Overview

Quality Report

How the records are checked, measured rather than asserted: fixed test suites with expected values, separate audit programs that recompute the numbers, and reviews by a model from another family.

78/78Common synthetic cases matched14 checks across 12 projects

6Harness audits passed24 gates in all

46Model review rows19 with a PASS verdict

0Auditor signaturesSign-off has not been run

Common synthetic suite

Each case has an expected value written before the run. A case counts as matched only when the observed value equals it.

Project	Checks	Cases	Matched
5-1 · AI drug discovery	2	11	11
8-1 · Synthetic data	1	5	5
18-2 · Medication safety	2	18	18
19-1 · Rare disease diagnosis	1	4	4
16-2 · Adaptive cognitive training	1	4	4
16-3 · Dementia diagnosis	1	5	5
18-1 · Fall early warning	1	6	6
18-3 · Surgical safety	1	5	5
12-1 · Acute crisis signals	1	5	5
12-2 · Safety-plan transitions	1	4	4
12-3 · Social isolation	1	5	5
12-4 · Student crisis handoffs	1	6	6

Open every case with its expected and observed value

Independent audit gates, by harness

An audit program recomputes the reported numbers without importing the solver. The module name tells the gates apart where a project has more than one.

Public-record audit · acmg19 · public_records_and_simulated_patients

19-1

G1 Frozen inputs and result file
G2 Ledger-derived claims pass
G3 False claims and ledger tampering blocked
G4 Literature findings verified and read twice

Figures come from ClinGen, ClinVar, gnomAD, HPO and published text, and from patients simulated out of those records. No patient data were used and no clinical performance was measured. No human auditor has signed, and this is not performance of the public call.

Synthetic harness · cog16 · synthetic_only

16-2 · 16-3

G1 Frozen inputs and submission structure
G2 Verdict definition, control completeness and structure rules
G3 Independent recalculation of rates, intervals and verdicts
G4 Virtual metric verdict and result hash

These are virtual metrics on synthetic cohorts. No ADNI, OASIS-3 or DementiaBank data were read, no real cohort has been evaluated, clinical validity has not been assessed and no auditor has signed the result.

Synthetic harness · dispense18 · synthetic_only

18-2

G1 Frozen inputs and submission structure
G2 Completeness of rules, fault transitions and the audit trail
G3 Independent recalculation of thresholds and benchmark verdicts
G4 Virtual metric verdict and result hash

These figures come from a synthetic gallery, simulated vision outputs and orders written in abstract codes. No pill image was analysed, no real prescription or DUR rule table was used, clinical performance was not assessed and no auditor has signed the result.

Synthetic harness · prevent12 · synthetic_only

12-1 · 12-2 · 12-3 · 12-4

G1 Frozen inputs and submission structure
G2 Schema, policy rules and transition completeness
G3 Independent recalculation of thresholds and metrics
G4 Virtual metric verdict and result hash

These are virtual metrics on synthetic inputs. No real cohort has been evaluated, clinical validity has not been assessed and no auditor has signed the result.

Synthetic replay · rare19 · synthetic_only

19-1

G1 Frozen inputs and submission structure
G2 Variant annotation and ACMG evidence attribution
G3 Synthetic reference replay and rank check
G4 Independent calculation and result hash

Scope: synthetic inputs and evidence-code review. Patient VCFs, clinical performance and auditor sign-off are not included.

Synthetic harness · synth8 · synthetic_only

8-1

G1 Frozen inputs and submission structure
G2 Schema, privacy ledger and rule tables
G3 Independent recalculation of metrics and bounds
G4 Known answers and release verdicts

These figures come from a synthetic multi-omics world generated by code. No real cohort (TCGA, GEO, MIMIC-IV) was evaluated, the main latent-diffusion generator was specified but not trained, and no auditor has signed the result.

Cross-family model review

A verdict covers only the evidence sent to the reviewer. UNCLEAR and FIX_REQUIRED rows are kept with the PASS rows that followed.

Date	Scope reviewed	Reviewer	Verdict
2026-11-14	19-1 phenotype score equivalence, the top-five boundary and synthetic ACMG evidence attribution	Claude Opus 5.5	PASS
2026-11-14	Registration of twelve projects, internal targets against the call and the scope of tests and sign-off	Claude Opus 5.5	PASS
2026-11-14	Four earlier Claude Haiku reviews of the same two scopes before corrections	Claude Haiku	2 UNCLEAR · 2 FIX_REQUIRED
2026-11-14	19-1 public time split, HPO inputs, metrics and ClinVar comparison	Claude Opus 5.5	PASS
2026-11-14	Shared registry and site check, first attempt; the response was cut off	Claude Opus 5.5	UNCLEAR
2026-11-14	Shared registry and site check, final	Claude Opus 5.5	PASS
2026-11-14	12-1 and 12-3 statistical guarantees, first review	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	12-2 and 12-4 safety invariants, first review	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	12-1 and 12-3 statistical guarantees, re-review after three fixes	azure/gpt-5.4-mini	PASS
2026-11-14	12-2 and 12-4 safety invariants, re-review after three fixes	azure/gpt-5.4-mini	PASS
2026-11-14	Site integration contract against its implementation: failed or unknown audits stop the build, nothing is written before every check passes	azure/gpt-5.4-mini	PASS
2026-11-14	IATHEON comparison record and the 12-1 to 12-4 summary wording, first review; no specific defect was named	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	IATHEON comparison record and the 12-1 to 12-4 summary wording, re-review after the scope of each section was stated	azure/gpt-5.4-mini	PASS
2026-11-14	Audit gate and rule table, first review	azure/gpt-5.4-mini	FIX_REQUIRED
2026-11-14	Audit gate and rule table, second review	azure/gpt-5.4-mini	PASS
2026-11-14	Literature step, first review	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	Literature step, second review	azure/gpt-5.4-mini	FIX_REQUIRED
2026-11-14	Move into APOSTOLUS, first review	azure/gpt-5.4-mini	FIX_REQUIRED
2026-11-14	Move into APOSTOLUS, second review	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	Move into APOSTOLUS, third review	azure/gpt-5.4-mini	PASS
2026-11-14	Site summary wording	azure/gpt-5.4-mini	PASS
2026-11-14	16-2 harness and specification, first review	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	16-3 harness and specification, first review	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	16-2 harness and specification, second review	azure/gpt-5.4-mini	FIX_REQUIRED
2026-11-14	16-3 harness and specification, second review	azure/gpt-5.4-mini	FIX_REQUIRED
2026-11-14	16-2 harness and specification, third review	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	16-3 harness and specification, third review	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	16-2 auditor and replay record	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	16-3 auditor, first review	azure/gpt-5.4-mini	FIX_REQUIRED
2026-11-14	16-3 auditor, second review	azure/gpt-5.4-mini	FIX_REQUIRED
2026-11-14	16-3 auditor, third review	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	16-2 abrupt-change detector variant	azure/gpt-5.4-mini	UNCLEAR
2026-11-14	18-2 specification and core modules, first review	azure/gpt-5.4-mini	2 FIX_REQUIRED
2026-11-14	18-2 gate mathematics, kernel, audit trail and rule engine, rounds before the final ones	azure/gpt-6.1-sol	8 FIX_REQUIRED
2026-11-14	18-2 Gate B and drift-monitor mathematics, final	azure/gpt-6.1-sol	PASS
2026-11-14	18-2 fail-closed kernel and audit trail, final	azure/gpt-6.1-sol	PASS
2026-11-14	18-2 DUR rule engine and MIMIC-IV mapping, final	azure/gpt-6.1-sol	PASS
2026-11-14	18-2 site summary wording against the audit file, final	azure/gpt-6.1-sol	PASS
2026-11-14	8-1 specification and core modules, first review	azure/gpt-5.4-mini	1 FIX_REQUIRED · 1 UNCLEAR
2026-11-14	8-1 differential-privacy core, rounds before the final one	azure/gpt-6.1-sol	3 FIX_REQUIRED
2026-11-14	8-1 differential-privacy core: sensitivities, accounting, AUROC bound, final	azure/gpt-6.1-sol	PASS
2026-11-14	8-1 preservation rule, privacy gate and measured results, earlier rounds	azure/gpt-6.1-sol	1 PASS · 1 FIX_REQUIRED
2026-11-14	8-1 preservation rule, privacy gate and measured results, final	azure/gpt-6.1-sol	PASS
2026-11-14	8-1 site summary wording against the audit file, first review	azure/gpt-6.1-sol	FIX_REQUIRED
2026-11-14	8-1 site summary wording against the audit file, final	azure/gpt-6.1-sol	PASS
2026-11-14	8-1 site summary after binding the baseline-ceiling sentence to the calibration evidence	azure/gpt-6.1-sol	PASS

Reviewers, costs and review records

What these checks do not establish

A matched synthetic case shows that code produced the value its author expected. A passing audit shows that a second implementation reached the same numbers from the same inputs. Neither measures performance on real patients, and neither replaces an external cohort or the auditor’s signature.