Every legal standard this system almost satisfies — and the one missing signature for each

Nothing in this build is a new idea. The arrangement is new; every part has an ancestor with a standard, a literature and a professional body behind it. This page maps each field to the frame it belongs to, and says in each row what is not

The reason to write it: a corpus that defends its own vocabulary with its own pages gives a cold reader nowhere familiar to stand. Roughly four hundred of this site's articles are about its own protocol. Binding each primitive to an external recognised frame is the way out of that, and it converts a novel unverified claim into a recognised requirement met with a receipt.

Evidence law: the findings are 902(14)-shaped by construction

frame	the field here	what satisfies it	what does NOT
FRE 902(13) — certified record of a regulularly conducted activity	the invocation receipt	every invocation writes id, timestamp, actor, object, input and output fingerprints and provider status, automatically, as a regular activity of the system	no custodian declaration has been drafted or signed
FRE 902(14) — certified data copied from an electronic device, authenticated by hash	artifact SHA-256, record SHA-256, rule-set hash, chain head, anchor id	hashes are published before deliberation and recomputable by anyone; the offline verifier rehashes every object	the qualified-person certification that makes the self-authentication operative is not written
FRCP 37(e) — failure to preserve ESI	RECORDS_ABSENT	absence is a positive assertion made before anyone knew there would be a dispute, naming the records a competent reviewer would have expected	nothing here proves the absent record ever existed
eIDAS Art. 41 — qualified electronic timestamp	the anchor packet	drand + Bitcoin give cryptographic anteriority	a qualified timestamp carries a legal presumption of time and integrity, and none is obtained; this is the largest missing item in this table

Audit standards: RECORDS_ABSENT is a mechanised scope limitation

frame	the field here	what satisfies it	what does NOT
ISA 500 — sufficient appropriate audit evidence	verdict tiers and the span obligation	a finding must quote the shortest verbatim span that carries it, or return SPAN: NONE	no evidence-sufficiency threshold is defined per rule set
ISA 705 — modified opinions arising from scope limitation	RECORDS_ABSENT + CANNOT_CONCLUDE	the instrument's analogue of a qualified opinion: name what was unavailable and decline to conclude on it	there is no equivalent of an adverse opinion, and no materiality concept
RADPEER-style imaging peer review	the panel plus the recorded adversary	blinded independent readings with disagreement recorded rather than resolved by seniority	no human reader has participated in any panel

EU AI Act: the clause-by-clause table, with what is not satisfied

Article 12 and Article 14 were supplied verbatim to a panel under the external-statutory rule set, hashed at 3e3bcd8d47d14c159f9d9303438b0839… before the run. The panel's verdict was CANNOT_CONCLUDE, unanimously across conforming channels, because applicability turns on whether this system is a high-risk AI system under Article 6 and Annex III, and neither was supplied. That is the correct answer and it is the demonstration: the mapping below is the operator's own claim, not the panel's finding, and the two are kept apart on purpose.

provision	what it requires	the surface here	what is NOT satisfied
Art. 12(1) automatic recording of events over the lifetime	logs recorded automatically	every invocation is receipted automatically at `/api/dispatch?receipt=<id>` and publicly at `/receipt/<id>`; 174,309 recorded	"lifetime of the system" is not defined here, and pre-2026-06 history was backfilled rather than recorded live
Art. 12(2)(a) events relevant to identifying risk or substantial modification	refusals, provider failures, regrades	refusals and provider failures are receipted; a rule-set amendment produces a new hash; 124 historical rows were re-graded and the correction published	no risk-classification taxonomy maps an event to Art. 79(1) risk
Art. 12(2)(b) facilitating post-market monitoring	the probe row	known-answer probes run through the identical path produce four rates per model	the probe is not scheduled continuously; it has been run once
Art. 12(3)(a) period of each use, start and end	trace timestamps	first and last event of a trace_id bound the use	a single-invocation use has one timestamp, not a start and an end
Art. 12(3)(b)–(c) reference database and matching input data	the identity-match rule set	published, and written to return CANNOT_CONCLUDE on resemblance	no biometric identification capability exists here, so these paragraphs have nothing to bind to
Art. 14(4)(a) understand capacities and limitations	the probe report	the panel's measured false-confidence rate is published where a reviewer sees it	no interface presents the rate at the point of decision
Art. 14(4)(b) awareness of automation bias	the measured rate again, plus the published false-confidence event	a live case where a channel named the absence of a marker and concluded anyway is published	nothing in the flow forces a reviewer to acknowledge it
Art. 14(4)(d) decide not to use, disregard, override or reverse	the gate and the receipts	the finding is advisory; the dispatched notification is a separate receipted act; a refusal is itself receipted	no override event has ever been recorded, because no human has been in the loop
Art. 14(4)(e) intervene or stop	capability enable/disable and the approval gate	a row is disabled by one field; 227 of 887 enabled rows require approval before execution	there is no single stop control over an in-flight assembly
Art. 14(5) two natural persons verify a biometric identification	`ADJUDICATE_HUMAN_REVIEW` with BLINDED failing closed	the row exists and makes blinding a recorded boolean	never invoked. No named human has returned a finding. This is the clearest unmet item in the table

Risk-management frameworks

frame	the surface here	what is NOT satisfied
NIST AI RMF — MEASURE	four rates per model per rule set on a stratified suite published at a hash	fourteen items is too small to characterise anything beyond this suite
NIST AI RMF — MANAGE	the deterministic gate, the escalation path, the sensitivity regrade on six capability rows	no residual-risk statement, and the escalation terminates at a role rather than a person
ISO/IEC 42001	rule sets as versioned objects, an objections ledger, a change record with hashes	no management-system documentation, no internal audit programme, no certification
IEC 61508 common-cause failure	verdict correlation by training family, measured at 0.893 same-family against 0.714 cross-family, and a gate that counts families rather than seats	no systematic-capability argument, no proof-test interval

The intellectual ancestry, named

ancestor	the correspondence	what is new here
Toulmin (1958) — claim, grounds, warrant, qualifier, rebuttal	claim → the assertion; grounds → the quoted span; warrant → the rule-set clause; qualifier → tier and verdict; rebuttal → the published falsifier and the recorded adversary	the warrant is content-addressed, so it cannot be substituted after the objection — sixty years of argument about warrant identification, settled by a hash
Preregistration	the rule set is pinned as bytes and anchored before the artifact is judged	this is preregistration applied to machine judgment, and it has no analogue in AI evaluation
Adversarial collaboration	two parties pre-commit to the rules that would settle a disagreement	the machinery operationalises it; it has not been run with a real second party, which is the honest state of that claim
Chow's reject option; conformal risk control	abstention in exchange for a distribution-free error bound	the bound is not computed; what exists is a measured escalation behaviour
Knight & Leveson (1986)	independently developed programs to one spec fail together	the same result, sharper, because channels share training data
Double reading with arbitration in population screening	independent readers, disagreement to arbitration, measured sensitivity gain	the readers are models and the arbitration trigger is a published deterministic function
Merkle trees; OpenTimestamps	hash-linked history with external inclusion proofs	the chain is sealed and anchored; no OTS proof exists, and `chain_inclusion` still reads NOT_YET_PROVEN_INCLUDED
Macaroons	caveat-bearing, attenuable, audience-bound capabilities	witness tokens are audience-bound and fail closed when forwarded
HATEOAS	representations carry their own next actions	every capability row publishes its own read, patch, invoke and skill routes

What is claimed as contribution, precisely: not any of these frames. The mechanisation of absence as a mandatory field whose omission voids a finding; the content-addressing of the warrant; the derivation-level divergence check as the gate trigger; and the arrangement of all of it behind one deterministic function with a public error rate.

References (3)

The system

Messaging

Ads and leads

Devices