verification-before-publication — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited verification-before-publication (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Claiming results are publication-ready without verification is not confidence — it is a liability to your co-authors, your field, and your career.
Core principle: Evidence before submission, always.
Violating the letter of this rule is violating the spirit of this rule.
NO SUBMISSION CLAIM WITHOUT FRESH VERIFICATION EVIDENCEIf you haven't re-run the verification in this session, you cannot claim it is ready for submission.
BEFORE claiming publication-ready:
1. IDENTIFY: What evidence proves each claim?
2. RUN: Re-run key analyses fresh (not cached results)
3. READ: Full output, check every number
4. VERIFY: Does output match manuscript claims?
- If NO: State actual values with evidence
- If YES: State verification WITH evidence
5. ONLY THEN: Claim ready for submission
Skip any step = asserting, not verifying| Claim | Requires | Not Sufficient |
|---|---|---|
| Results are accurate | claims-audit PASS | "I checked last week" |
| Figures are correct | Re-generated figures match manuscript | "They look right" |
| Analysis is reproducible | Single command reproduces all results | "It worked on my machine" |
| Statistics are proper | hypothesis-first INTERPRET standards met (effect size + CI + correction) | "p-values are significant" |
| All results reported | Experiment log cross-referenced with manuscript | "Main results are in" |
| Excuse | Reality |
|---|---|
| "Should be ready now" | RUN the verification |
| "I'm confident in the numbers" | Confidence is not evidence |
| "Checked last week" | Last week's check is not this week's evidence |
| "Just minor revisions since last check" | Minor changes can invalidate figures and tables |
| "Deadline pressure" | A retraction is worse than a delayed submission |
| "The reviewer will catch errors" | Peer review is not your QC pipeline |
| "I know this codebase" | Knowing the codebase does not prevent stale outputs |
| "Different wording so rule doesn't apply" | Spirit over letter |
| "Co-authors approved it" | Co-authors approved what they saw, not what is current |
| "The analysis passed review last month" | Results are only as valid as the last verified run |
| "We've published this method before" | This manuscript's numbers still need fresh verification |
| "The figures look identical to the last version" | Look identical is not verified identical |
Complete every item before claiming submission-ready. No skipping.
Claims and Accuracy
docs/eureka/audits/YYYY-MM-DD-claims-audit.md frontmatter. Verify manuscript_hash and results_hash match the current manuscript + results. If hashes mismatch, soft-warn: the audit may be stale; consider re-running. Proceed at user's discretion (soft warning, not hard block)status: passed):docs/eureka/audits/YYYY-MM-DD-claims-audit.mddocs/eureka/reviews/YYYY-MM-DD-review.md (research-reviewer ≥ 95/100)docs/eureka/novelty-audits/YYYY-MM-DD-novelty-audit.md (novelty-competitive-audit PASS)Figures
n, statistical test, error bar type (SEM/SD/95%CI), and center value (or confirms N/A for the figure type)Reproducibility
Review
Reporting Completeness
Manuscript Metadata
These must pass independently before the main checklist is considered complete.
reproduce.sh or equivalent exists, is committed, and runs end-to-endFailure on any sub-check item means the analysis is not reproducible. An irreproducible analysis cannot be submitted.
Results match manuscript:
PASS: [Re-run analysis] [Compare output values to Table 2] [All numbers match]
FAIL: "The numbers should match — I haven't changed anything"Figures regenerated:
PASS: [Run figure scripts] [Diff outputs against manuscript PDFs] [Match confirmed]
FAIL: "The figure looks the same as last time"Reproducibility verified:
PASS: [Run reproduce.sh from scratch in clean environment] [All results regenerated]
FAIL: "It worked on my machine before"Statistics verified:
PASS: [Verify every result has effect size + 95% CI + correction method per docs/references/statistical-guide.md] [No uncorrected multiple comparisons]
FAIL: "The p-values are significant — statistics are fine"Completeness verified:
PASS: [Cross-reference experiment log] [Every run in log accounted for in manuscript]
FAIL: "The main results are all in there"Publication errors propagate. A wrong number in a published table becomes a cited number in the next paper, a misinterpreted figure becomes a consensus, and an irreproducible analysis becomes a failed replication. Post-publication corrections damage credibility and waste the field's resources. The gate function exists because the cost of verifying before submission is hours; the cost of a published error is years.
ALWAYS before:
Rule applies to:
eureka:claims-audit PASS — every claim must be traceable before this skill's checklist beginseureka:requesting-research-review PASS at ≥ 95/100 — scientific rigor validated by independent revieweureka:hypothesis-first INTERPRET standards met — every statistical result reports effect size, CI, and correction methodeureka:novelty-competitive-audit PASS — external competitiveness verified against recent literature (internal rigor is necessary but not sufficient for submission)eureka:using-eureka when submission intent is detecteddocs/references/statistical-guide.md — statistical reporting checklistdocs/references/data-checklist.md — data version locking, preprocessing reproducibility, raw→processed regenerationdocs/references/figure-guide.md — see sections "Figure Legend Requirements (Reviewer-Grade)" and "Common Reviewer Rejection Reasons for Figures" for the legend reporting checklist and the dynamite-plot anti-patterndocs/references/novelty-audit-guide.md — search strategy, preemption rubric, differentiation templates, verdict decision tree for the novelty-competitive-audit prerequisiteNo shortcuts for submission verification.
Re-run the analyses. Regenerate the figures. Check every number. THEN claim it is ready.
This is non-negotiable.
RIGID — The checklist is not optional. The gate function sequence is enforced. The Iron Law is not a guideline.
The only flexibility is in the tooling used to re-run analyses — adapt scripts, commands, and environments to your domain. The requirement to produce fresh verification evidence before claiming submission-readiness does not flex.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.