Do we need this task? Honest sizing review

Status (2026-04-21): Written in response to a user question — "is this task really necessary, and would it significantly de-risk the BSI response?" — after Claude noticed that the task's CLAUDE.md was seeded from a lightly corrupted copy of the bsi-clinical-auditor and celine-clinical-consultant agent briefs. The answer below is a standalone sizing review; it does not replace the main brief, it re-scopes it.

TL;DR

• The underlying gap is real but narrow.
• The gap that really matters is a documentation self-inconsistency, not a regulatory-breadth argument.
• BSI did not raise this in round 1; the finding is anticipatory and comes from the bsi-clinical-auditor agent's own rubric.
• The task as currently scoped is 3–5× larger than necessary for round-1 resubmission.
• Recommendation: keep the task, drastically shrink the brief to Option D-lite. Treat Options A / C / D-blend as round-2 contingency only.

The three facts that drive the sizing

Fact 1 — The CEP contains a real textual self-inconsistency

R-TF-015-001 §Subject Matter Expert coverage across the indication scope (line 399) states, verbatim:

"… the Subject Matter Expert's documented consultation network for ad-hoc subspecialty input on case-level adjudication where required."

R-TF-015-001 §Annexes (line 1339) declares Annex I as:

"Annex I: CV AND DECLARATIONS OF INTEREST."

The four evaluator-team members (Barrachina, Vidal, Ugidos, Martorell) have CVs and Declarations of Interest in Annex I. No consultation-network roster, no named subspecialty consultants, no engagement procedure is listed in Annex I or anywhere else in the CEP.

The body text therefore promises documentation that the annexes do not contain. That is an internal-consistency bug, visible on a close reading without any regulatory expertise. It converts a soft breadth-of-coverage concern into a hard "you said it's documented; show me" question.

Fact 2 — BSI did not raise this in round 1

A verbatim grep across every question.mdx under clinical-review/round-1/ and across the 2026-03-25 Erin + Nick clarification-call transcript returns zero hits for:

• evaluator qualifications
• team competence
• SME coverage
• subspecialty expertise
• consultation network

Round 1's seven items address:

• Item 1 — CER update frequency
• Item 2 — device description and claims
• Item 3 — clinical data (equivalence, PMS, sufficiency, gaps)
• Item 4 — usability
• Item 5 — PMS Plan
• Item 6 — PMCF Plan
• Item 7 — risk

None of them touch Annex XIV Part A §1(d)/(f) evaluator qualifications or MDCG 2020-13 Section A administrative-completeness on team expertise.

The round-1 preamble does warn: "Significant gaps were identified which prevented in-depth review … more in-depth and unrelated questions may be asked during later rounds after scope is defined." Round 2 will probably test CEAR Section A more aggressively. But that is an anticipatory reason to act, not a round-1 closure reason.

Fact 3 — The finding comes from an internal-agent rubric, not from BSI

Line 72 of .claude/agents/bsi-clinical-auditor.md hard-codes this exact pattern as a finding:

"SME coverage must be proportionate to the breadth of clinical indications: a single SME for a multi-domain MDSW (malignancy + severity assessment + paediatric + rare-disease) is a finding. Where the claim scope spans beyond the primary SME's documented expertise, named subspecialty consultants must be cited with their contributions traceable to specific sections of the CER. Unnamed 'consultation networks' are not sufficient."

That rubric is genuinely backed by MDCG 2020-13 Section A and MEDDEV 2.7/1 Rev 4 §6.4, so it isn't fabricated. But "the agent would flag this" is not equivalent to "BSI has flagged this." The current task exists because the internal pre-audit caught a real textual defect, not because Erin asked for a fix.

What this means for task sizing

The current CLAUDE.md frames four options:

• Option A — document the consultation network (recruit + named CVs + DOIs + procedure).
• Option B — narrow the intended purpose (IFU ripple; downstream CER/performance-claims edits).
• Option C — add subspecialty SMEs to the evaluation team (recruitment + formal evaluator sign-off).
• Option D — structured indirect evidence + bounded advisory panel.

Three of the four (A, C, D-blend) require recruiting real external people into audit-visible documents with signed CVs and DOIs. That is load-bearing work with human dependencies — weeks-to-months to close, fragile under the resubmission clock, and not actually necessary to fix the round-1 textual defect.

The minimum-viable action that closes the actual exposure (Fact 1) does not require recruiting anyone.

Minimum-viable fix — Option D-lite

Two surgical edits to R-TF-015-001:

Edit 1 — Delete the unsubstantiated claim

In §Subject Matter Expert coverage across the indication scope, remove the phrase "and (c) the Subject Matter Expert's documented consultation network for ad-hoc subspecialty input on case-level adjudication where required."

Rationale: no such network is documented in Annex I. Deleting the claim removes the self-inconsistency. You cannot be held to a commitment you never made.

Edit 2 — Replace with structured indirect-evidence scaffolding

Add, in place of the deleted clause, a paragraph that cites evidence the company already has. Every cross-reference below points to an artefact that is real, present in the repo, and either complete or scoped:

Coverage gap	Existing evidence to cite	Artefact
Dermatopathology / malignancy	Biopsy-confirmed histopathology as reference standard in MC_EVCDAO_2019; NMSC_2025 Clinical Performance manuscript	R-TF-015-003 Pillar 3 §4.4; CER malignancy-evidence section
Severity assessment	Published peer-reviewed severity-validation literature (APASI_2025, AUAS_2023, AIHS4_2023, ASCORAD_2022)	R-TF-015-003 Pillar 2 severity-validation section; R-TF-015-011 SotA
Paediatric dermatology	Paediatric performance-claims rows (Item 2b Jordi work-in-progress) + paediatric age-cutoff definition	R-TF-015-003 performance-claims table; CER pediatric-age section
Autoimmune / genodermatoses	Task-3b5 triangulation; task-3b7 per-epidemiological-group ICD V&V (autoimmune AUC 0.948; genodermatoses 0.905)	R-TF-028-006 §Per-Epidemiological-Group Performance
Darker phototypes (FP V–VI)	MAN_2025 Fitzpatrick V–VI MRMC (149 atlas images; 19 enrolled / 16 primary; data lock 2026-04-17)	R-TF-015-006 MAN_2025 CIR; R-TF-015-010 Annex E
Surrogate-endpoint anchoring	Task-3b6 VCA surrogate-endpoint literature review (Pillar 1 anchor)	R-TF-015-011 Pillar 1 section (once task-3b6 completes)
Indication breadth / RWE	Legacy-device RWE study (R-TF-015-012, 60 respondents, 21 sites) + ~250,000-report legacy PMS corpus	R-TF-015-012; R-TF-007-003 legacy PMS Report

Edit 3 (optional) — Honest one-sentence statement of engagement practice

A single sentence can describe what actually happens on the ground without inventing documentation:

"Where case-level subspecialty adjudication is required during a clinical investigation, external dermatologists are engaged on a study-specific basis and their adjudication is recorded in the investigation records of the relevant study (see R-TF-015-004 / R-TF-015-006 per-study records)."

This is factually true (it reflects what the third-party CROs did), testable against the investigation records, and does not commit the company to producing a standing consultation-network roster.

Effort comparison

Route	Effort	Human dependency	Round-1 closure timing	Residual round-2 exposure
Current brief (Option A / C / D-blend)	2–6 weeks	High (recruit named consultants)	Uncertain	Low
Option D-lite (Edits 1–3 above)	2–4 hours	None	Same-day	Medium
Do nothing	0 hours	None	Same-day	High — textual self-inconsistency ships as-is

Option D-lite is a 10–20× effort reduction vs. the current brief for the bulk of the round-1 risk mitigation. The remaining round-2 exposure (Erin / Nick subjectively judging whether a single-SME footprint is enough for 346-ICD-11 breadth) is a soft finding — a conversation, not a documented gap — and can be escalated to Option A or Option C if and only if BSI raises it in round 2.

Residual risk under Option D-lite

Two genuine risks remain even after Edits 1–3:

1. Round-2 Section-A finding. BSI could still argue under MDCG 2020-13 Section A that a single general-dermatology SME is insufficient for a 346-ICD-11 scope. Mitigation: the structured indirect-evidence scaffolding makes a defensible argument, and Option A / C can be executed in round 2 with a ~6-8 week lead time if Erin pushes back.
2. Nick's breadth-of-indications concern interacting with single-SME footprint. Nick's documented view on 346 indications is "unusual in my experience" and wanted either narrowing or fuller evidencing. A single SME makes that optics slightly worse. Mitigation: the per-epidemiological-group V&V (task-3b7) and task-3b5 triangulation directly answer Nick on evidence breadth; team-breadth is a secondary Nick concern.

Neither residual risk is load-bearing enough to justify Options A / C / D-blend in round 1.

Recommendation

1. Keep the task folder. The gap is real.
2. Replace the main CLAUDE.md brief with Option D-lite as the round-1 scope.
3. Demote Options A, C, D-blend to a "round-2 contingency" section — written down, but not triggered unless BSI raises it.
4. Estimate: 2–4 hours of editing R-TF-015-001 + one pass through all three reviewer agents.
5. Priority: medium. Worth doing before resubmission because of the textual self-inconsistency; not worth doing at the size the current brief describes.

What NOT to do

• Do not invent named consultants for round 1 under pressure. Unnamed is better than fabricated, and "Dermatopathology consultant: [TBC]" is worse than both.
• Do not narrow intended purpose (Option B) as a reflex response to a soft finding. Scope narrowing ripples through the IFU, performance-claims table, CER evidence hierarchy and PMCF plan; the cost is disproportionate to the risk.
• Do not let this task block items that BSI actually asked about in round 1 (Item 2b pediatric claims, Item 3a literature gaps, Item 3a zip of referenced papers). Those are the round-1 critical path.

Ownership

• User: decides whether to adopt Option D-lite or the larger brief.
• Claude: if user adopts Option D-lite, drafts Edits 1–3 into R-TF-015-001, runs the three reviewer agents, propagates consistent language to the CER §Evaluator qualifications discussion (if any).