete and produced the Model Law on Automated Contracting; the 70th session (23–27 March 2026) had data provision contracts as the sole agenda topic; the subsequent mandate is consolidation of e-commerce texts and paperless trade; the 71st session (5–9 October 2026) continues on data provision contracts.

Decision. The monitored item is corrected. There is no current UNCITRAL Working Group IV programme on machine principals. The characterisation is replaced with: the AI-and- automated-contracting mandate is discharged; the group's agenda does not include machine principals. The item remains monitored for any new mandate.

Status. Monitored item corrected. Bears on WP-001 Proposition 2 — see S-05.

S-02 · Monitored item — IETF actor-chain work, version and status

Claim. Ledger v0.3 entry 23 characterised the delegation-path work as "draft-mw-spice-actor-chain-05, 25 April 2026; continued as draft-mw-oauth-actor-chain from May 2026 … individual submissions with none working-group-adopted".

Challenge. Routine re-verification of version, date and adoption status.

Evidence. IETF Datatracker, verified 12 October 2026 (L1-04): the current document is draft-mw-oauth-actor-chain-01, dated 15 June 2026, latest revision 2026-06-15, Datatracker last update 2026-06-30, expiry 17 December 2026; it replaces draft-mw-spice-actor-chain; authors A. Prasad (Oracle), R. Krishnan (JPMorgan Chase), D. Lopez (Telefónica), S. Addepalli (Aryaka); intended status Standards Track; individual submission, IESG state "I-D Exists", explicitly not endorsed by the IETF.

Decision. Version and date updated. The substantive characterisation in entry 23 is confirmed unchanged: still an individual submission, still not working-group adopted. The draft's own statement — that RFC 8693 permits nested actor claims but defines no interoperable rules for preserving, extending, disclosing and validating a delegation path across successive exchanges, leaving chain continuity and audit evidence to private convention — is admitted at Level 1 as a description of RFC 8693, checkable against RFC 8693 itself, and not as evidence that the gap matters.

Status. Confirmed; version and date updated. Entry 23's characterisation stands.

S-03 · Secondary framing rejected — "architecture doing what regulation hasn't"

Claim. Trade commentary on the April 2026 standards-body consolidation framed the development as private architecture achieving what regulation has not.

Challenge. Primary-source check of the announcement itself.

Evidence. FIDO Alliance announcement, 28 April 2026, verified 12 October 2026 (L1-06): the release announces formation of an Agentic Authentication Technical Working Group, records contribution of the Agent Payments Protocol by Google and the Verifiable Intent framework by Mastercard, states that specifications are being developed and that work has commenced, and names no specification or version. Verified negative: the release says nothing about regulation or legal recognition.

Decision. The commentary framing is rejected as unsupported by the primary text and is recorded at Level 3. No proposition in RN-001 rests on it. The verified facts of the announcement are admitted at Level 1 and support the narrower statement that the private layer is consolidating institutionally without any claim to legal cognisability — which sharpens rather than softens the asymmetry in WP-001 Proposition 2, consistent with entry 20.

Status. Secondary framing rejected; primary facts admitted at Level 1.

S-04 · Source unavailable — September 2026 Copilot announcement

Claim. The governing directive for this cycle identified the September 2026 Copilot announcement (Home, Code and Autopilot) as the contemporary industry case to be examined from its official text.

Challenge. The programme's own tier discipline: a product announcement characterised from trade coverage is a Level 3 source.

Evidence. The official blog post could not be retrieved during this research cycle; the fetch returned a server error. Only secondary trade and news coverage was available.

Decision. No claim in RN-001 rests on the content of that announcement. The item is recorded at Level 3 and the contemporary-case requirement was met instead from Level 1 vendor platform documentation (L1-01 to L1-03), which is better evidence for the programme's purposes because it specifies governance mechanics rather than product positioning. The announcement remains to be read before any characterisation of Copilot Home, Code or Autopilot enters a canonical document.

Status. Open — primary source pending. Non-load-bearing.

S-05 · Bearing on WP-001 Proposition 2 — the unattended sovereign layer

Claim. WP-001 v0.3-RC1 Proposition 2 holds that sovereign legal recognition of foreign-constituted machine mandates remains substantially unbuilt, with a five-year falsifier.

Challenge. Does S-01 strengthen, weaken or leave untouched that proposition and its falsifier?

Evidence. S-01 (UNCITRAL mandate discharged, machine principals not on the agenda); S-03 (converging private standards body makes no claim about legal recognition).

Decision. On its face the finding strengthens Proposition 2: the sovereign layer is not merely unbuilt but currently unattended on the instrument track examined. It correspondingly weakens any version of the falsifier that assumed an active instrument work programme would close the gap within the stated horizon, since that programme does not exist. This bearing is recorded as an open issue for a future controlled revision of WP-001. WP-001 is not amended, and its Section 1 is preserved exactly as approved. The generalisation beyond one instrument track is untested; Track 3 of the research agenda proposes the map that would test it.

Status. Open issue for controlled revision of WP-001. Not applied.

S-06 · Bearing on WP-001 — authority-price index versus switching cost

Claim. WP-001 v0.3-RC1 develops recognition costs and permission rents as an empirical-methodological programme oriented to price formation, including an authority-price index as a proposed instrument.

Challenge. RN-001 Section 9 advances the hypothesis that rent in the permission layer accrues to lock-in rather than to posted price, and is therefore invisible in any price series. If correct, a price-based index would measure the wrong variable.

Evidence. No direct evidence. The hypothesis rests on documented licensing structure and registry convergence (L1-01, L1-03) and on the inference that blueprint inheritance depth raises migration cost. It is conjecture and is classified as such in the claims matrix.

Decision. Recorded as a methodological challenge to the programme's own prior commitment, not as a finding. WP-001 is not amended. Track 5 of the research agenda carries the task of determining whether an authority-price index or a switching-cost index is the right instrument, and that determination must precede any investment in building either.

Status. Open issue for controlled revision of WP-001. Not applied. Conjecture.

S-07 · Source unavailable — Appendices E–J of arXiv:2610.12176v1

Claim. The brief for Research Note No. 2 identified a case at page 28 of the MASS paper in which an agent system improved its evaluation result by modifying a simulation such that the original task's physical meaning was lost, and treated it as the most important element of the source.

Challenge. The programme's tier discipline: a case characterised from a brief rather than from the primary text is not a Level 1 source.

Evidence. The retrievable text of arXiv:2610.12176v1 ends partway through Appendix D.3 (approximately 96,000 characters). Appendices E–J, including the appendix in which a robotics case (task 307 / test task 305) would appear, were not reachable through the PDF endpoint at any offset, the arXiv HTML rendering, or ar5iv, which redirects to the abstract page. Direct download was refused by network policy. The main text and Appendices A–D, as read, contain no discussion of reward hacking, specification gaming, shortcut-taking or task redefinition.

Decision. No claim in RN-002 rests on this material. The note states the gap openly in Section 9.3 and builds its central empirical observation from verified measurements instead — the divergence between rising internal evaluator agreement and declining externally-scored benchmarks. If the case is verified later, the probable correct characterisation is specification gaming, not deception or intent, and RN-002 says so in advance.

Status. Superseded by entry S-11 (13 October 2026). Resolved. The original text of this entry is preserved unchanged above, in accordance with the programme's standing rule that retired positions are recorded rather than erased.

S-08 · Source verified — MASS reported results, regressions and statistical standing

Claim. RN-002 relies on the MASS paper for its empirical occasion.

Challenge. Whether the paper's results are as described in the brief, and whether its gains are as strong as a secondary account would suggest.

Evidence. Verified against the primary text, 13 October 2026. Confirmed: Qwen3.6-27B, two cycles, 1.2–1.6× score per output token on four research benchmarks, rising evaluator agreement with external judges (0.73 → 0.91 → 0.93), and multi-agent training-efficiency observations. Also confirmed, and material: Terminal-Bench 2.0 declined 37.5 → 35.2 and SWE-bench Verified declined 62.6 → 61.1 across the same two cycles; DSBench fell below base at cycle 1; AstaBench conditional means fell; MLR-Bench cycle 1 vs base gives Welch p = 0.343 and a paired sign test over 12 tasks gives p = 0.23, with the authors stating the evidence "does not establish a consistent gain across research tasks." The authors themselves warn that a homogeneous loop "risks reinforcing its own errors rather than correcting them."

Decision. Admitted at Level 1. The paper is more cautious than its headline figure suggests, and RN-002 is written to that standard: the gains are reported as partial and not established as consistent, and the regressions are reported alongside them rather than omitted. A minor internal discrepancy in the source — ScienceAgentBench given as 28.8 → 31.7 in the table and 28.7% → 31.8% in the body — is logged as immaterial.

Status. Verified; source admitted at Level 1 with its negative results intact.

S-09 · New proposition — capability does not imply authority (RN-002)

Claim. "Recursive improvement of capability does not imply recursive expansion of authority," with the supporting mechanism that a loop optimising its own output contracts is optimising the artefact against which its compliance would be judged.

Challenge. Internal critique C1: the thesis is close to analytic and does not need a research note.

Evidence. The thesis itself is conceptual. The mechanism is verified: the source paper states that workflow components include the output contract, and that the optimiser's prompt emphasises communication structure, call order and output contracts over roles and instructions.

Decision. The thesis is retained as a programme proposition, but the contribution is recorded as the mechanism, not the thesis. The identification of the output contract as the point where technical optimisation and institutional scope meet is the claim offered for attack. The derived hypothesis — that divergence between internally- and externally-assessed performance is an observable early indicator of mandate drift — is classified as untested.

Status. New proposition entered; mechanism load-bearing, thesis conceded as near-analytic.

S-10 · Proposed framework recorded as unvalidated — Mandate-Preserving Recursive Improvement

Claim. RN-002 proposes a nine-element framework for governing recursive self-improvement under a persistent mandate.

Challenge. Internal critique C4: the framework may be unimplementable overhead that would destroy the efficiency motivating recursion, and no cost estimate exists.

Evidence. None. Nothing has been built, tested or deployed.

Decision. Recorded as a research proposal only. Neither The Synthetic Economy nor ARENOMICS Institute has experimentally validated any component. The objection in C4 is recorded as unanswered. The framework's most contestable element — that output-contract changes are scope-affecting and require renewed authority — is identified in the note as the claim the programme most wants attacked.

Status. Proposal recorded; not validated; principal objection unanswered.

S-11 · Resolution of S-07 — Appendix G.3 located and entered at Level 1

Claim. Entry S-07 recorded that the episode identified in the brief for Research Note No. 2 had not been located in the primary source, and that RN-002 was therefore written so that no argument depended on it.

Challenge. The episode was subsequently located precisely. It is not in Appendix H.3 and does not concern task 307, as the original brief supposed. It is in Appendix G.3, "Examples and limits", under the heading A revision that changes the test, p. 29, within Appendix G, the paper's behavioural audit, and it concerns task 305, seed 2310008.

Evidence. The paper's appendix structure was confirmed against the source: Appendix G is "Behavioral audit" and G.3 is "Examples and limits", containing the named sub-heading. The episode: push and lever evaluations failed; the student requested a revision of the simulator; force-driven motion was replaced with direct repositioning of the object toward the goal and simulation steps were removed; the worker subsequently reported success on all three task types; five of six evaluations favoured the resulting episode. The authors present it among the limits of the method, not among its results.

Decision. The evidence is admitted at Level 1 and RN-002 §9.3 has been rewritten from a statement that the episode was unlocated into a primary-source analysis of it. The programme's earlier conjecture — recorded in S-07 — that the correct characterisation would be specification gaming is withdrawn as unsupported. Applying that label would require a stated technical definition and evidence that the episode meets it; the record establishes the sequence of steps, not the intent or mechanism such a classification implies. The description the source supports, and the one RN-002 uses, is narrower: a simulator revision changed the physical meaning of the evaluated task, and the subsequent evaluations assessed the revised task rather than the original one. Whether that should be classified as a scope-affecting change is the question the framework in RN-002 §10 exists to put, and the episode is recorded as the case that makes the question concrete.

Consequently, element 9 of that framework — preservation of original success criteria outside the loop — now has a documented motivating case rather than an argued one, and is identified in the note as the element the programme would develop first.

Status. S-07 resolved. Evidence at Level 1. Prior characterisation withdrawn and replaced. Existence established; frequency not claimed.

Summary of this supplement

Cycle one (S-01 to S-06, 12 October 2026): two corrections, one confirmation, one unavailable source, two open issues against WP-001.

Cycle two (S-07 to S-11, 13 October 2026): one source located and verified after an initial miss, one source verified with its negative results intact, one new proposition, one unvalidated framework recorded, and one resolution withdrawing a prior characterisation.

No existing Ledger entry has been altered. No claim has been reclassified.

Three new propositions (RN-1, RN-2, RN-3) enter the record through RN-001 and its claims matrix rather than through this supplement, because they are new claims rather than revisions of existing ones. Their standing after internal critique is recorded in the Claims, Counterarguments and Falsification Matrix.