- Home
- Skills
- Research & Analysis
- Integrity Forensics: Paper Audit Trail
More screenshots
Works with the AI Agents you already use
Integrity Forensics: Paper Audit Trail
A pre-submission sweep for research papers: numbers checked against the result files, citations verified as real and relevant, claims audited against evidence, findings kept in an append-only ledger.
$5
Integrity Forensics: Paper Audit Trail
Example session with this skill installed
Run an integrity sweep on my workshop paper draft before I submit it. The abstract says our method improves accuracy by 37%. There's a citation to a 2024 survey I copied from a related paper, and the conclusion says the method is "robust across domains."
- Read your context and instructions
- Compiled the integrity forensics
INTEGRITY SWEEP — draft frozen as v0.9 (submission candidate). Scope: all sections; ground truth available (result files supplied). Three findings opened.
F-001 — MAJOR, numeric mismatch. Abstract claims 37% improvement; Results §4.2 reports 35%; the run logs show 34.6%. Fix order is ground truth outward: recomputed from the logs (34.6%), rounding policy set to whole percent (35%), abstract and results both updated to 35%. Disposition: FIXED, evidence pointer to the log file recorded. Status: closed with receipt.
F-002 — CRITICAL, citation does not support its claim. The "2024 survey" is actually a 2023 workshop paper, and it does not contain the result it is cited for. A real paper cited for something it never said is worse than no citation. Disposition: REPLACE — the actual source of the quoted result was located, verified for existence and relevance, and substituted; the replacement was itself checked before this finding closed. Status: closed with receipt.
F-003 — MAJOR, overstated claim. "Robust across domains" rests on experiments in two domains. The evidence supports two, not "domains" in general. Disposition: NARROWED — conclusion now reads "robust across the two domains evaluated." Narrowing is an honest fix, not a defeat. Status: closed with receipt.
Gate verdict: CLEAR TO SUBMIT for v0.9 as swept — zero open findings. Verdict scope note: this means no issues found in the evidence at hand; it is not a proof of honesty, and any revision round re-opens the numeric checks for every section it touches.
Connects securely to your tools. The creator never sees your data.
About this skill
Most paper integrity failures are not fraud — they are drift. A number updated in the results section but not the abstract. A citation copied from a related paper and never actually read. A claim that quietly outgrew its evidence across three revisions. This skill systematizes the hostile pre-submission reading that catches drift before reviewers do. It runs in five steps: freeze the draft at a named version and inventory everything the sweep covers; check numeric self-consistency deterministically (internal agreement across sections, arithmetic recomputed from the raw result files rather than the paper's own tables, statistical plausibility, table/figure/text alignment); verify every citation for existence, relevance, version, and self-citation balance, with an explicit keep/fix/replace disposition for each; audit every substantive claim against its evidence and bucket it as supported, overstated (narrow it — an honest fix), or unsupported (withdraw it or run the experiment). Every finding goes into an append-only obligations ledger, anchored to the exact text span, with severity and an explicit disposition — findings are never silently dropped, rewording never counts as a fix, and edited evidence reopens the finding it closed. A severity rubric (Critical / Major / Minor, with escalation rules) drives a gate verdict: blocked while findings are open, clear only with the sweep's scope stated. The skill also names its forbidden loop — never edit-and-resweep until clean, which trains authors to defeat their own audit — and ships a finding template, a pre-sweep checklist, and a fully worked example with three findings closed on receipts.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Research & Analysis.
- The sweep checks consistency, citations, and claim-to-evidence fit; it does not re-run experiments or judge whether the research itself is sound or novel.
- Citation verification needs scholarly database access; citations that cannot be accessed are marked unverified, and unverified load-bearing citations block the gate.
- Without the raw result files, the sweep degrades to internal-consistency checking only, and its verdict says so — it can never report "clear" on that basis.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
Permissions
No special permissions declared or detected