Release 1.2.0 audit enhancements

Author: nufegia

CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## v1.2.0
+
+- Added public report export support with schema metadata and CLI coverage.
+- Expanded project audit checks for references, image signals, routing, raw data, and summary-stat crosschecks.
+- Improved benchmark fixtures, benchmark reports, and documentation for the updated audit behavior.
+- Added development tooling configuration for pytest, ruff, mypy, and type stubs.
+
 ## v1.1.0
 
 - Renamed the project to `pre-check-research` and standardized the `pcr` CLI/output prefix.

README.md

@@ -75,7 +75,7 @@ The report is designed to support a defensible review process. It does not repla
 | **P-values** | `p_value_collection` | Domain validity (p outside [0,1]), just-significant clustering |
 | **Statistical text** | `statcheck` | APA/NHST in-text statistic vs reported p-value consistency |
 | **Images** | `image_audit` | Internal duplicates (aHash/dHash/pHash), rotated/flipped copies, copy-move triage, western blot/gel review |
-| **References** | `reference_audit` | DOI/PMID parsing, Crossref/OpenAlex/NCBI metadata queries, citation claim extraction |
+| **References** | `reference_audit` | DOI/PMID parsing, Crossref/OpenAlex/PubPeer/NCBI metadata queries, citation claim extraction |
 | **Code** | `code_audit`, sandbox | Pattern scanning (hardcoded paths, exclusion clues), Python/R script rerun with output capture |
 | **Corpus** | `corpus_signals` | Cross-manuscript text similarity (simhash, Jaccard), reference overlap, papermill phrase signals |
 | **Provenance** | `provenance` | SHA-256 file hashing, append-only JSONL ledger, verify/diff change detection |
@@ -283,7 +283,7 @@ Built-in example projects for testing and demonstration:
 ## Privacy and Security
 
 - **All computation is local.** No data is uploaded to external services.
-- External lookups (Crossref, OpenAlex, NCBI) only query public identifiers (DOI, PMID) and can be disabled with `--no-external-lookups`.
+- External lookups (Crossref, OpenAlex, PubPeer, NCBI) only query public identifiers (DOI, PMID) and can be disabled with `--no-external-lookups`.
 - Code reruns execute in temporary project copies with timeouts and minimal environment variables. This is not a strong security sandbox — treat unknown code accordingly.
 - SHA-256 provenance ledgers are append-only and never transmit file contents.
 

benchmark/BENCHMARK.md

@@ -20,7 +20,7 @@ From the repo root:
 python3 benchmark/run_benchmark.py
 ```
 
-To skip external Crossref/OpenAlex/NCBI calls:
+To skip external Crossref/OpenAlex/PubPeer/NCBI calls:
 
 ```bash
 python3 benchmark/run_benchmark.py --no-network
@@ -36,7 +36,7 @@ python3 benchmark/run_benchmark.py --regenerate
 
 The suite covers raw data rules, including digit distribution, high-similarity rows/columns, column relationships, rare categories, and ordinal concentration; summary-stat crosscheck; R scrutiny; R statcheck; R rsprite2; p-value collection checks; reference parsing; external metadata lookup; citation claim extraction; papermill light/network signals; image duplicate/copy-move/metadata review; code scan/rerun; unsupported code recording; data trace crosscheck; provenance record/verify; and local corpus screening.
 
-Network coverage uses `inputs/project_external` and expects evidence from Crossref, OpenAlex, and NCBI. Network failures should be interpreted separately from detector regressions because external APIs can be unavailable, rate-limited, or return changed metadata.
+Network coverage uses `inputs/project_external` and expects evidence from Crossref, OpenAlex, PubPeer, and NCBI. Network failures should be interpreted separately from detector regressions because external APIs can be unavailable, rate-limited, require credentials, or return changed metadata.
 
 ## Interpretation
 

benchmark/BENCHMARK_REPORT.md

@@ -16,7 +16,7 @@ Conclusion: The core detection pipeline is stably covered by automated benchmark
 - Raw data: Covers duplicate/highly similar rows and columns, fixed steps, high-frequency values, missing-concentrated-by-group, terminal digit distribution, inter-column relationships, and non-continuous variable anomalies; clean controls maintain 0 risk signals.
 - Summary statistics: Covers SE/SD/N, CI, percent/count, p/t/df, p-value domain, and R scrutiny/SPRITE feasibility checks.
 - In-text statistics: Covers R statcheck p-value consistency checks on APA/NHST expressions.
-- Literature & network: Covers DOI/PMID parsing, Crossref/OpenAlex/NCBI metadata queries, and citation claim extraction.
+- Literature & network: Covers DOI/PMID parsing, Crossref/OpenAlex/PubPeer/NCBI metadata queries, and citation claim extraction.
 - Images: Covers image discovery, internal duplicates, local copy-move, metadata quality, and Western blot/gel review checklist.
 - Code & project: Covers Python/R script reruns, Stata/SPSS/SAS read-only prompts, cross-material data reconciliation, project manifest, provenance version chain, and local corpus screening.
 
@@ -36,18 +36,18 @@ Not executed (--no-network used).
 
 | Case | Type | Pass | Seconds | Risk Signals | Info | Missing Tools | Missing Checks |
 |---|---:|---:|---:|---:|---|---|
-| raw_suspicious | single_run | Yes | 1.16 | 16 | 0 |  |  |
-| raw_clean_control | single_run | Yes | 1.046 | 0 | 0 |  |  |
-| summary_suspicious | single_run | Yes | 2.083 | 17 | 2 |  |  |
-| p_values_suspicious | single_run | Yes | 1.012 | 2 | 0 |  |  |
-| apa_stats_suspicious | single_run | Yes | 2.193 | 2 | 0 |  |  |
-| paper_refs_and_claims_offline | single_run | Yes | 1.011 | 0 | 4 |  |  |
-| analysis_suspicious | single_run | Yes | 1.378 | 1 | 1 |  |  |
-| analysis_manual_unsupported | single_run | Yes | 1.012 | 0 | 3 |  |  |
-| figures_project | project | Yes | 1.205 | 11 | 13 |  |  |
-| project_full | project | Yes | 2.555 | 12 | 19 |  |  |
-| corpus_screen | corpus | Yes | 2.355 | 4 | 0 |  |  |
-| provenance_change | provenance_change | Yes | 2.067 | 1 | 5 |  |  |
+| raw_suspicious | single_run | Yes | 1.284 | 16 | 0 |  |  |
+| raw_clean_control | single_run | Yes | 1.147 | 0 | 0 |  |  |
+| summary_suspicious | single_run | Yes | 2.279 | 17 | 2 |  |  |
+| p_values_suspicious | single_run | Yes | 1.039 | 2 | 0 |  |  |
+| apa_stats_suspicious | single_run | Yes | 2.156 | 2 | 0 |  |  |
+| paper_refs_and_claims_offline | single_run | Yes | 1.072 | 0 | 4 |  |  |
+| analysis_suspicious | single_run | Yes | 1.42 | 1 | 1 |  |  |
+| analysis_manual_unsupported | single_run | Yes | 1.054 | 0 | 3 |  |  |
+| figures_project | project | Yes | 1.201 | 11 | 13 |  |  |
+| project_full | project | Yes | 2.471 | 12 | 19 |  |  |
+| corpus_screen | corpus | Yes | 2.104 | 4 | 0 |  |  |
+| provenance_change | provenance_change | Yes | 2.072 | 1 | 5 |  |  |
 | external_refs_online | project_network | Yes | 0.0 | 0 | 0 |  |  |
 
 ## Tool Coverage
@@ -93,5 +93,5 @@ Not executed (--no-network used).
 ## Interpretation Boundaries
 
 The high/medium/low levels in this report are benchmark risk signals, not conclusions of academic misconduct, fabrication, or fraud. `info` records are run statuses, dependency states, skip reasons, or coverage notes; they do not count toward risk conclusions.
-Network test cases depend on real-time availability, certificate chains, and rate limiting of Crossref, OpenAlex, and NCBI. If network cases fail, first check HTTP/SSL/rate-limit information in evidence before concluding it is a detector regression.
+Network test cases depend on real-time availability, certificate chains, credentials, and rate limiting of Crossref, OpenAlex, PubPeer, and NCBI. If network cases fail, first check HTTP/SSL/rate-limit information in evidence before concluding it is a detector regression.
 All weak-signal tools are only for surfacing human review directions. Final review should return to original data, scripts, image source files, literature metadata, and audit logs.

benchmark/benchmark_manifest.json

@@ -110,8 +110,8 @@
       "kind": "project_network",
       "input": "inputs/project_external",
       "expected_tools": ["reference_audit", "citation_claim_check", "provenance_hash"],
-      "expected_checks": ["DOI title mismatch", "PMID title mismatch", "DOI external metadata unverifiable", "PMID metadata verification"],
-      "expected_external_services": ["crossref", "openalex", "ncbi"],
+      "expected_checks": ["DOI title mismatch", "PMID title mismatch", "DOI external metadata absent", "PMID metadata verification"],
+      "expected_external_services": ["crossref", "openalex", "pubpeer", "ncbi"],
       "min_risk_findings": 3
     }
   ]

benchmark/generate_synthetic_benchmark.py

@@ -1,7 +1,6 @@
 from __future__ import annotations
 
 import json
-import math
 import shutil
 from pathlib import Path
 
@@ -283,7 +282,7 @@ def write_ground_truth() -> None:
                 "known_limitations": [
                     "R tool output depends on statcheck/scrutiny/rsprite2 parsing of column names and text formats.",
                     "Image copy-move is a weak signal; low-texture or regularly repeating graphics may produce false positives/negatives.",
-                    "Project-level external reference queries are disabled in this benchmark; Crossref/OpenAlex/NCBI network reliability is not tested.",
+                    "Project-level external reference queries are disabled in this benchmark; Crossref/OpenAlex/PubPeer/NCBI network reliability is not tested.",
                 ],
             },
             ensure_ascii=False,

benchmark/ground_truth.json

@@ -23,6 +23,6 @@
   "known_limitations": [
     "R tool output depends on statcheck/scrutiny/rsprite2 parsing of column names and text formats.",
     "Image copy-move is a weak signal; low-texture or regularly repeating graphics may produce false positives/negatives.",
-    "Project-level external reference queries are disabled in this benchmark; Crossref/OpenAlex/NCBI network reliability is not tested."
+    "Project-level external reference queries are disabled in this benchmark; Crossref/OpenAlex/PubPeer/NCBI network reliability is not tested."
   ]
-}
+}

benchmark/reports/pcr.benchmark_summary.json

@@ -7,7 +7,7 @@
       "kind": "single_run",
       "ok": true,
       "returncode": 0,
-      "seconds": 1.16,
+      "seconds": 1.284,
       "json_path": "benchmark/reports/pcr.raw_suspicious.json",
       "markdown_path": "benchmark/reports/pcr.raw_suspicious.md",
       "risk_findings": 16,
@@ -38,7 +38,7 @@
       "kind": "single_run",
       "ok": true,
       "returncode": 0,
-      "seconds": 1.046,
+      "seconds": 1.147,
       "json_path": "benchmark/reports/pcr.raw_clean_control.json",
       "markdown_path": "benchmark/reports/pcr.raw_clean_control.md",
       "risk_findings": 0,
@@ -56,7 +56,7 @@
       "kind": "single_run",
       "ok": true,
       "returncode": 0,
-      "seconds": 2.083,
+      "seconds": 2.279,
       "json_path": "benchmark/reports/pcr.summary_suspicious.json",
       "markdown_path": "benchmark/reports/pcr.summary_suspicious.md",
       "risk_findings": 17,
@@ -90,7 +90,7 @@
       "kind": "single_run",
       "ok": true,
       "returncode": 0,
-      "seconds": 1.012,
+      "seconds": 1.039,
       "json_path": "benchmark/reports/pcr.p_values_suspicious.json",
       "markdown_path": "benchmark/reports/pcr.p_values_suspicious.md",
       "risk_findings": 2,
@@ -113,7 +113,7 @@
       "kind": "single_run",
       "ok": true,
       "returncode": 0,
-      "seconds": 2.193,
+      "seconds": 2.156,
       "json_path": "benchmark/reports/pcr.apa_stats_suspicious.json",
       "markdown_path": "benchmark/reports/pcr.apa_stats_suspicious.md",
       "risk_findings": 2,
@@ -135,7 +135,7 @@
       "kind": "single_run",
       "ok": true,
       "returncode": 0,
-      "seconds": 1.011,
+      "seconds": 1.072,
       "json_path": "benchmark/reports/pcr.paper_refs_and_claims_offline.json",
       "markdown_path": "benchmark/reports/pcr.paper_refs_and_claims_offline.md",
       "risk_findings": 0,
@@ -162,7 +162,7 @@
       "kind": "single_run",
       "ok": true,
       "returncode": 0,
-      "seconds": 1.378,
+      "seconds": 1.42,
       "json_path": "benchmark/reports/pcr.analysis_suspicious.json",
       "markdown_path": "benchmark/reports/pcr.analysis_suspicious.md",
       "risk_findings": 1,
@@ -186,7 +186,7 @@
       "kind": "single_run",
       "ok": true,
       "returncode": 0,
-      "seconds": 1.012,
+      "seconds": 1.054,
       "json_path": "benchmark/reports/pcr.analysis_manual_unsupported.json",
       "markdown_path": "benchmark/reports/pcr.analysis_manual_unsupported.md",
       "risk_findings": 0,
@@ -211,7 +211,7 @@
       "kind": "project",
       "ok": true,
       "returncode": 0,
-      "seconds": 1.205,
+      "seconds": 1.201,
       "json_path": "benchmark/reports/pcr.figures_project.json",
       "markdown_path": "benchmark/reports/pcr.figures_project.md",
       "risk_findings": 11,
@@ -255,7 +255,7 @@
       "kind": "project",
       "ok": true,
       "returncode": 0,
-      "seconds": 2.555,
+      "seconds": 2.471,
       "json_path": "benchmark/reports/pcr.project_full.json",
       "markdown_path": "benchmark/reports/pcr.project_full.md",
       "risk_findings": 12,
@@ -308,7 +308,7 @@
       "kind": "corpus",
       "ok": true,
       "returncode": 0,
-      "seconds": 2.355,
+      "seconds": 2.104,
       "json_path": "benchmark/reports/pcr.corpus_screen.json",
       "markdown_path": "benchmark/reports/pcr.corpus_screen.md",
       "risk_findings": 4,
@@ -332,7 +332,7 @@
       "kind": "provenance_change",
       "ok": true,
       "returncode": 0,
-      "seconds": 2.067,
+      "seconds": 2.072,
       "json_path": "benchmark/reports/pcr.provenance_change.json",
       "markdown_path": ".",
       "risk_findings": 1,

benchmark/reports/pcr.benchmark_summary.md

@@ -16,7 +16,7 @@ Conclusion: The core detection pipeline is stably covered by automated benchmark
 - Raw data: Covers duplicate/highly similar rows and columns, fixed steps, high-frequency values, missing-concentrated-by-group, terminal digit distribution, inter-column relationships, and non-continuous variable anomalies; clean controls maintain 0 risk signals.
 - Summary statistics: Covers SE/SD/N, CI, percent/count, p/t/df, p-value domain, and R scrutiny/SPRITE feasibility checks.
 - In-text statistics: Covers R statcheck p-value consistency checks on APA/NHST expressions.
-- Literature & network: Covers DOI/PMID parsing, Crossref/OpenAlex/NCBI metadata queries, and citation claim extraction.
+- Literature & network: Covers DOI/PMID parsing, Crossref/OpenAlex/PubPeer/NCBI metadata queries, and citation claim extraction.
 - Images: Covers image discovery, internal duplicates, local copy-move, metadata quality, and Western blot/gel review checklist.
 - Code & project: Covers Python/R script reruns, Stata/SPSS/SAS read-only prompts, cross-material data reconciliation, project manifest, provenance version chain, and local corpus screening.
 
@@ -36,18 +36,18 @@ Not executed (--no-network used).
 
 | Case | Type | Pass | Seconds | Risk Signals | Info | Missing Tools | Missing Checks |
 |---|---:|---:|---:|---:|---|---|
-| raw_suspicious | single_run | Yes | 1.16 | 16 | 0 |  |  |
-| raw_clean_control | single_run | Yes | 1.046 | 0 | 0 |  |  |
-| summary_suspicious | single_run | Yes | 2.083 | 17 | 2 |  |  |
-| p_values_suspicious | single_run | Yes | 1.012 | 2 | 0 |  |  |
-| apa_stats_suspicious | single_run | Yes | 2.193 | 2 | 0 |  |  |
-| paper_refs_and_claims_offline | single_run | Yes | 1.011 | 0 | 4 |  |  |
-| analysis_suspicious | single_run | Yes | 1.378 | 1 | 1 |  |  |
-| analysis_manual_unsupported | single_run | Yes | 1.012 | 0 | 3 |  |  |
-| figures_project | project | Yes | 1.205 | 11 | 13 |  |  |
-| project_full | project | Yes | 2.555 | 12 | 19 |  |  |
-| corpus_screen | corpus | Yes | 2.355 | 4 | 0 |  |  |
-| provenance_change | provenance_change | Yes | 2.067 | 1 | 5 |  |  |
+| raw_suspicious | single_run | Yes | 1.284 | 16 | 0 |  |  |
+| raw_clean_control | single_run | Yes | 1.147 | 0 | 0 |  |  |
+| summary_suspicious | single_run | Yes | 2.279 | 17 | 2 |  |  |
+| p_values_suspicious | single_run | Yes | 1.039 | 2 | 0 |  |  |
+| apa_stats_suspicious | single_run | Yes | 2.156 | 2 | 0 |  |  |
+| paper_refs_and_claims_offline | single_run | Yes | 1.072 | 0 | 4 |  |  |
+| analysis_suspicious | single_run | Yes | 1.42 | 1 | 1 |  |  |
+| analysis_manual_unsupported | single_run | Yes | 1.054 | 0 | 3 |  |  |
+| figures_project | project | Yes | 1.201 | 11 | 13 |  |  |
+| project_full | project | Yes | 2.471 | 12 | 19 |  |  |
+| corpus_screen | corpus | Yes | 2.104 | 4 | 0 |  |  |
+| provenance_change | provenance_change | Yes | 2.072 | 1 | 5 |  |  |
 | external_refs_online | project_network | Yes | 0.0 | 0 | 0 |  |  |
 
 ## Tool Coverage
@@ -93,5 +93,5 @@ Not executed (--no-network used).
 ## Interpretation Boundaries
 
 The high/medium/low levels in this report are benchmark risk signals, not conclusions of academic misconduct, fabrication, or fraud. `info` records are run statuses, dependency states, skip reasons, or coverage notes; they do not count toward risk conclusions.
-Network test cases depend on real-time availability, certificate chains, and rate limiting of Crossref, OpenAlex, and NCBI. If network cases fail, first check HTTP/SSL/rate-limit information in evidence before concluding it is a detector regression.
+Network test cases depend on real-time availability, certificate chains, credentials, and rate limiting of Crossref, OpenAlex, PubPeer, and NCBI. If network cases fail, first check HTTP/SSL/rate-limit information in evidence before concluding it is a detector regression.
 All weak-signal tools are only for surfacing human review directions. Final review should return to original data, scripts, image source files, literature metadata, and audit logs.