{"paper":{"title":"DOCFORGE-BENCH: A Comprehensive 0-shot Benchmark for Document Forgery Detection and Analysis","arxiv_id":"2603.01433","primary_cat":"cs.CV","submitted_at":"2026-03-02T04:26:57+00:00","version_compiled":"v3"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v3","sentences":[{"i":0,"p":1.0,"id":"s0","own":true,"role":"contribution","text":"We present DOCFORGE-BENCH, the first unified zero-shot benchmark for document forgery detection, evaluating 14 methods across eight datasets spanning text tampering, receipt forgery, and identity document manipulation.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.06,"quantitative":true,"quantitative_p":0.93,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":1.0,"interpretation":0.0}},{"i":1,"p":0.77,"id":"s1","own":true,"role":"method","text":"Unlike fine-tuning-oriented evaluations such as ForensicHub [Du et al., 2025], DOCFORGE-BENCH applies all methods with their published pretrained weights and no domain adaptation -- a deliberate design choice that reflects the realistic deployment scenario where practitioners lack labeled document training data.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.78,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.49,"role_probabilities":{"gap":0.0,"method":0.77,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.23,"interpretation":0.0}},{"i":2,"p":1.0,"id":"s2","own":true,"role":"result","text":"Our central finding is a pervasive calibration failure invisible under single-threshold protocols: methods achieve moderate Pixel-AUC (>=0.76) yet near-zero Pixel-F1.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.79,"quantitative":true,"quantitative_p":0.97,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":3,"p":0.88,"id":"s3","own":true,"role":"interpretation","text":"This AUC-F1 gap is not a discrimination failure but a score-distribution shift: tampered regions occupy only 0.27-4.17% of pixels in document images -- an order of magnitude less than in natural image benchmarks -- making the standard tau=0.5 threshold catastrophically miscalibrated.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.89,"main_claim_p":0.21,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.12,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.88}},{"i":4,"p":0.52,"id":"s4","own":true,"role":"result","text":"Oracle-F1 is 2-10x higher than fixed-threshold Pixel-F1, confirming that calibration, not representation, is the bottleneck.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.14,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.52,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.48}},{"i":5,"p":0.73,"id":"s5","own":true,"role":"result","text":"A controlled calibration experiment validates this: adapting a single threshold on N=10 domain images recovers 39-55% of the Oracle-F1 gap, demonstrating that threshold adaptation -- not retraining -- is the key missing step for practical deployment.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.12,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.01,"result":0.73,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.01,"interpretation":0.25}},{"i":6,"p":0.81,"id":"s6","own":true,"role":"interpretation","text":"Overall, no evaluated method works reliably out-of-the-box on diverse document types, underscoring that document forgery detection remains an unsolved problem.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.84,"main_claim_p":0.49,"quantitative":false,"quantitative_p":0.06,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.19,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.81}},{"i":7,"p":0.55,"id":"s7","own":false,"role":"future_work","text":"We further note that all eight datasets predate the era of generative AI editing; benchmarks covering diffusion- and LLM-based document forgeries represent a critical open gap on the modern attack surface.","hedge":{"p":0.94,"label":"asserted"},"own_p":0.3,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.24,"role_probabilities":{"gap":0.06,"method":0.0,"result":0.0,"background":0.0,"limitation":0.39,"future_work":0.55,"contribution":0.0,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":0.51,"text":"Overall, no evaluated method works reliably out-of-the-box on diverse document types, underscoring that document forgery detection remains an unsolved problem.","location":"abstract s6","hedge_strength":{"p":1.0,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":6631,"cost_usd_reported":0.0},"compiled_at":"2026-10-27","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.95,"states_limitation":0.93,"human_participants":0.04,"controlled_comparison":0.43,"randomized_assignment":0.01,"states_quantitative_result":0.99,"compares_with_prior_results":0.9,"mentions_code_or_data_availability":0.04},"evidential_basis":{"p":0.87,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.12,"observation_or_experiment":0.87,"simulation_or_computation":0.01}},"contribution_type":{"p":1.0,"value":"dataset or benchmark","probabilities":{"review":0.0,"method_or_tool":0.0,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":1.0,"position_or_commentary":0.0}},"contribution_statement":{"p":1.0,"text":"We present DOCFORGE-BENCH, the first unified zero-shot benchmark for document forgery detection, evaluating 14 methods across eight datasets spanning text tampering, receipt forgery, and identity document manipulation.","location":"abstract s0"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v3"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":8,"sentences_in_abstract":8},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}