agentsope/SkillAlchemy / skills/agentsop-bio-fraud-forensics
agentsop-bio-fraud-forensics
VerifiedScreens biomedical / life-science papers for signs of data fabrication, image manipulation, and statistical anomalies, using the detection techniques distilled from the field's canonical exposure platforms (PubPeer, Data Colada, Science Integrity Digest, For Better Science) and tools (ImageTwin/Proofig, statcheck, GRIM/GRIMMER, Problematic Paper Screener, Seek & Blastn). Use when asked to check a paper/figure for image duplication, blot splicing, impossible statistics, paper-mill or tortured-phrase signals, research integrity, or "is this data faked"; or when a user shares a figure, Western blot, supplementary dataset, or DOI and asks whether it looks manipulated. Reports observable anomalies as questions for clarification — it never accuses anyone of fraud.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 2.9 KB |
| SKILL.md | skill md | 12.5 KB |
| USAGE.md | markdown | 8.3 KB |
| examples/demo_screening.md | markdown | 4.2 KB |
| references/R01-misconduct-taxonomy.md | markdown | 19.9 KB |
| references/R02-image-forensics.md | markdown | 19.0 KB |
| references/R03-statistical-forensics.md | markdown | 19.3 KB |
| references/R04-exposure-sites-method.md | markdown | 21.0 KB |
| references/R05-evidence-red-lines.md | markdown | 21.4 KB |
| references/R06-screening-workflow.md | markdown | 21.7 KB |
| references/R07-paper-mill-signals.md | markdown | 19.5 KB |
| references/research_notes.md | markdown | 7.9 KB |
| references/sop_models.md | markdown | 17.4 KB |
references/research_notes.md
View on GitHub# Research Notes — Evidence Summary, Routing Table, Seed Lists
Human-readable distillation of the 7 research dossiers (R01–R07). Use for source
citations, the verification routing table, and the bundled detection word-lists.
## The one-sentence ethic
> Report what you can see from the evidence in front of you, interpret it with explicit
> hedges and alternatives, and let an authorized body decide intent — never accuse.
## Anchor facts (with sources)
- **Image duplication is the #1 biomedical signal.** Bik, Casadevall & Fang, *mBio* 2016 (10.1128/mbio.00809-16): 20,621 papers screened, 3.8% with inappropriate duplication, ~half with features suggesting deliberate manipulation. Type I (simple, 29%) / II (repositioned, 46%) / III (altered, 25%).
- **Most flags are honest errors.** AACR/Proofig screen (Scholarly Kitchen 2024): 1,367 papers → 207 author contacts → 204 honest mistakes, only 4 withdrawn. Calibrates against over-reading.
- **Automated image tools carry a substantial false-positive load.** ASM/ImageTwin publisher pilot, *mBio* 2025 (PMC12505991): a large share of automated flags resolve as benign reuse, and only a small fraction of screened manuscripts had acceptance affected — every flag needs human verification. Tools are halftone-only and database-limited. (Cite the paper's published rates directly rather than a second-hand figure.)
- **Statistical impossibility from text alone.** GRIM/GRIMMER (Brown & Heathers, SPPS 2017); statcheck (Nuijten & Epskamp 2016): ~½ of psych articles ≥1 inconsistency, ~1/8 a decision-flipping one.
- **calcChain spreadsheet forensics.** Data Colada [109] "Clusterfake" (datacolada.org/109): `.xlsx`→`.zip`, read `xl/calcChain.xml`, formula-order vs row-order = manual relocation. [98] = odometer fabrication (uniformity + rounding asymmetry + font-twin Monte Carlo).
- **Legal safe harbor = disclosed facts + hedge + alternative.** Gino v. Data Colada dismissed Sept 11 2024 (protected opinion). Sarkar v. Doe, Michigan COA 2016 (ACLU): opinion on disclosed facts not defamatory; PubPeer commenters not unmasked. *Form* of the statement matters.
- **Paper-mill fingerprints.** Tortured phrases — Cabanac/Labbé Problematic Paper Screener (~130M pubs/week, 473+ phrases). Seek & Blastn — Byrne & Labbé, PLOS ONE 2019 (wrong gene reagents ~3.8–4% in cancer journals). Tadpole mill — Bik, Science Integrity Digest 2020 (400+ papers). Authorship-for-sale network shape — Scientific Reports 2024 (PMC11604963). Hindawi/Wiley 8,000+ retractions 2023.
- **Regulatory ground truth.** ORI / 42 CFR 93: misconduct = FFP + significant departure + intent/recklessness + preponderance; "does not include honest error." 2024 final rule (effective Jan 2025) keeps FFP unchanged.
## Site Function Map — exposure (method) vs database (status)
| Site | Type | What it gives you | How to use |
|------|------|-------------------|------------|
| PubPeer | exposure / METHOD | Per-paper comments; the exact catch steps; author rebuttals | Search DOI/author/title; read each comment as a recipe; browser extension shows flags inline |
| For Better Science (Schneider) | exposure / METHOD + narrative | How one flag expands to a lab/mill-wide pattern | Read for the investigation arc; mine motif-clustering technique |
| Data Colada | exposure / METHOD (stats) | calcChain, distributional tests, duplicate-row forensics | Read numbered posts as replicable protocols |
| Science Integrity Digest (Bik) | exposure / METHOD (image) | Type I/II/III taxonomy; ~1 min/paper visual workflow | Adopt the taxonomy; use the Preview brightness/contrast trick |
| Retraction Watch DB (Crossref-hosted) | database / STATUS | Retraction status, reason codes, retraction DOI | Query Author/Journal/DOI in UI; bulk via Crossref Labs API |
| ORI case summaries | database / STATUS | US PHS adjudicated misconduct findings (active actions) | Search researcher name; confirm via Federal Register / NIH Guide |
| PubMed | database / STATUS | Retraction/EoC/correction banners | Read the record banner; filter pub-type "Retracted Publication" |
| scite.ai | database / STATUS+signal | Retraction/notice flags; cited-but-retracted refs | DOI lookup or Reference Check on a PDF |
| Crossref / Crossmark | database / STATUS | Authoritative publisher update status | REST API by DOI; check `update-to` / Crossmark |
## Red-flag → Verification-platform Routing Table
| Red flag (input) | Primary platform | How to query | Confirm with |
|------------------|------------------|--------------|--------------|
| Suspected image duplication (blot/micrograph/FACS) | PubPeer | Search DOI/author; read prior overlay analyses | Replay overlay/flip yourself; classify Bik I/II/III |
| Same panel across DIFFERENT papers / template motif | For Better Science + PubPeer | Search motif/author; look for cross-author clusters | Check co-author/affiliation plausibility (M5) |
| Suspected impossible mean/SD | apply GRIM/GRIMMER directly | recompute granularity vs N | Data Colada for analogous distributional tests |
| Excel supplied, suspect manipulation | Data Colada method | OP-CALCCHAIN: unzip, read calcChain.xml | fonts, RANDBETWEEN-uniform distribution; note sorting caveat |
| Suspect already retracted | Retraction Watch DB | Author/Journal/RetractionDOI | PubMed banner + Crossref/Crossmark |
| Suspect correction/EoC (not full retraction) | PubMed + Crossref/Crossmark | record banner / Crossmark `update-to` | Retraction Watch blog narrative |
| Suspect adjudicated misconduct (US, PHS) | ORI case summaries | search researcher name | Federal Register / NIH Guide full text |
| Relying on a possibly-retracted reference | scite.ai | DOI lookup / Reference Check on PDF | Crossref canonical status |
| Author has a pattern | Retraction Watch DB (author) + PubPeer (author) | author query in both | For Better Science for lab-level narrative |
## Seed list — tortured phrases (extend from Problematic Paper Screener)
`bosom peril` = breast cancer · `counterfeit consciousness` = artificial intelligence ·
`mean square blunder` = mean square error · `flag to clamor`/`flag to commotion` = signal-to-noise ·
`profound neural organization` = deep neural network · `amino corrosive` = amino acid ·
`irregular esteem` = random value · `lung malignancy` = lung cancer (when paired with other tells).
Any hit is a strong text-layer flag; bundle with other signals (ESL writing alone is not a mill).
## Seed list — LLM-leakage strings
`as an AI language model` · `as a large language model` · `I'm sorry, but as an AI` ·
`Certainly, here is` · `regenerate response` · `I cannot fulfill this request` (in body text) ·
fabricated/phantom references that resolve to nothing on Crossref/PubMed.
## Banned words in Tier-1/2 output (M6 filter)
fraud · fraudulent · fabricated · faked · falsified · doctored · manipulated (as a verdict) ·
misconduct · lied · cheated · guilty. — Allowed only inside a quoted Tier-3 adjudication.
Allowed neutral verbs: appears · shows · is consistent with · is identical to · overlaps ·
cannot be explained by · warrants clarification.
## Named sleuth specialties (route the flag to the matching technique)
- **Elisabeth Bik** — visual image-duplication; Type I/II/III; large-scale eyeballing.
- **Smut Clyde** — image duplication + paper-mill motif clustering.
- **Tiger BB8** — Chinese paper-mill linguistics/affiliation patterns.
- **Cheshire** — anonymous tip intake; PubPeer flagging.
- **Nick Brown** — statistical consistency (GRIM/GRIMMER) in psych/nutrition.
- **Data Colada (Simonsohn/Nelson/Simmons)** — spreadsheet + distributional data forensics.
## Field-branching rule
Biology/medicine → image-forensics path (blot overlay + Bik taxonomy). Psych/econ/nutrition →
statistical-forensics path (GRIM/GRIMMER + calcChain). Cancer molecular biology / gene-knockdown
preclinical → highest paper-mill base rate (run M5 reagent + image layers).
*Sources: full URLs and confidence ratings in `R01-…` through `R07-…`. Information current to May 2026.*
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |