Reviewer-agents pass — task-3b13 CEP/Stat-summary MAN_2025 alignment

Internal artefact. Three reviewer agents fired in parallel on 2026-04-21 after the CEP and R-TF-015-013 factual-alignment edits were applied. This note records the findings, the verdict on each, and the follow-up edits applied.

Agents invoked

• bsi-clinical-auditor — Erin/Nick voice; regulatory-substance and MDCG/MEDDEV/ISO citation integrity.
• audit-deliverable-reviewer — PDF-leak gate; confirms no internal-engineering or BSI-workflow leaks in audit-visible files.
• celine-clinical-consultant — pillar-mapping, Rank-vs-Pillar orthogonality, indirect-benefit causal-chain integrity (Horiana/Ugidos framework).

All three agents scoped to R-TF-015-001 (CEP) and r-tf-015-013-statistical-summary.mdx.

Findings — summary

Critical (2 raised, 1 confirmed, 1 dismissed); Warnings (11 raised, 6 accepted, 5 noted); Positive observations across all three agents.

Critical C1 — CEP Appendix-III Rank 11 row, stale reader count (bsi-clinical-auditor)

File:line: R-TF-015-001-Clinical-Evaluation-Plan.mdx:1275. Quote (before): MRMC studies: ... MAN_2025 (≥5 dermatologists × 149 images, Fitzpatrick V-VI). Verdict: Confirmed — internal self-contradiction with the Confirmatory-Phase row at line 1212 (16 HCPs) and with the R-TF-015-013 §MAN_2025 body (16 HCPs). An auditor cross-checking Appendix-III rank membership against the main Confirmatory-Phase table would flag this inside 30 seconds. Fix applied: row updated to MAN_2025 (16 HCPs × 149 images in the primary analysis cohort; 19 enrolled with 3 documented screen failures, data-lock 17 April 2026; Fitzpatrick V–VI).

Critical C2 — CER Clinical Investigations table, "In preparation" (bsi-clinical-auditor)

File:line: R-TF-015-003-Clinical-Evaluation-Report.mdx:1837, Publication Status column. Quote: In preparation. Verdict: Dismissed — the agent misread the column. The column header is "Publication Status" and tracks manuscript / journal publication status, not CIR status. Every other row in the same column uses the same convention (MCEVCDAO_2019: "Under review", IDEI_2023 and BI_2024: "Published (...)", COVIDX / DAO_O / DAO_PH / PH_2024 / SAN_2024 / AIHS4_2025: "In preparation"). The CIR itself is separately identified in the "Clinical Investigation Report" column as R-TF-015-006 (MAN_2025). "In preparation" is factually correct for MAN_2025's _manuscript, which has not yet been written. No edit applied. If future review again flags ambiguity, the mitigation is to extend the column header ("Peer-reviewed publication status") rather than to change the cell values.

Warning W3 — ISO 14155 citation scoping (bsi-clinical-auditor)

File:line: R-TF-015-001-Clinical-Evaluation-Plan.mdx:39. Verdict: Accepted — the tightened line-39 prose, as first written, unconditionally asserted that any PMCF-phase investigations listed with status "Planned" are governed by this Plan for execution under ISO 14155:2020. MAN_2025 is a simulated-use MRMC reader study on retrospective anonymised atlas images, outside the scope of MDR Article 2(45), and therefore NOT run under ISO 14155:2020. The line-39 clause needed to carve this out explicitly. Fix applied: added a second sub-clause: ...Rank 11 simulated-use MRMC reader studies on retrospective anonymised images are outside the scope of MDR Article 2(45) and are governed by the CEP and by their respective CIPs without a dedicated ISO 14155:2020 regime.

Warning W4 — "ICC 0.15" abbreviation (audit-deliverable-reviewer + bsi-clinical-auditor)

File:line: R-TF-015-001-Clinical-Evaluation-Plan.mdx:1212, Sample-size rationale sentence. Verdict: Accepted — "ICC 0.15" alone is dev-note compression; a PDF reviewer should not have to leave the CEP to decode "ICC". Also accepted Erin's complementary W4: the ICC-sensitivity range (0.05–0.30) is what anchors the robustness claim and should appear at CEP level. Fix applied: sentence expanded to ...after accounting for an assumed reader-level intraclass correlation coefficient (ICC) of 0.15, with the primary-endpoint conclusion confirmed robust across an ICC sensitivity range of 0.05 – 0.30.

Warning W5 — Reference-standard limitation cross-ref missing (bsi-clinical-auditor)

File:line: R-TF-015-001-Clinical-Evaluation-Plan.mdx:1212, Reference-standard sentence. Verdict: Accepted — atlas diagnoses are not all histopathologically confirmed; the CIP (R-TF-015-004 §Reference standard) explicitly documents this limitation and three mitigants. Surfacing the limitation and its cross-reference at CEP level, as a single clause, is standard regulatory hygiene. Fix applied: sentence extended to Atlas diagnoses may not all be histopathologically confirmed; the limitation and three-point mitigation narrative (image-set size, self-controlled paired design, cross-study methodological consistency) are documented in R-TF-015-004 §Reference standard (ground truth).

Warning W6 — "Ten clinical investigations" enumeration drift (celine-clinical-consultant W1)

File:line: r-tf-015-013-statistical-summary.mdx:11 vs R-TF-015-001-Clinical-Evaluation-Plan.mdx:39. Verdict: Accepted — the two documents both landed on "ten" but with different compositions (CEP excluded AIHS4 as a proof-of-concept pilot and included NMSC_2025 as the tenth; stat summary included AIHS4 as study 9 and excluded NMSC_2025 from its table). This is a visible drift at cross-document reading. Fix applied: stat-summary scope sentence rewritten to (a) break the composition out explicitly — "five prospective real-patient studies, four MRMC simulated-use reader studies, and one retrospective proof-of-concept / pilot feasibility study (AIHS4_2025)"; and (b) explicitly state that NMSC_2025 and R-TF-015-012 are out of scope for this pre-market statistical summary (with their respective loci: the CER and the legacy umbrella PMS Report). This simultaneously resolves Celine's W2 (R-TF-015-012 out-of-scope needs to be explicit, not ambiguous).

Warnings / notes — noted, not actioned

• N1 (audit-deliverable W1): :::info admonition at stat-summary line 113. PDF-rendering style. The block had pre-existed the task-3b13 edit and is not a new leak. Left untouched.
• N2 (audit-deliverable style 4): percent-spacing and Greek-letter inconsistency (60 % vs 60%; Unicode α β vs Roman alpha/beta). Deliberate choice — Greek notation and SI-standard spacing are appropriate for statistical / regulatory register. Not changed.
• N3 (bsi-clinical-auditor W3): "Date of the study report" cross-reference to the CIR signature block. Agent flagged that other rows use a concrete calendar date. Reviewed and kept as a cross-reference because the CIR's <Signature /> block is dynamic by construction — hard-coding a calendar date would drift each time the CIR is re-signed. The cross-reference phrasing is intelligible to a PDF reader and does not invite a flag on its own.
• N4 (audit-deliverable style): line 113 admonition style. Same as N1. Not changed.

Pillar-mapping verification (Celine Check 7)

Check	Status	Note
MAN_2025 present and anchored Pillar 3 §4.4 Rank 11 (supporting)	PASS	R-TF-015-001:1212 Type-of-study cell preserved verbatim; r-tf-015-013:470 heading updated to (completed) without pillar change.
Rank-vs-Pillar orthogonality preserved	PASS	No elevation to Rank 2/4 or to Pillar 2; Sample-size and Reference-standard paraphrases do not over-claim.
Causal-chain integrity	PASS	R-TF-015-001:1171 mandated-integration-outputs paragraph untouched; surrogate anchoring via R-TF-015-011 preserved.
Integrator-responsibility language	PASS	R-TF-015-001:1171 integrator-is-a-co-controlled-risk-control-agent framing untouched; no soft-delegation language introduced.
Surrogate-endpoint anchoring (Class IIb indirect-benefit)	PASS	Row's 10 %/14 %/11 % acceptance criteria remain surrogate-endpoint deltas vs standard clinical practice, not direct-benefit claims.
R-TF-015-012 Rank 8 primary / Rank 4 supplementary posture	PASS	CEP line 39 explicit; stat-summary scope sentence now explicitly out-of-scopes the legacy RWE study.

Verdict

All actionable findings applied or dismissed with stated rationale. The CEP and stat-summary are now internally consistent and mutually consistent with the CIP (R-TF-015-004 MAN_2025 instance) and the CIR (R-TF-015-006 MAN_2025 instance). Pillar 3 §4.4 Rank 11 anchoring preserved. No new audit-visible leaks introduced.