self-review

name	self-review
description	Pre-submission self-review for the user's own manuscripts, applying a reviewer perspective. Systematic check across 10 categories with research-type branching. Outputs Anticipated Major/Minor Comments with severity framing and optional R0 numbering for /revise pipeline integration.
triggers	self-review, pre-submission check, check my paper, reviewer perspective, manuscript self-check
tools	Read, Write, Edit, Grep, Glob
model	inherit

Self-Review Skill

You are helping a medical researcher check their own manuscript before journal submission. The goal is to anticipate reviewer comments by applying the same critical lens used in peer review across medical journals.

This is NOT about writing a review. It's about producing an actionable list of anticipated reviewer comments with specific fix suggestions, so the manuscript can be strengthened before reviewers ever see it.

Optional Flags

• --fix: After generating the review report, automatically apply fixes for all issues where fixable_by_ai is true. Edits the manuscript in place, then reports a diff summary. Does NOT fix issues marked fixable_by_ai: false (e.g., missing data, design flaws). Maximum 2 fix-and-re-review iterations.
• --json: Output the structured JSON block (see Phase 3c below) in addition to the markdown report. Default when called from /write-paper Phase 7.
• --panel: Run the multi-agent panel review (Phase 2.6) — several domain-expert reviewers in parallel plus an editor synthesis — instead of the single-pass review. Opt-in and off by default (a panel spawns N reviewer agents + 1 editor, so it costs several times more tokens). Reserve it for a high-stakes pre-submission final pass on a top-tier target. Do not combine with --fix: a panel diagnoses and prioritizes; run --fix as a separate follow-up pass once the author has triaged the panel's findings.

Severity Framing

When flagging issues, classify severity:

• Fatal: Fundamental design flaw that cannot be fixed with existing data (e.g., data leakage that invalidates all results, absence of any reference standard, label-feature circularity). The manuscript likely needs redesign. Submission would likely result in Reject.
• Fixable: Significant but addressable with existing data (e.g., missing calibration analysis, unclear exclusion criteria, absent CIs, incomplete reporting). These are the most actionable findings.

Most issues are Fixable. Reserve Fatal for true design-level problems.

Workflow

Phase 1: Intake

1. Get the manuscript -- PDF, Word doc, or pasted text.

2. Ask the user:

   • Target journal? (affects reporting standards and scope expectations)
   • Manuscript type? (original research / review / technical note / letter / meta-analysis / case report)
   • Anything they're already worried about?
   • Review depth? The default is a single-pass review. For a high-stakes pre-submission final pass, a multi-agent panel (--panel, Phase 2.6) is available — several domain-expert reviewers run independently, then an editor consolidates them (more thorough, but it spawns several agents so it costs several times more tokens). On an interactive run, surface this option once in one line and offer it; then proceed with the single-pass review unless the user opts in. Do not surface or auto-apply the panel when invoked with --json or from /write-paper — those stay single-pass.

3. Read the full manuscript.

4. SSOT gate — confirm there is one manuscript, not several. Self-review reads a single input file, so a divergence between a legacy working copy and the live submission copy is structurally invisible to it. Before a --panel run (or any pre-submission pass), check for multiple copies and reconcile first:

   find . \( -path '*manuscript*' -o -path '*main_document*' \) -name '*.md' | grep -v node_modules
   

   If more than one manuscript-like file exists, confirm which is the SSOT and run /sync-submission's divergence gate before reviewing — a STALE_COPY (an SSOT numeric claim or heading that did not propagate to the other copy) is a P0 that must clear first:

   python3 "${MEDSCI_SKILLS_ROOT:-$HOME/workspace/medsci-skills}/skills/sync-submission/scripts/detect_copy_divergence.py" \
     --ssot <ssot>.md --copy <other-copy>.md
   

   Review the SSOT copy; do not review a stale copy and pass it.

   In --panel mode this is a blocking precondition, not advice. A panel spawns N reviewer agents + an editor, so reviewing a stale copy wastes the whole pass (a prior panel's top finding was literally "you reviewed the wrong file"). If the find above returns more than one manuscript-like .md and the SSOT is not pinned — no SSOT.yaml with truth.manuscript_md and no explicit --ssot <path> argument — STOP before spawning any reviewer and have the user name the SSOT (and clear any STALE_COPY). Do not auto-pick the longest/newest file. The single-pass review may proceed on the one file it was given, but the panel must not.

Phase 2: Systematic Check

Run the manuscript through each applicable category below. For each item, assess whether a reviewer would raise it as a Major or Minor comment.

Use the Research-Type Adaptation table (below) to determine which categories apply fully, partially, or not at all for the given manuscript type.

A. Study Design & Data Integrity

Check	What to look for
Patient-level splitting	Are train/val/test splits at the patient level? Is this explicitly stated?
Leakage risk	Any postoperative variable used in a preoperative model? Cohort-wide preprocessing before split?
Input-text contamination	For NLP/LLM extraction tasks, does any supplied report text (clinical history, indication, impression, prior diagnosis, referral text) already contain the target label? If yes, mark as Major unless the input was masked or a no-leaky-field sensitivity analysis is reported.
Temporal independence	Random split within same institution = no temporal independence. Acknowledged?
Analysis unit clarity	Patient vs exam vs lesion vs image -- is the unit consistent throughout?
Sample size per class	For the test set specifically -- are there enough cases per class for stable metrics?

B. Reference Standard & Ground Truth

Check	What to look for
Definition specificity	Is the reference standard precisely defined? (e.g., "pathological T stage" vs vague "staging")
Timing	Interval between index test and reference standard reported?
Independence	Were ground truth annotators independent from the comparator readers?
Annotation protocol	Number of readers, consensus method, blinding, inter-reader agreement reported?

C. Validation & Statistical Reporting

Check	What to look for
Confidence intervals	All primary metrics have 95% CIs?
Calibration [CRITICAL]	Prediction models: calibration plot + Brier score or slope/intercept MUST be present. AUC alone is insufficient -- mark as Major if absent
Clinical comparator	Is there a clinical-only baseline to show incremental value?
DCA / net benefit	For clinical decision tools: decision curve analysis present?
Fine-tuning baseline	For LLM/NLP fine-tuning, LoRA, prompt-engineering, or multi-agent claims, is there a same-backbone zero-shot or few-shot comparator on the same input, schema, and test split?
Multiple comparisons	If many tests: acknowledged as exploratory, or correction applied?
Paired statistics	If same patients compared across modalities: paired tests used (McNemar, DeLong)?
Effect-size meaningfulness	Scored separately from significance: is each primary effect (OR, HR, beta, Cohen's d, correlation) translated to a real-world unit shift and compared to a minimal clinically important difference? Is significance driven by magnitude rather than sample size?
Power-aware null interpretation	Scored separately from significance, for any non-significant primary result (p > 0.05, 95% CI crossing the null): is the analysis powered to exclude a clinically meaningful effect? An underpowered null is "not yet established," not "no effect" -- if the upper CI bound still includes a meaningful effect size, a flat "X was not associated with Y" claim overreads the data. Look for reported observed power or a minimum detectable effect that justifies a negative conclusion, and watch for bilateral over-correction (a prior "independently associated" overclaim swinging to an equally unsupported "not associated" claim during revision). Undocumented null = Minor; a null that drives a clinical recommendation or a headline negative conclusion without power/CI-compatibility justification = Major.
Equivalence-margin discipline	A claim that two groups/methods are "equivalent," "non-inferior," "indistinguishable," or show "no difference" requires a pre-stated margin — a TOST procedure, or the CI compared against a declared MCID. Grep indistinguishable|equivalent|non-inferior|no difference and check for an adjacent margin|TOST|MCID|non-inferiority; a margin-free equivalence claim is a Major (it converts a failure to reject into positive evidence of no effect).
Interaction-anchor discipline	When synergy / interaction / effect-modification is the research question, the null must be anchored to the interaction parameter (a likelihood-ratio test of the interaction term, or the interaction OR/HR on one consistent scale), not to a main-effect OR whose upper CI is then read as "no synergy." Grep synergy|interaction|joint effect|effect modification; if present, confirm Results carries an OR_int|β_int|LRT|p_interaction term. A synergy conclusion resting on a main-effect estimate is a model mis-specification (Major), even when each main effect is individually correct.
Difference-in-significance discipline	A between-group claim that an association is "more X / stronger / more pronounced in group A than group B" must rest on a formal interaction test, not on group A being significant (p < 0.05) while group B is not (p = NS). The difference between "significant" and "non-significant" is not itself significant. Grep more (clearly|strongly|pronounced)|stronger in|(only|chiefly) in (men|women|older|younger|the [A-Za-z]+ subgroup) near two stratum-specific estimates with discordant p-values; if no interaction term (p_interaction|OR_int|LRT) is reported for that contrast, flag it (difference-in-significance fallacy). A subgroup-difference conclusion built this way is a Major; the fix is to report the interaction test or soften to "associations were observed in group A; the interaction was not formally tested."

D. Clinical Framing & Importance

Check	What to look for
Intended use	Is the clinical decision point clearly stated? (triage vs diagnosis vs prognosis vs monitoring)
Overclaiming	Does language match evidence? ("will improve" -> "may potentially"; "superior" with overlapping CIs?)
Terminology precision	Key terms defined? (e.g., "perioperative" = when exactly?)
Title-content alignment	Does the title accurately reflect what was actually done?
Novelty statement	What does this study add beyond existing literature? Is this explicitly stated?
Substantive novelty differentiation	For AI/LLM extraction papers, does the Introduction name 2-3 close prior papers/systems and state the concrete delta (new task, dataset, workflow, method, validation, or clinical decision point), rather than merely saying the method is novel?
Clinical importance	Would the findings change clinical practice or research direction? Is this articulated?
Decision impact	Does the manuscript state what decision, workflow step, or downstream action would change if the model is correct? A text-only phenotype that does not alter triage, treatment, surveillance, enrichment, or research operations has weak clinical utility even if accuracy is high.
Added value / actionability	Scored separately from novelty: does the finding add value over a measure already in routine use, or is it "real but redundant" (restates a standard test)? At the typical effect size, would a clinician act on it for an individual?
Endpoint↔conclusion scope [CRITICAL]	Does the conclusion's action exceed what the design or endpoint supports? A cross-sectional / single-visit study cannot license a prognostic or surveillance claim (rescreen interval, disease progression); a binary surrogate endpoint (present/absent, >0) is risk stratification, not a care directive (defer/withhold/initiate therapy). Both are documented anti-patterns.

Run the deterministic scope gate:

python3 "${CLAUDE_SKILL_DIR}/scripts/check_scope_coherence.py" \
  --manuscript manuscript.md --out qc/scope_coherence.json --strict

CROSS_SECTIONAL_PROGNOSTIC and SURROGATE_CARE_DIRECTIVE are Anticipated Major Comments (category: D. Clinical Framing). CROSS_SECTIONAL_YIELD_LANGUAGE is an Anticipated Minor Comment — a cross-sectional / prevalence design using incidence-flavored screening vocabulary ("yield", "detection rate", "number-needed-to-screen/image", "rescreen interval") without defining "yield" once as cross-sectional report-positive prevalence. The gate is conservative — it fires only when a design/endpoint signal and a conclusion-region action verb (or the yield lexicon) co-occur.

E. Reproducibility

Check	What to look for
Preprocessing details	All steps listed in order? Normalization, augmentation, resampling specified?
Model details	Architecture, optimizer, LR, batch size, epochs, early stopping reported?
Segmentation protocol	ROI definition, reader experience, blinding, tool used?
Hardware/software	Inference environment, software versions, code availability?
Scanner/protocol info	For imaging studies: scanner model, sequence parameters, contrast protocol?
Data/code availability	Is a data availability statement included? Code shared or reason for not sharing stated?

F. Reporting Completeness

Check	What to look for
Abstract-body consistency	Numbers in Abstract match Tables/Results?
Table/Figure accuracy	Cross-check key values between tables, figures, and text
Follow-up duration	For survival/prognosis: median follow-up with IQR reported?
Ethics	All participating institutions' IRB approval documented? Patient consent described?
Missing data	Handling of incomplete cases described?
CONSORT/STARD/TRIPOD flow	Appropriate flow diagram present with patient counts at each step?
Body word count vs journal cap	Is the body within the target journal's word limit? A revise loop monotonically adds words and silently breaches the cap. Run /sync-submission scripts/check_wordcount_cap.py (--journal-profile or --limit; the binding number is the rendered DOCX count). Over cap → Major; within 0.95× → Minor (a further pass will likely breach).
Funding & COI	Funding sources and competing interests disclosed?

G. Reporting Guideline Compliance

Match the manuscript type to the appropriate checklist and verify key items:

Manuscript type	Checklist	Critical items to verify
Diagnostic accuracy	STARD / STARD-AI	Flow diagram, reference standard, spectrum
Prediction model (non-AI)	TRIPOD 2015	Model development vs validation, calibration, missing data
Prediction model (AI/ML)	TRIPOD+AI 2024	Model development vs validation, calibration, leakage, fairness
AI / Radiomics	CLAIM 2024 / CLEAR	Feature selection transparency, external validation
RCT	CONSORT / CONSORT-AI	Randomization, blinding, ITT
Systematic review (interventions)	PRISMA 2020	Search strategy, screening, risk of bias
Meta-analysis (observational)	MOOSE + PRISMA 2020	Confounding assessment, heterogeneity, publication bias
Observational	STROBE	Confounding, selection bias, missing data
Reliability / agreement	GRRAS	ICC model/type, rater description, measurement protocol
Educational	SQUIRE 2.0	Intervention description, outcome measures, context
Case report	CARE	Timeline, diagnostic reasoning, informed consent
Surgical	STROBE-Surgery	Surgeon experience, technique details, complications

For a full item-by-item audit, run /check-reporting on this manuscript. If it has already been run, reference its results and flag any MISSING items as Anticipated Major/Minor Comments. If not yet run, flag: "Full reporting guideline compliance not yet audited -- run /check-reporting before submission for item-level assessment."

H. Circularity

Check	What to look for
Label-feature overlap	Is the prediction label derived from the same data source as any input features? (e.g., NLP-extracted label + text-derived features from same reports)
Tautological prediction	Does the model predict something that is already encoded in its inputs?
Circular validation	Is the validation set constructed using information from the training process?

I. Protocol Heterogeneity

Check	What to look for
Multi-site acquisition	If multi-site: are scanner models, protocols, and acquisition parameters reported per site?
Harmonization	For imaging or lab features: was harmonization applied (ComBat, z-scoring)? If not, acknowledged?
Temporal protocol drift	For longitudinal data: did acquisition protocols change over the study period?

J. Method Transparency

Check	What to look for
Model provenance	Is it clear where the model came from? (in-house vs vendor-provided vs open-source)
Training vs fine-tuning	If pre-trained: was the model fine-tuned on study data? If vendor-provided: any access to training data composition?
Proprietary limitations	For commercial AI or tools: are known limitations acknowledged? Can results be independently reproduced?
Classical-style body conventions	Does the body carry an AI tell or a policy violation a senior reviewer flags on sight — a § symbol, an in-body AI-disclosure paragraph, eligibility criteria as prose, mixed OR/HR decimal places, or em-dash overuse?

Run the deterministic classical-style lint (these are all greps, so they belong in a gate, not eyeballing):

python3 "${CLAUDE_SKILL_DIR}/scripts/check_classical_style.py" \
  --manuscript manuscript.md --out qc/classical_style.json --strict

SECTION_SYMBOL and INBODY_AI_DISCLOSURE are Major (the § count must be 0; the AI-disclosure paragraph belongs on the title page for a classical / senior-MA target, not the body). ELIGIBILITY_PROSE, DECIMAL_INCONSISTENCY, and EM_DASH_OVERUSE are Minor. This is the self-review-side mirror of /write-paper Step 7.1's classical QC (manuscript-style-classical §5/§6/§7/§8).

K. Reviewer-team consistency (SR/MA-only; fabrication-grade)

Check	What to look for
DUAL vs SINGLE conjunction [CRITICAL]	Methods or PROSPERO claims dual independent reviewers AND Discussion/Limitations admits single primary reviewer + 20% sample (or "deferred to before submission")? Mark as MAJOR, fabrication-grade.
LLM-as-reviewer [CRITICAL]	A per-study extraction JSON whose reviewer/screener/extractor field is an LLM (Claude, GPT-4, Gemini, "LLM")? An LLM is a tool, not an independent reviewer — listing it as one misrepresents the team. Fatal, regardless of the prose.
Deferred mitigation	A future-tense mitigation promise — "a 20% sample will be completed before submission" — unmet at circulation? The future tense is the tell that the work is not done. MAJOR.

Run the deterministic check at Phase 2 entry (pass the extraction JSON — a file or a directory of per-study JSONs — so the prose↔JSON↔confession 3-way is covered):

python "${CLAUDE_SKILL_DIR}/scripts/check_reviewer_team_consistency.py" \
    --manuscript manuscript.md \
    --prospero prospero/record.md \
    --extraction-json extraction/ \
    --out _audit_self/reviewer_team_consistency.md

Exit 1 = MAJOR red flag. The JSON sidecar carries dual_hits, single_hits, llm_reviewer_hits, and deferred_mitigation_hits. Any of the DUAL+SINGLE conjunction, an LLM reviewer field, or a deferred mitigation trips it. Either of the dual/single claims alone is fine; the conjunction is read by reviewers as fabrication. Resolution path:

1. Honest Methods/PROSPERO update (single-reviewer execution disclosed), OR
2. Limitations confession rewritten if dual review was actually completed.

Research-Type Adaptation

Not all categories apply equally to every study type. Use this routing table:

Category	AI/ML	Observational	Educational	Meta-Analysis	Case Report	Surgical
A. Study Design	Full	Full	Partial	N/A	N/A	Full
B. Reference Standard	Full	Full	N/A	Per-study	Partial	Full
C. Validation & Stats	Full	Full	Full	Special*	Partial	Full
D. Clinical Framing	Full	Full	Full	Full	Full	Full
E. Reproducibility	Full	Partial	Partial	Partial	N/A	Full
F. Reporting	Full	Full	Full	Full	Full	Full
G. Guideline Compliance	Full	Full	Full	Full	Full	Full
H. Circularity	Full	Partial	N/A	N/A	N/A	Partial
I. Protocol Heterogeneity	Full	Full	N/A	Per-study	N/A	Full
J. Method Transparency	Full	Partial	Partial	N/A	N/A	Partial
K. Reviewer-team consistency	N/A	N/A	N/A	Full	N/A	N/A

*Meta-analysis: Replace C with heterogeneity assessment (I-squared, prediction intervals), publication bias (funnel plot, Egger), and sensitivity/subgroup analyses.

Type-Specific Additional Checks:

• Observational studies: Confounding assessment (DAG or adjustment strategy), selection bias, exposure measurement validity. Run Phase 2.5e (Confounding Completeness) and apply the O1–O14 probes in references/domain-probes/observational_confounding.md — including O7 (over-adjustment: do not adjust for a consequence/mediator of the outcome, e.g. serum uric acid in an eGFR model — the opposite-direction failure to O1), O8 (analysis unit & clustering — run check_cohort_arithmetic.py --id-col for records-vs-subjects), O9 (construct validity of a report-/registry-derived outcome), O10 (an inferential effect-size gradient across overlapping/nested subsets needs a difference/interaction test, not descriptive refinement alone), and — for complex-survey data (NHANES/KNHANES/CHNS) — O11 (design-based weighting: the right weight + strata + PSU, subpopulation-not-subset) and O12 (data-driven inflection-point/'saturation' threshold mining needs a breakpoint CI + pre-specification, not a quoted cutoff), O13 (a cross-sectional mediation claim cannot establish X→M→Y order and needs an unmeasured-M–Y-confounding sensitivity), and O14 (a synergy/joint-effect/effect-modification claim needs the additive scale — RERI/AP/S with CIs — not a multiplicative-only interaction or joint-category ORs). If the manuscript develops or compares a clinical prediction model (TRIPOD / TRIPOD+AI, nested predictor-set comparison), also apply the CP1–CP4 probes in references/domain-probes/clinical_prediction_model.md (apparent-vs-optimism-corrected calibration/DCA, the incremental-value-vs-marginal-effect two-null distinction, EPV per nested model, net benefit as model comparison not policy).
• Educational studies: Learning outcome measurement validity, Kirkpatrick level, control group adequacy, curriculum fidelity
• Meta-analyses: Search comprehensiveness (2+ databases), screening reproducibility (2 reviewers), RoB assessment per study, GRADE certainty
• Case reports: Diagnostic reasoning transparency, timeline completeness, informed consent, generalizability disclaimer
• Surgical studies: Learning curve consideration, surgeon volume/experience, complication grading (Clavien-Dindo), operative detail completeness

Domain probe modules (load when the manuscript type matches):

These modules carry the same domain-specific critique probes used by /peer-review, vendored here so self-review reaches the same depth (in particular, survival/time-to-event manuscripts now get a dedicated probe set that the routing table above does not otherwise cover).

Manuscript type / signal	Probe module
Systematic Review / Meta-Analysis	references/domain-probes/sr_ma.md (P0–P10)
Time-to-event / survival / prognostic model (Cox, Fine-Gray, DeepSurv, nomogram, risk-stratification cutoff)	references/domain-probes/survival_prognostic.md (S1–S9)
Radiomic feature reproducibility / acquisition-parameter sweep / reliability-based feature filtering	references/domain-probes/radiomics.md (R1–R4)
Cross-modality image synthesis (MRI→PET / MRI→CT / non-contrast→contrast / low-dose→full-dose) claiming functional/molecular information or target-modality substitution	references/domain-probes/image_synthesis.md (IS1–IS4)
Narrative / review article / primer / state-of-the-art	references/domain-probes/narrative_review.md (RV1–RV9)
AI/ML primary study with a clinical claim (generalizable / outperforms clinicians / deployment-ready / can replace a reader)	references/domain-probes/ai_overclaiming.md (AO0–AO5)
Randomised controlled trial (parallel / crossover / cluster / stepped-wedge)	references/domain-probes/rct_trial.md (RC0–RC7)
Diagnostic test accuracy (DTA) primary study / multi-reader multi-case (MRMC) reader study (index test vs reference standard, AI-vs-reader, modality comparison)	references/domain-probes/diagnostic_accuracy.md (D1–D6)
Case report / case series / single-patient clinical narrative (incl. adverse-event/pharmacovigilance and imaging-led radiology/nuclear-medicine/IR reports)	references/domain-probes/case_report.md (CR1–CR9)
AI/ML, prediction, or diagnostic study claiming cross-population performance (generalizable / deployment-ready / "works for patients"), or presenting subgroup analyses as a fairness/equity argument	references/domain-probes/equity_fairness.md (EQ0–EQ6)

When the manuscript matches a row, read ${CLAUDE_SKILL_DIR}/references/domain-probes/<module>.md and apply each probe as an additional source of Anticipated Major / Minor Comments. The module severity words (MAJOR / MINOR) map to this skill's framing as follows: a conclusion-threatening or design-level finding becomes a Fatal Anticipated Major Comment, a reporting-level finding becomes a Fixable Anticipated Minor Comment, and each is tagged with the closest category letter (A–K). These probes complement categories A–K above; they do not replace them. (The modules are vendored byte-identical from /peer-review; do not edit one copy only — run python3 scripts/check_domain_probe_sync.py --sync.)

Phase 2.5: Numerical Cross-Verification (Internal)

Before generating the report, verify internal consistency:

1. Abstract vs Body: Do all numbers in the Abstract match the Results section and Tables?
2. Table vs Text: Cross-check key metrics (sample sizes, primary outcomes, p-values) between tables and narrative text.
3. Figure vs Text: Do figure legends match the data described in Results?
4. Percentage arithmetic: Verify that n/N percentages are calculated correctly (e.g., 23/150 = 15.3%, not 15.0%).
5. CI plausibility: Do confidence intervals seem reasonable given sample sizes?
6. Rate back-calculation: every reported rate must invert to its own numerator/denominator — an incidence rate ≈ events / person-years × scale (±rounding). A rate that does not recompute from the stated events and person-time (or that implies more events than the cohort can supply) is a Major, not a Minor.
7. Exclusion-cascade and complete-case arithmetic (cohort/observational): the STROBE flow must balance — start N − Σ(exclusions) == final analytic N — and any complete-case statement must balance — total − missing == complete. A footnote N that does not equal the subtraction is a Major.

For cohort/observational manuscripts, run the deterministic gate instead of eyeballing it (it parses prose equations + GFM tables, and recomputes from a committed CSV when given one):

python3 "${CLAUDE_SKILL_DIR}/scripts/check_cohort_arithmetic.py" \
  --manuscript manuscript.md --data analysis/cohort.csv --id-col mockid \
  --out qc/cohort_arithmetic.json --strict

RATE_BACKCALC / CASCADE_SUM / PARTITION_OVERLAP rows are Anticipated Major Comments (category: A. Study Design & Data Integrity); the partition check is the Phase 2.5b cohort branch below. Pass --id-col (or let it auto-detect a subject-ID column) on health-screening / EMR / registry data so the gate also runs the analysis-unit check: when records > unique subjects and the manuscript states neither the analysis unit nor a one-record-per-subject sensitivity, it emits ANALYSIS_UNIT_UNDISCLOSED (Major — non-independent observations give anti-conservative CIs; probe O8). Flag any remaining internal-consistency discrepancies as Anticipated Minor Comments (category: F. Reporting Completeness).

Phase 2.5a: Numerical Source-Fidelity Audit (External)

Internal consistency (Phase 2.5) is necessary but not sufficient. Numbers can be fully self- consistent across Abstract / Table / Text and still be wrong at the source — a single transcription error propagates cleanly through every downstream stage.

Precedent failure pattern:

A revision-era comparative meta-analysis reported a safety-outcome 2x2 with the arm-level events direction-reversed relative to the primary-source Table. Internal consistency passed because Abstract, Discussion, Table, and the R script all echoed the same wrong values. The reversal was caught only by an explicit second-pass audit that randomly sampled claims and traced each back to the primary paper.

When to run: MA revisions, submissions, or any review where the user mentions "check against the source," "verify extraction," or "random sample."

Inputs the reviewer should expect:

• manuscript.md (or .docx converted to .md)
• extraction_final.csv (or equivalent data-extraction spreadsheet)
• A directory of primary-source PDFs (or equivalent accessible text)

Procedure:

1. Inventory numerical claims in Abstract, Results, and Discussion (patterns: \\d+/\\d+, \\d+\\.\\d+%, (95% CI:, p\\s*=\\s*0\\., I\\^2, n\\s*=, etc.).

2. Stratified random sample — draw 5 claims across: (a) pooled estimates, (b) subgroup / sensitivity results, (c) comparative-arm specific values, (d) study-level numbers (first-cited in narrative), (e) a claim introduced during revision if the draft is post-v1. Comparative-arm specific values and revision-introduced numbers are the two highest- yield strata — always include one of each.

3. For each sampled claim, traverse 3 layers:

   • Layer 1 (Manuscript → CSV): Find the row / column in the extraction CSV.
   • Layer 2 (CSV → Primary source): Locate the exact Table, Figure, or paragraph in the original paper. Record page number.
   • Layer 3 (Analysis script → CSV): If the claim came from an analysis script, read the script and confirm its input value matches the CSV cell.

4. Record results in a table and append to the report:

   Claim (manuscript location)	CSV row/col	Primary source (paper, Table/Fig, page)	Script input	Match?

5. Any mismatch is a Major Comment (M-level), not Minor. Mismatches that reverse a direction or change a significance boundary are P0 blockers for submission.

Revision-specific rule: If the manuscript contains [VERIFY-CSV] tags, treat each as a mandatory audit item regardless of the sampling size. The tag exists precisely because that number was introduced after the initial extraction pass and has not yet been independently checked.

Hand-entered analysis-script inputs are a code smell. When Layer 3 reveals a matrix(...), c(1, 2, 3), or data.frame(...) line with numerical data and no CSV-coordinate comment, escalate to a Major Comment even if the audited values happen to match — the next revision will re-introduce the same risk.

Statistic-type fidelity (not just the value). A prose sentence must match the table/CSV not only on the number but on the statistic type. A body sentence that reports a median ("median eGFR 92.8") while Table 1 reports a mean ("mean 91.3") for the same variable cannot be reconciled by a reviewer comparing the two — and the mismatch usually means one of them was not regenerated after a Table 1 rule change (see the mean/median-by-skewness rule in /analyze-stats table-types/table1_demographics.md). Treat a prose↔table statistic-type mismatch (mean vs median, SD vs IQR, n vs %) as a Minor Comment, or Major if it sits on a primary characteristic the conclusion leans on. Also re-check that any descriptive figure the prose quotes (e.g. "78.4% male") matches the current table value, not a stale earlier one.

Stale derived CSVs after a model/adjustment-set change (n mismatch). When the primary model or adjustment set changes mid-revision, every derived CSV (Table 2, sensitivity tables, supplements) must be regenerated, or a stale file silently contradicts the new primary. The fastest tell is the analytic n: if a derived CSV's n differs from the manuscript's current primary n, suspect it is stale — and the conflict can flip a result's significance (a proteinuria sensitivity CSV left at the old n = 4,914 / OR 4.52 contradicted the new primary n = 4,214 / OR 3.99, significant ↔ not). Grep each derived CSV's n against the primary n; any divergence that is not explained by a stated sub-analysis restriction is a Major Comment, requires_reanalysis (re-run, not a prose edit — see Phase 4).

Phase 2.5a-2: Design & Power Statistic Provenance (computed, not extracted)

Phase 2.5a traces data-derived numbers back to a CSV and a primary source. Design and power statistics are a different class and a common blind spot: the minimum detectable effect (MDE), a-priori or post-hoc power, the required sample size for a future trial, and the a-priori effect-size assumptions behind them are computed, not extracted, so they have no CSV row or source-paper Table to trace to. They routinely escape both the internal-consistency check and the source-fidelity audit above.

Precedent failure pattern:

A pilot study reported a minimum detectable effect of d = 1.67. No standard two-sample method reproduces it (the correct value at the stated n, alpha, and power was about 1.24). It survived several review rounds because no committed script computed it — the value had been hand-entered — and one reviewer even cited the figure approvingly. In the same manuscript, a set of future-trial sample sizes was numerically correct but had been produced with an exact noncentral-t tool, while the committed script used a normal approximation and printed different numbers: right value, no reproducible provenance.

Procedure:

1. Inventory design/power claims. Search for: "minimum detectable", "detectable effect", "MDE", "power" (80% / 90% / "1 − beta"), "sample size", "n = N per arm/group", "to detect", "powered to", "a priori", and any a-priori planning effect size (Cohen's d / f / OR used for sizing).

2. Require a reproducible source for each. Every such value must be produced by committed code (e.g. statsmodels TTestIndPower, a G*Power-equivalent, or an explicit noncentral-t computation), with the inputs stated in the manuscript: n per arm, alpha, power, allocation ratio, and one- vs two-sided. A value with no committed-code source is the highest-risk case.

3. Recompute independently with a standard tool, then classify:

   • Not reproducible by any standard method → likely a calculation error (Major; P0 if it is a headline claim). This is the d = 1.67-vs-1.24 case above.
   • Reproducible only by a method the committed script does not implement (e.g. the manuscript value is noncentral-t but the script is a normal approximation) → provenance / method drift. The number may be correct, but update the committed code so it reproduces the reported value (Major: reproducibility, not correctness).

4. Method-consistency across the manuscript. All power, sample-size, and MDE statistics in one paper should share a single method family (e.g. all noncentral-t). A mix of normal approximation and exact-t within one manuscript signals that some values were computed in an ad-hoc side tool.

5. Any non-reproducible design/power value is a Major Comment; a non-reproducible headline power or MDE claim is a P0 submission blocker.

Hand-entered design/power statistics are a code smell even when correct. If no committed function emits the value, flag it: the next revision will re-introduce the risk, and a reviewer who recomputes will not match the manuscript.

POWER_MODEL_MISSPEC — the power/MDE simulation's adjustment set must match the primary model. For cohort "negative findings," the whole conclusion leans on the MDE ("the literature effect of 1.2–1.5 cannot be excluded"), so the MDE must be computed under the same covariate set as the primary analysis. When a committed power/MDE script exists, read its model formula: if it fits y ~ exposure + age (2 covariates) while the primary model adjusts for 6, it overstates power (omitted covariates inflate the apparent precision) — the MDE is too small and the negative claim too strong. Re-running a parametric bootstrap under the full model is the fix (in one worked case MDE moved from a 2-covariate "OR 1.67" to a full-model "OR ≈ 1.70"). A power/MDE whose script omits primary-model covariates → Major (P0 when the MDE is a headline). This is requires_reanalysis (re-simulate, not a prose edit). POWER_VALUE_INTERPOLATED — any interpolat/approx/interp token in a power/MDE CSV's provenance column means the headline value was never simulated on the grid; treat a non-reproducible headline power/MDE as Major.

Phase 2.5b: Screening-Count Reconciliation from ID Sets (SR/MA + observational tier/stratum)

Internal consistency across Abstract/Methods/Results (Phase 2.5) + source fidelity of 2×2 and effect-size numbers (Phase 2.5a) do not cover study-count arithmetic. The latter is a separate failure mode: a prior-draft prose total ("30 → 32 after FLAG consensus") can survive every downstream pass because Abstract, Methods, Results, Discussion, Figure 1 caption, and even the supplementary consensus file all cite the same wrong number back to each other.

Precedent failure pattern (a PRISMA-DTA meta-analysis revision):

A late-revision manuscript reported study counts of k_qualitative = 32, k_narrative-only = 10, k_FT-excluded = 46. An ID-level recount against the screening TSV and consensus sheet (with FLAG additions reconciled) yielded k_qualitative = 24 with only 2 narrative-only studies (k_FT-excluded = 54). The original 32/10/46 figures came from an early-draft assumption that was never reconciled against the ID-level artifacts; downstream files (consensus markdown, supplementary tables, edit plans) propagated the same wrong total. Caught only by an explicit ID-set recount against the screening TSV and consensus spreadsheet, verified independently by an adversarial audit.

When to run: any SR/MA manuscript revision, regardless of stage. Run before Phase 3.

Inputs:

• Screening TSV with one row per full-text-reviewed record and an include/exclude column
• Consensus spreadsheet (Excel/CSV) with one row per record requiring adjudication and a Consensus column (typical values: Exclude, Include-qualitative, Include-bivariate)
• Any FLAG-adjudicated inclusion log documenting records added to the qualitative pool outside the primary screening TSV
• The manuscript's Table 1 (or equivalent): the definitive list of studies contributing to the primary quantitative synthesis

Procedure:

1. Enumerate the ID sets:

   • A = set of IDs marked INCLUDE in the screening TSV
   • B = set of IDs marked Exclude in the consensus spreadsheet
   • C = set of IDs marked Include-qualitative in the consensus spreadsheet
   • T = set of IDs represented in Table 1 (via author/year cross-match)

2. Derive canonical totals:

   • k_qualitative = |A \ B| + |C|
   • k_bivariate = |T|
   • k_narrative-only = k_qualitative − k_bivariate = |(A ∪ C) \ B \ T|
   • k_FT-excluded = |screening TSV rows| − |A| + |B ∩ A| + |(B \ A) encountered at FT stage|

3. List the narrative-only IDs explicitly — this is the highest-yield cross-check. A manuscript claiming "10 narrative-only studies" while the (A ∪ C) \ B \ T set contains only 2 IDs is an immediate P0 finding.

4. Compare each derived total against the manuscript's prose claim in Abstract, Methods §Study Selection, Results §Study Selection, Figure 1 caption, Discussion §Limitations, and any References §Narrative-Only heading. Any mismatch between derived total and manuscript prose = P0 Major Comment, blocking submission.

5. Record results in a short reconciliation block and append to the report:

   | Quantity | Manuscript claim | ID-derived value | Status |
   |---|---|---|---|
   | k_full-text | 78 | 78 | ✓ |
   | k_qualitative | 32 | 24 | ✗ P0 |
   | k_bivariate | 22 | 22 | ✓ |
   | k_narrative-only | 10 | 2 (IDs 120, 474) | ✗ P0 |
   | k_FT-excluded | 46 | 54 | ✗ P0 |
   

Any "N → M" transition claim in a consensus summary (e.g., "30 → 32 after FLAG consensus") that is not backed by an enumerable ID addition/subtraction set is itself a Major Comment, because the transition is unverifiable by downstream audit. Require conversion of every such claim to explicit ID lists before closing the report.

Observational tier/stratum branch. The same set-recount logic applies when a cohort manuscript presents an ordinal tier or mutually-exclusive stratum split. A partition that is claimed to be disjoint must satisfy Σ(stratum N) == unique total and Σ(stratum events) == total events; denominators that sum above the unique cohort double-count subjects, and a table where every stratum n equals the grand total is a stratum-total mis-entry rather than a partition. Run check_cohort_arithmetic.py (Phase 2.5 above) with the stratum CSV — its PARTITION_OVERLAP verdict is the cohort analogue of an ID-set mismatch and is a P0 Major:

python3 "${CLAUDE_SKILL_DIR}/scripts/check_cohort_arithmetic.py" \
  --manuscript manuscript.md --data analysis/strata.csv --strict

Also confirm the reference (baseline) row of any stratified hazard/odds table is present and labelled; a missing reference category makes the other strata uninterpretable.

Phase 2.5c: Reference Hallucination Scan

Numerical audits (2.5/2.5a/2.5b) cover in-text numbers; they do not cover reference-list integrity. LLM-drafted or co-author-handed-in bibliographies frequently contain fabricated DOIs, wrong author/year combinations for a real DOI, or plausible-looking references that never existed. These slip past human proofreading because the surface form looks canonical.

When to run: every manuscript at self-review, regardless of stage. Mandatory before submission and before any revision circulation to co-authors or the editor.

Procedure:

1. Locate the bibliography. From SSOT.yaml → truth.refs_bib (fallback manuscript/_src/refs.bib for legacy projects). If SSOT.yaml is absent, scan references/library.bib as a last resort.

2. Invoke /verify-refs on the resolved bib. The skill writes qc/reference_audit.json with a per-entry verdict (VERIFIED / FABRICATED / UNVERIFIED) and a top-level submission_safe boolean.

   # equivalent CLI form (same result as invoking the skill).
   # verify_refs.py takes a positional input (the .bib path) and writes its audit
   # to <project-root>/qc/reference_audit.json (path derived from --project-root).
   BIB="$(python3 -c "import yaml; print(yaml.safe_load(open('SSOT.yaml'))['truth']['refs_bib'])")"
   python3 skills/verify-refs/scripts/verify_refs.py "$BIB" --project-root . --strict
   

   When both reference QC and cross-reference QC are needed in one pass, prefer the master orchestration entry point in /manage-refs — it chains check_citation_keys.py → verify_refs.py --strict → render_pandoc.sh (optional) → check_xref.py --strict and writes qc/pre_submission_gate.json as the single submission-readiness artifact:

   bash "${MEDSCI_SKILLS_ROOT:-$HOME/workspace/medsci-skills}/skills/manage-refs/scripts/pre_submission_gate.sh" \
       --md manuscript/manuscript.md \
       --bib manuscript/_src/refs.bib \
       --docx submission/<journal>/manuscript.docx \
       --allow-separate-attachments  # see Phase 2.5d for when this is appropriate
   

3. Read qc/reference_audit.json. For each entry not marked VERIFIED, add a row to the reconciliation block below. FABRICATED entries are P0 Major Comments (block submission). UNVERIFIED entries are Minor Comments unless the manuscript is at a circulation/submission gate, in which case they escalate to Major. For each duplicate_findings[] entry (category duplicate_pmid / duplicate_doi), add a Major Comment row noting the duplicated ref_ids pair and recommend cite renumbering — duplicates block submission (P0 Major) regardless of per-record VERIFIED status.

4. Cross-check placeholder + pagination drift. Run, on every round:

   grep -nE '\[@NEW:|\[N\]|\[N–N\]|e0{3}.{0,5}e0{3}|in[ .]?press|\bTBD\b|forthcoming' manuscript/
   

   Two failure classes:

   • Citation-queue placeholders ([@NEW:topic], [N], [N–N]): a citation slot that was never resolved. Any remaining at self-review is a P0.
   • Pagination placeholders (e000–e000, in press, TBD, forthcoming): /verify-refs (Phase 2.5c step 2) marks these UNVERIFIED with note = "pagination_placeholder" but cannot judge centrality from the .bib alone. Here, with the manuscript in hand, decide centrality: if the unresolved reference supports a method choice or a headline claim (grep the citekey/marker against the Abstract, the Statistical Analysis subsection, and the first Results paragraph), escalate it to a P0 Major rather than a generic Minor. A method-load-bearing citation that is still "in press / e000" at submission is a blocker. Include each in the reconciliation block.

5. Record results in a short reconciliation block and append to the Phase 3 report:

   | Citekey | Verdict | Source check | Status |
   |---|---|---|---|
   | Kim_2024_Validation | VERIFIED | DOI + PubMed match | ✓ |
   | Park_2023_Radiomics | FABRICATED | DOI resolves to unrelated paper | ✗ P0 |
   | Lee_2022_DeepLearning | UNVERIFIED | No DOI/PMID, title not found | △ Major before submission |
   | [@NEW:segmentation_review] | PLACEHOLDER | unresolved citation queue | ✗ P0 |
   

Short-circuit rule: if qc/reference_audit.json already exists with a bib-hash match within 60s (P9 cache TTL, pending), the scan MAY reuse it; otherwise re-run. Never consume a stale audit from a prior manuscript revision.

Do NOT fabricate replacement references if any entry fails. Fix-forward belongs to /search-lit and /lit-sync, not to this skill. Self-review only reports the failure and blocks submission.

Phase 2.5c-2: Reference Adequacy Scan

Phase 2.5c covers reference integrity — are the cited references real (fabricated / unverified / duplicate / placeholder)? It does not ask whether there are enough references, in the right sections, grounding every named method. That is reference adequacy, and it is the failure mode behind a draft with thirteen references where the Statistical Analysis subsection names a competing-risk model, multiple imputation, the E-value, and an eGFR equation with zero citations. Keep the two strictly separate: an integrity failure blocks because a citation is wrong; an adequacy failure flags because a citation is missing.

When to run: every manuscript at self-review, after the integrity scan. The two share the manuscript and the resolved bib path.

Procedure:

1. Run the deterministic checker. Resolve the article type from project.yaml (passed verbatim; the script's alias map handles repo paper-type names) and the journal cap from the target journal profile when known:

   python3 "${MEDSCI_SKILLS_ROOT:-$HOME/workspace/medsci-skills}/skills/self-review/scripts/check_reference_adequacy.py" \
     --manuscript manuscript/manuscript.md --bib "$BIB" \
     --article-type "$TYPE" ${CAP:+--journal-cap "$CAP"} \
     --out qc/reference_adequacy.json --strict
   

   It reports the cited-reference count vs the article-type target, the section distribution (Introduction / Methods / Results / Discussion), every named method found in the Methods/Statistical-Analysis block, which of them lack a citation in their paragraph, and a methods_zero_citations flag.

2. Fold findings[] into the review. Each finding becomes a standard issues[] entry (so /revise and downstream consumers ingest adequacy and other comments uniformly), additively carrying the machine-readable issue_type + subtype alongside the usual fields, under category: "F" / category_name: "Reporting Completeness":

   {"id":"M2","severity":"major","category":"F","category_name":"Reporting Completeness",
    "issue_type":"reference_adequacy","subtype":"methods_named_method_uncited",
    "location":"Methods - Statistical Analysis",
    "description":"Fine-Gray competing-risk model is named without a canonical citation.",
    "fixable_by_ai":false,
    "suggested_fix":"Run /search-lit for the canonical Fine-Gray competing-risk source, sync via /lit-sync, then rerun /verify-refs --strict."}
   

   Severity: methods_zero_citations (original / AI-validation / meta-analysis) and each uncited statistical method → Major (a P0 candidate before submission when the method is central to the primary or a sensitivity analysis); each uncited reporting/diagnostic standard → Minor; a total count below the article-type target → Major when far below (under half the floor), otherwise Minor, scaled also by stage (escalate at a submission/circulation gate).

3. Fix-forward, not fabricate. As in Phase 2.5c, this skill never writes replacement references. Every adequacy finding carries fixable_by_ai: false; the remedy is /search-lit (Manuscript Paper Reference Pool mode) → /lit-sync → /verify-refs --strict, which the author runs.

Phase 2.5d: Cross-Reference QC (Manuscript ↔ rendered DOCX)

Reference-list integrity (Phase 2.5c) does not cover Table/Figure cross-references. This is a separate failure mode where in-text citations ("Supplementary Table S4 reports a sensitivity analysis") resolve to a different caption in the rendered DOCX ("Supp Table S4 = a diagnostics table") because the build script carries its own legacy SSOT. Internal consistency (Phase 2.5) cannot detect it — both the prose and the build artifact echo their own divergent truths cleanly.

Precedent failure pattern (an STROBE cohort manuscript revision):

Body prose cited Supp Table S4 as a sensitivity analysis; the rendered DOCX S4 instead contained a diagnostics table. S1, S6, S7 also mismatched. S8 and S9 were cited in the manuscript but absent from the rendered DOCX entirely. Caught only on co-author circulation review.

When to run: every manuscript at self-review when a rendered DOCX exists (e.g., circulation drafts, post-build pre-submission checks). Skip only if no DOCX build has occurred yet (early drafts).

Procedure:

1. Locate inputs. manuscript/manuscript.md (or the SSOT truth.manuscript_md) and the rendered DOCX (typically manuscript/manuscript_final.docx or the most recent circulation .docx).

2. Invoke the shared script (lives in /manage-refs):

   python3 "${MEDSCI_SKILLS_ROOT:-$HOME/workspace/medsci-skills}/skills/manage-refs/scripts/check_xref.py" \
     --md manuscript/manuscript.md \
     --docx manuscript/manuscript_final.docx \
     --out qc/xref_audit.json \
     [--allow-separate-attachments]
   

   The script writes qc/xref_audit.json with per-label rows tagged OK | MISSING_DOCX | MISSING_BODY | MISMATCH | UNCITED | NOT_CITED_NO_BODY, a top-level submission_safe boolean, and a policy.allow_separate_attachments field that records which severity policy applied.

3. Translate findings to anticipated comments. Severity mapping depends on the journal's figure/table submission policy. Many radiology and medical journals (e.g., European Radiology, Radiology, AJR) accept figures and tables as separate attachment files rather than inline in the manuscript DOCX; for those workflows pass --allow-separate-attachments so MISSING_DOCX is not treated as a P0 blocker. MISSING_BODY and MISMATCH remain P0 regardless, because they indicate SSOT drift between body markdown and rendered DOCX rather than a legitimate attachment style.

   Status	Default policy	With --allow-separate-attachments
   MISSING_DOCX	Major (P0) — cited Table/Figure absent from rendered output	Minor — figure/table is separately attached per journal policy
   MISSING_BODY	Major (P0) — build SSOT drift; rendered caption has no body definition	Major (P0) (no change)
   MISMATCH	Major (P0) — caption text disagrees between body and rendered DOCX	Major (P0) (no change)
   UNCITED	Minor — orphan caption that should be cited or removed	Minor (no change)

4. Append a reconciliation block to the Phase 3 report:

   | Label | Status | Body caption | DOCX caption | Verdict |
   |---|---|---|---|---|
   | Supplementary Table S4 | MISMATCH | Sensitivity analysis | Diagnostics table | ✗ P0 |
   | Supplementary Table S8 | MISSING_DOCX | (defined in body) | — | ✗ P0 |
   | Figure 2 | UNCITED | Forest plot of subgroups | Forest plot of subgroups | △ Minor |
   

5. Emit each P0 row as a separate M-numbered Major Comment with category: "F" (Reporting Completeness) and fixable_by_ai: false (build script changes are out of scope for the auto-fix loop — they require pipeline-side fixes per /write-paper Step 7.6a routing).

Do NOT auto-fix cross-reference defects in --fix mode. Caption rewrites in the body without re-running the DOCX build will simply move the mismatch. Surface as Major Comments and let the user route to /write-paper Step 7.6a.

Phase 2.5e: Confounding Completeness (observational only)

For an observational study, the highest-yield reviewer finding is also the most mechanical, and a prose pass misses it because the manuscript text is internally consistent: a covariate that was measured, is imbalanced across exposure groups in the baseline table, and is absent from the adjustment set is residual confounding by a measured variable. Only a join of the exposure- stratified Table 1 against the Methods adjustment set exposes it. This is probe O1 of references/domain-probes/observational_confounding.md, run here as a deterministic gate so the finding lands without the --panel cost.

When to run: manuscript type is observational (cohort, case-control, cross-sectional, health-screening registry) and the central claim is an adjusted exposure–outcome association. Skip for RCTs and descriptive studies.

Precedent failure pattern:

A cross-sectional screening-cohort manuscript reported an adjusted association while Table 1 showed uric acid, smoking pack-years, HDL, total cholesterol, and HbA1c all significantly imbalanced across the exposure groups — none of which were in the age/sex/BMI/hypertension/diabetes adjustment set. The single-pass review passed it; only an epidemiology panel reviewer who read the Table 1 CSV against the Methods caught the gap. After refitting with extended adjustment the primary estimate held, but the manuscript had claimed robustness it had not shown.

Procedure:

1. Locate the exposure-stratified baseline table as a CSV (e.g. table1_by_<exposure>.csv from /analyze-stats) and the Methods adjustment set (the variables after "adjusted for ...").

2. Run the deterministic gate:

   python3 "${CLAUDE_SKILL_DIR}/scripts/check_confounding_completeness.py" \
     --table1 table1_by_<exposure>.csv \
     --adjusted-list "age, sex, BMI, hypertension, diabetes" \
     --exposure-defining-list "body mass index, waist, fasting glucose, triglycerides, HDL cholesterol" \
     --out qc/confounding_completeness.json --strict
   

   It emits a reconciliation table (covariate | imbalance p | SMD | in adjustment set? | verdict) and flags each measured-but-unadjusted imbalanced covariate as an UNADJUSTED_IMBALANCED Major candidate. The gate resolves DB column codes against a prose adjustment set (alias map), and when the Table 1 has no p / SMD column it computes the SMD from per-stratum "mean ± SD" group columns (--group-cols A,B, or auto-detected). When the CSV is unavailable, apply probe O1 by hand from the published Table 1.

   Guideline-defined exposures (MASLD / metabolic syndrome / CKM / sarcopenia / frailty): pass --exposure-defining-list (the components of the exposure's own diagnostic criteria). Those rows are marked EXPOSURE_DEFINING_EXEMPT, not Major — adjusting for them is over-adjustment (probe O7), not a confounding fix. Without the exemption the gate false-positives a Major on every metabolic-criteria covariate. The residual-confounding remedy is an extended-adjustment model adding only non-defining prognostic covariates.

3. Each UNADJUSTED_IMBALANCED covariate is an Anticipated Major Comment (category: A. Study Design & Data Integrity), with the suggested fix: report an extended-adjustment sensitivity model that adds the omitted covariates and states whether the primary estimate is materially unchanged; the original model stays primary only if the extended model agrees.

4. Then apply the rest of the observational probe set (O2 adjustment-set provenance, O3 selection/collider bias, O4 exposure measurement validity, O5 missing-data mechanism & complete-case collapse, O6 residual-confounding E-value, O7 over-adjustment, O8 analysis unit, O9 outcome construct validity, O10 overlapping-subset gradient) from references/domain-probes/observational_confounding.md — these are prose probes (O1/O7/O8 are the data-checkable ones), and complement the generic Phase 2 categories rather than replacing them.

5. Extended-adjustment frame discipline. When the extended-adjustment model adds covariates that carry missingness, its analytic n shrinks. Comparing the adjusted estimate to the full-frame unadjusted estimate confounds adjustment with case-concentrated missingness ("adjustment inflated the estimate" when the drift is who-was-dropped). The fair anchor is the unadjusted estimate refit on the reduced complete-case frame; flag any "adjustment changed the estimate" claim that compares across different frames, and route the refit to /analyze-stats (requires_reanalysis).

Adjustment-set matching is fuzzy (a table row "Smoking, pack-years" vs an adjustment token "smoking"): read the reconciliation table rather than trusting the count, and confirm each flagged covariate is a plausible cause of the outcome (not a mediator or collider, which O2 covers) before raising it.

Phase 2.5f: Claim-vs-Artifact Cross-Check

Phases 2.5–2.5e check numbers and adjustment sets. This phase checks claims against the external artifacts they should trace to — the pre-registration, the protocol, the analysis outputs. These are the errors that survive a single-pass review because the manuscript prose is internally consistent yet disagrees with the registration or the analysis it reports. The first scope is the two highest- value, deterministic instances; figure/flow-count reconciliation, Methods-promised- analysis completeness, and imputation-input integrity are separate subchecks (run /make-figures legend reconciliation and /write-paper's Methods-promised gate).

Precedent failure pattern:

A manuscript reported a null primary association from a multiple-imputation model and described it as "pre-specified," while the registered primary had been the complete-case model that was significant — the primary had been re-designated after the results were known. In the same paper an E-value of 2.79 was attached to the primary HR of 1.34, but 2.79 does not recompute from 1.34 (it came from a different, non-primary estimate), and a second E-value bounded an exploratory cancer-specific hazard, not the headline contrast. None of these tripped the internal-consistency checks; all three are deterministic against the registration and the arithmetic.

Procedure:

1. Run the cross-check with the manuscript and (if available) the pre-registration / protocol / project.yaml:

   python3 "${CLAUDE_SKILL_DIR}/scripts/check_claim_artifact.py" \
     --manuscript manuscript.md --prereg prereg.md \
     --out qc/claim_artifact.json --strict
   

2. Estimand provenance. The script flags PRIMARY_REASSIGNED when the manuscript admits re-designating the primary after results were known, and ESTIMAND_DRIFT when the manuscript's primary statement does not match the registered one. Both are Anticipated Major Comments (category: A. Study Design & Data Integrity); a primary re-designated post-hoc is a P0 issue. The fix is to report the pre-specified and the revised models coequally and disclose the change in the Abstract and Limitations, not to silently lead with the more favourable estimate.

3. E-value. EVALUE_ARITHMETIC means the reported E-value does not recompute from its adjacent effect estimate (the value was likely produced for a different estimate); EVALUE_NON_PRIMARY means the E-value is attached to a secondary/exploratory estimate but presented as if it bounded the headline claim. Both warrant a Major/Minor comment — recompute the E-value for the declared primary estimate and its near-null confidence limit, and quote it there.

4. Primary-change guard. Independently of the script, if the manuscript reports two models for the same contrast where one is significant and the other null and the significant one is foregrounded, confirm which was pre-specified; an outcome-dependent choice of primary model is a Major comment even when each model is individually correct.

5. Headline vs own-sensitivity direction. Read the sensitivity series (S1 etc.) the manuscript itself reports. If the headline causal/association claim points the opposite way from the authors' own adjusted or sensitivity estimate — a positive lead sentence over a sensitivity model that attenuates to the null, or vice versa — that is a Major: the paper is contradicting its own robustness check. This is a prose judgement, not a script verdict.

6. Methods ↔ Results ↔ disk coverage. Run the deterministic coverage gate:

   python3 "${CLAUDE_SKILL_DIR}/scripts/check_artifact_coverage.py" \
     --manuscript manuscript.md --analysis-dir output/analysis \
     --out qc/artifact_coverage.json --strict
   

   PROMISED_ABSENT (an analysis named in Methods that never reaches Results) and DISK_UNREPORTED (an analysis output on disk — an added-value DeLong CSV, a calibration table — never mentioned in the manuscript) are Anticipated Major Comments. The reverse direction matters because a run-but-unreported result can be the one that undercuts the headline. When an _analysis_outputs.md manifest exists the gate uses it as the source of truth; otherwise it globs --analysis-dir and only escalates analysis-bearing file names.

The script is deterministic but its provenance match is fuzzy (token overlap): read the reconciliation in qc/claim_artifact.json and confirm against the actual registration before raising ESTIMAND_DRIFT. For time-to-event manuscripts, also apply probe S8 (estimand provenance) of references/domain-probes/survival_prognostic.md.

Phase 2.6: Multi-Agent Panel Review (--panel, opt-in)

Run this phase only when --panel is passed. The default single-pass review (Phases 2–2.5d) stays the fast path; the panel is the high-cost, high-precision option for a pre-submission final pass on a top-tier target. Run it after the numerical audits (Phases 2.5–2.5d) so the reviewers see source-verified numbers, and before the Phase 3 report, which it feeds.

Precondition (blocking): the SSOT must be singular. Before spawning any reviewer, enforce the Phase 1 step 4 SSOT gate: if more than one manuscript-like .md exists and none is pinned (no SSOT.yaml truth.manuscript_md, no explicit --ssot), halt and ask the user which file is the SSOT — a panel is too expensive to spend on a stale copy. Clear any STALE_COPY from detect_copy_divergence.py first.

The panel simulates independent peer reviewers who do not see each other's comments, then an editor who consolidates them — the same structure a journal uses. It reuses the vendored domain-probe modules so every reviewer applies the same criteria.

Step 1 — Compose the reviewer set by research type. Auto-detect the manuscript type (Phase 1 input + the Research-Type Adaptation table). Each reviewer loads the matching domain-probe module so the panel's criteria are single-sourced.

Research type	Reviewer set (each is one reviewer)	Domain-probe module each loads
Survival / prognostic cohort	R1 Biostatistics & Study Design · R2 Clinical (domain) · R3 Imaging/Radiology (if an imaging exposure)	references/domain-probes/survival_prognostic.md
Systematic review / meta-analysis	R1 Methodology (search/screening/PRISMA) · R2 Clinical · R3 Statistics (pooling/heterogeneity)	references/domain-probes/sr_ma.md
Radiomics / feature reproducibility	R1 Imaging physics & acquisition · R2 ML / Statistics · R3 Clinical translation	references/domain-probes/radiomics.md
Diagnostic-accuracy / AI model	R1 Study design & leakage · R2 Statistics (DeLong, calibration) · R3 Clinical / reference standard	references/domain-probes/sr_ma.md (P1 DTA cells) + references/domain-probes/ai_overclaiming.md (AO0–AO5, for AI clinical claims) + categories A–C
Observational (STROBE)	R1 Epidemiology / confounding · R2 Clinical · R3 Statistics	references/domain-probes/observational_confounding.md (O1/O8 run as the Phase 2.5e / check_cohort_arithmetic.py --id-col deterministic gates; O7 over-adjustment) + references/domain-probes/clinical_prediction_model.md (CP1–CP4, when it is a prediction-model paper) + categories A–J + the effect-size / added-value axes
Narrative / review article	R1 Domain-content expert · R2 Methodology / SANRA · R3 Technical accuracy · R4 Adversarial reject-hunter (structural: RV9 curated-base circularity, RV6 single-anchor overload, RV8 self-citation architecture)	references/domain-probes/narrative_review.md
Case report	R1 Clinical case-report reviewer · R2 Ethics / de-identification · R3 Literature-context reviewer	references/domain-probes/case_report.md + CARE items + categories D/F/G

If the type is ambiguous, ask the user before composing the set.

Step 2 — Run the reviewers (portable execution). When the host provides a parallel subagent / Task capability (Claude Code, or any harness exposing an Agent tool), spawn the reviewer set as independent parallel subagents, each blinded to the others, then run the editor as a final synthesis agent. Fallback (no subagent capability — e.g. a minimal Codex/Cursor harness): a single agent role-plays each reviewer sequentially and in isolation — it completes and writes out reviewer R1's full structured review before reading the manuscript "fresh" as R2, so a later reviewer never sees an earlier reviewer's comments. The panel is defined by these instructions; it does not depend on the Workflow tool or any Claude-Code-only orchestration.

A reusable reviewer schema, a generic harsh-but-fair reviewer prompt skeleton with per-domain focus checklists, and the editor synthesis prompt skeleton live in ${CLAUDE_SKILL_DIR}/references/panel_review_template.md.

Each reviewer returns: reviewer_id, expertise_area, an overall_assessment (name the single biggest threat to the conclusions), strengths (2–3), major[] (each with heading, comment, location, severity, suggested_fix), and minor[]. Map severity onto this skill's own scale — a conclusion-threatening / design-level finding is Fatal, a reporting-level finding is Fixable — rather than introducing a separate vocabulary.

Step 3 — Editor synthesis. One editor pass (a final agent, or the main agent in the fallback) consolidates the reviews:

1. Dedupe findings by theme across reviewers.
2. Flag CONSENSUS for any theme raised by ≥2 reviewers, with R1/R2/R3 attribution (e.g., [CONSENSUS: R1+R3]); single-reviewer findings are attributed to the one reviewer.
3. Decide an internal readiness verdict (this sets the Phase 3c verdict / overall_score; it is not printed as a journal recommendation).
4. Rank the concrete pre-submission actions the author should complete first.
5. State a one-line readiness verdict (ready for the target tier now / fix specific items first / consider a different tier).

Step 3.5 — Lens-diversity gate (deterministic). A panel only earns its cost if its reviewers span distinct axes rather than echo one theme louder. Before the editor finalizes, serialize the reviewers' structured outputs (the schema above) to a JSON file — either a top-level list or {"reviewers": [...], "research_type": "..."} — and run the gate:

python3 ${CLAUDE_SKILL_DIR}/scripts/check_panel_diversity.py \
    --panel panel_reviews.json \
    --research-type {survival|sr_ma|radiomics|dta|observational|narrative} --strict

It reports three diversity failures, each mapped onto a concern family aligned to the focus checklists:

• UNCOVERED_AXIS (Major) — an axis the research type is expected to probe (e.g. heterogeneity/pooling for an SR/MA) drew zero major findings. The editor re-probes it with the owning reviewer before finalizing, or records in the synthesis why the gap is acceptable.
• FAMILY_MONOCULTURE (Major) — the majority of majors fall in one concern family; the lenses converged rather than spanned the manuscript.
• LENS_COLLAPSE (Flag) — a reviewer raised only families another reviewer already covered, adding no independent axis.

Healthy CONSENSUS is preserved — agreement on some themes is a strength (Step 3 flags it), and the gate fires LENS_COLLAPSE only on a fully redundant reviewer and the Major checks on panel-level coverage, never on agreement per se. Do not silently ship a monoculture: resolve every Major before the synthesis verdict.

Step 4 — Feed Phase 3. The consolidated panel output flows into the Phase 3 report, Phase 3b R0 numbering (preserved, so /revise still consumes it), and Phase 3c JSON. CONSENSUS flags and reviewer attribution are additive annotations on the existing M/m comments (and the optional consensus JSON field); they do not change the report or JSON structure.

Re-run the panel after a large revision. A panel is high-yield not only before the first submission but again after any large edit — a word-count compression, a primary-model or adjustment-set change, or resolving a batch of majors. Such edits introduce new drift (a compression drops a caveat; a re-fit leaves a derived CSV stale; a relocation orphans a cross-reference), and the second panel's findings shift character accordingly (method → compression-drift → residual). If the author has just compressed or re-modelled, recommend one more --panel pass rather than assuming the prior panel still holds; in practice each post-revision round surfaces real, distinct errors.

Phase 3: Report

Before writing the Anticipated Comments, skim references/exemplar_findings/ for the finding at hand (cohort-arithmetic mismatch, unadjusted confounder, cross-sectional scope overreach, post-hoc primary / estimand drift). Each models the full shape — which gate fired, the comment in the reviewer's own words, Fatal/Fixable severity, the closest category letter, the concrete fix, fixable_by_ai, and an R0-ready line for Phase 3b. They are synthetic teaching models — match the structure, not the wording.

Generate a concise report with this structure:

# Self-Review Report: {manuscript title}

**Target journal**: {journal}
**Manuscript type**: {type}
**Date**: {date}
**Overall assessment**: {1-2 sentences: key vulnerability and overall readiness}

## Anticipated Major Comments (fix before submission)

M1. **{Issue title}** [{Category letter}]
{1-2 sentences: what a reviewer would likely say, with specific manuscript location}
**Severity**: {Fatal | Fixable}
**Suggested fix**: {specific, actionable fix using existing data}

M2. ...

## Anticipated Minor Comments (address proactively)

m1. **{Issue}** [{Category}]: {1 sentence with location + fix}
m2. ...

## Strengths (emphasize in cover letter)

- {Specific strength 1}
- {Specific strength 2}
- ...

Conciseness targets:

• Anticipated Major Comments: 3-7 items, each 3-5 lines
• Anticipated Minor Comments: 3-6 items, each 1-2 sentences
• Strengths: 3-5 items, each 1 sentence
• Total report: 400-800 words (excluding optional R0 section)

Phase 3b: R0 Numbering (Optional)

If the user plans to use /revise after receiving actual reviews, offer to append R0-numbered output for pipeline compatibility:

## R0 Pre-Submission Findings (for /revise cross-reference)

R0-1 [MAJ] {mapped from M1}: {issue title}
R0-2 [MAJ] {mapped from M2}: {issue title}
R0-3 [MIN] {mapped from m1}: {issue title}
...

When actual reviewer comments arrive as R1-N, the user can cross-reference which issues were anticipated (R0) vs. novel (R1-only).

Phase 3c: Structured JSON Output

When --json is passed, or when invoked by /write-paper Phase 7, append a machine-readable JSON block after the markdown report. Fence it with triple backticks and the json language tag so downstream parsers can extract it.

{
  "self_review_version": "1.0",
  "manuscript_title": "...",
  "date": "YYYY-MM-DD",
  "overall_score": 72,
  "verdict": "REVISE",
  "fatal_count": 0,
  "major_count": 3,
  "minor_count": 4,
  "issues": [
    {
      "id": "M1",
      "severity": "major",
      "category": "C",
      "category_name": "Validation & Stats",
      "location": "Methods, paragraph 5",
      "description": "Calibration plot and Brier score absent for prediction model",
      "fixable_by_ai": true,
      "suggested_fix": "Add calibration analysis paragraph after discrimination results. Generate calibration plot via /make-figures."
    },
    {
      "id": "m1",
      "severity": "minor",
      "category": "F",
      "category_name": "Reporting Completeness",
      "location": "Abstract, line 3",
      "description": "Abstract reports AUC 0.91 but Table 2 shows 0.912 -- rounding inconsistency",
      "fixable_by_ai": true,
      "suggested_fix": "Change abstract to match table: AUC 0.91 (95% CI: 0.87-0.95)"
    }
  ]
}

Field definitions:

• overall_score: Integer 0-100 reflecting manuscript submission readiness
• verdict: "PASS" (score >= 85, no fatal issues) or "REVISE"
• severity: "fatal", "major", or "minor"
• category: Letter code from the 10-category system (A-J)
• fixable_by_ai: true if the issue can be resolved by editing manuscript text with existing data; false if it requires new data, analyses, or human judgment (e.g., design changes, IRB decisions, missing experiments)
• requires_reanalysis (optional, default false): true when closing the finding needs a committed analysis re-run against the real data, not a prose edit — power/MDE re-simulation under the full model, first-visit/one-record-per-subject dedup, an extended- or reduced-adjustment sensitivity model, optimism correction of calibration. Always implies fixable_by_ai: false. Additive and backwards-compatible; parsers that do not expect it must ignore it. Route these to /analyze-stats (see Phase 4).
• suggested_fix: Specific, actionable instruction. If fixable_by_ai is true, this must be concrete enough for the fixer to execute without ambiguity.
• consensus (optional, panel mode only): array of reviewer ids that raised the issue, e.g. ["R1","R3"]. Additive and backwards-compatible — present only when Phase 2.6 ran; parsers that do not expect it must ignore it.

Phase 4: Fix Support

Standard mode (no --fix flag)

After presenting the report, offer to help fix specific issues:

• Rewrite overclaiming sentences
• Draft missing limitation statements
• Suggest statistical additions (e.g., calibration analysis code via /analyze-stats)
• Draft intended use, decision-impact, or novelty-delta statements
• Check specific tables/figures for consistency
• Generate missing flow diagrams via /make-figures

requires_reanalysis findings route to /analyze-stats, not a prose edit (observational/cohort). For cohort and observational manuscripts, the highest-value fixes are usually data-level: a power/MDE re-simulation under the full primary model, a first-visit / one-record-per-subject dedup sensitivity, an extended- or reduced-adjustment (over-adjustment) sensitivity model, or optimism correction of calibration. These are not fixable_by_ai text edits — --fix is text-only and will silently skip them. Tag each such finding requires_reanalysis: true and route it to /analyze-stats for a committed script + CSV, then feed the regenerated numbers back into the manuscript and re-run the relevant Phase 2.5 gate. Surface these explicitly to the author rather than letting an auto-fix pass appear to "resolve" them.

Auto-fix mode (--fix flag)

When --fix is passed:

1. Filter fixable issues: Select all issues where fixable_by_ai is true.
2. Apply fixes sequentially: For each fixable issue, edit the manuscript file directly:
   • Text rewrites (overclaiming, missing sentences, terminology) → Edit in place
   • Missing reporting items (ethics statement, data availability) → Insert at suggested location
   • Numerical inconsistencies (abstract-table mismatch) → Correct to match tables
   • Do NOT attempt: new statistical analyses, new figures, design changes, IRB-dependent items, or any issue tagged requires_reanalysis (route those to /analyze-stats)
   • Do NOT invoke other skills (/make-figures, /analyze-stats) during fix — text edits only
3. Report changes: After all fixes, output a summary:

   ## Auto-Fix Summary
   - Fixed: {N} issues
   - Skipped (requires human): {M} issues
   - Changes: {list of id + one-line description of what was changed}