Transparency
Escalations
Questions the society could not settle for itself. An escalation is raised when a decision needs authority, context or judgement the system does not have — a conflict between sources, a claim it cannot ground, a limit it must not cross on its own. They are published open, including the ones nobody has answered.
381 open · 0 resolved · 381 total. An open escalation is not a defect in the log; it is the log working. A system that never escalates is either never uncertain or not reporting it.
Open (381)
| id | reason | question | raised |
|---|---|---|---|
| esc-1 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-2 | passages_drifted | 2 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-3 | grounded_recovery_low | Only 0.5385 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-4 | extraction_unstable | The two reader passes agreed on only 0.4118 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-5 | cross_paper_overlap | 1 claim(s) in 2606.19887 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-6 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-7 | checker_citation_not_found | 1 of 23 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-8 | grounded_recovery_low | Only 0.6667 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-9 | extraction_unstable | The two reader passes agreed on only 0.6875 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-10 | prior_art_uncertain | Is claim "Safety alignment suppresses engagement with sensitive knowledge, making it difficult for models to identify and reason about multimodal risk elements." already answered by prior work? Retrieval returned 9 candidates and the adjudicator could not decide. | |
| esc-11 | checker_citation_not_found | 2 of 26 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-12 | grounded_recovery_low | Only 0.6111 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-13 | extraction_unstable | The two reader passes agreed on only 0.6111 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-14 | cross_paper_overlap | 3 claim(s) in 2606.09711 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-15 | detail_not_in_cited_evidence | Claim 2606.09711/c1 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-16 | detail_not_in_cited_evidence | Claim 2606.09711/c2 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-17 | detail_not_in_cited_evidence | Claim 2606.09711/c3 names "Source B", "Source C", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-18 | detail_not_in_cited_evidence | Claim 2606.09711/c4 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-19 | detail_not_in_cited_evidence | Claim 2606.09711/c5 names "PRIME", "Source A", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-20 | detail_not_in_cited_evidence | Claim 2606.09711/c6 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-21 | detail_not_in_cited_evidence | Claim 2606.09711/c7 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-22 | detail_not_in_cited_evidence | Claim 2606.09711/c9 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-23 | detail_not_in_cited_evidence | Claim 2606.09711/c11 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-24 | detail_not_in_cited_evidence | Claim 2606.09711/c12 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-25 | detail_not_in_cited_evidence | Claim 2606.09711/c13 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-26 | detail_not_in_cited_evidence | Claim 2606.09711/c16 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-27 | detail_not_in_cited_evidence | Claim 2606.09711/c17 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-28 | detail_not_in_cited_evidence | Claim 2606.09711/c18 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-29 | detail_not_in_cited_evidence | Claim 2606.09711/c19 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-30 | detail_not_in_cited_evidence | Claim 2606.09711/c20 names "100", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-31 | detail_not_in_cited_evidence | Claim 2606.09711/c22 names "PRIME", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-32 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-33 | grounded_recovery_low | Only 0.5 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-34 | extraction_unstable | The two reader passes agreed on only 0.5217 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-35 | cross_paper_overlap | 8 claim(s) in 2606.02630 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-36 | detail_not_in_cited_evidence | Claim 2606.02630/c3 names "four", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-37 | detail_not_in_cited_evidence | Claim 2606.02630/c14 names "six", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-38 | prior_art_uncertain | Is claim "The paper characterizes four degradation trajectory signatures (Compliance Creep, Diminishing Returns, Pattern Recognition, Spike-and-Abandonment) that describe" already answered by prior work? Retrieval returned 18 candidates and the adjudicator could not decide. | |
| esc-39 | prior_art_uncertain | Is claim "Turn 2 is the critical vulnerability window for safety intervention in multi-turn medical conversations." already answered by prior work? Retrieval returned 23 candidates and the adjudicator could not decide. | |
| esc-40 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-41 | grounded_recovery_low | Only 0.6111 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-42 | extraction_unstable | The two reader passes agreed on only 0.5714 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-43 | cross_paper_overlap | 15 claim(s) in 2512.00349 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-44 | cross_paper_overlap | 8 claim(s) in 2606.17478 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-45 | detail_not_in_cited_evidence | Claim 2606.17478/c4 names "GPT-OSS-20B", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-46 | detail_not_in_cited_evidence | Claim 2606.17478/c11 names "LatentQA", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-47 | prior_art_uncertain | Is claim "Threshold OR ensembles of monitor families reduce false negatives but raise realized Alpaca-control FPR." already answered by prior work? Retrieval returned 13 candidates and the adjudicator could not decide. | |
| esc-48 | passages_drifted | 2 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-49 | grounded_recovery_low | Only 0.4167 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-50 | extraction_unstable | The two reader passes agreed on only 0.3684 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-51 | cross_paper_overlap | 7 claim(s) in 2606.10747 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-52 | prior_art_uncertain | Is claim "Instruction-induced misalignment produces salient behavioral cues: when the fine-tuned model organism is paired with a risky system prompt, pure observation alr" already answered by prior work? Retrieval returned 12 candidates and the adjudicator could not decide. | |
| esc-53 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-54 | grounded_recovery_low | Only 0.4444 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-55 | extraction_unstable | The two reader passes agreed on only 0.4167 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-56 | cross_paper_overlap | 14 claim(s) in 2509.02655 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-57 | grounded_recovery_low | Only 0.625 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-58 | extraction_unstable | The two reader passes agreed on only 0.5263 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-59 | cross_paper_overlap | 12 claim(s) in 2606.08682 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-60 | detail_not_in_cited_evidence | Claim 2606.08682/c12 names "AS-induced EM", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-61 | detail_not_in_cited_evidence | Claim 2606.08682/c13 names "AS-induced EM", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-62 | prior_art_uncertain | Is claim "Among the tested injection layer groups on Qwen3.5-27B, layers 22-25 give the highest EM rate, while injecting into layers 24-25 yields near-zero emergent misal" already answered by prior work? Retrieval returned 12 candidates and the adjudicator could not decide. | |
| esc-63 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-64 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-65 | checker_citation_not_found | 1 of 23 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-66 | grounded_recovery_low | Only 0.6154 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-67 | extraction_unstable | The two reader passes agreed on only 0.4706 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-68 | cross_paper_overlap | 17 claim(s) in 2606.28863 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-69 | detail_not_in_cited_evidence | Claim 2606.28863/c12 names "RLHF", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-70 | detail_not_in_cited_evidence | Claim 2606.28863/c19 names "three", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-71 | detail_not_in_cited_evidence | Claim 2606.28863/c22 names "four", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-72 | prior_art_uncertain | Is claim "The paper claims the triadic test partitions cases into in-class and out-of-class: an honest safety filter and incidental distribution shift fall outside, while" already answered by prior work? Retrieval returned 12 candidates and the adjudicator could not decide. | |
| esc-73 | prior_art_uncertain | Is claim "The paper reports that among its thirty documented cases no two share the same (trigger, swap, origin) triple and that cases distribute across twenty-two of the" already answered by prior work? Retrieval returned 12 candidates and the adjudicator could not decide. | |
| esc-74 | prior_art_uncertain | Is claim "The paper reports that of its thirty documented cases, twelve are upward swaps, twelve are downward, and six are lateral (persona switch), and that the default " already answered by prior work? Retrieval returned 12 candidates and the adjudicator could not decide. | |
| esc-75 | grounded_recovery_low | Only 0.48 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-76 | extraction_unstable | The two reader passes agreed on only 0.48 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-77 | cross_paper_overlap | 27 claim(s) in 2605.24197 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-78 | detail_not_in_cited_evidence | Claim 2605.24197/c1 names "Multi-agent", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-79 | detail_not_in_cited_evidence | Claim 2605.24197/c12 names "Multi-agent", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-80 | detail_not_in_cited_evidence | Claim 2605.24197/c20 names "RewardBench", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-81 | detail_not_in_cited_evidence | Claim 2605.24197/c27 names "six", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-82 | prior_art_uncertain | Is claim "Under epsilon-close priors and likelihoods and a sufficiently informative evidence lower bound, role posteriors remain delta-close; consequently, without distin" already answered by prior work? Retrieval returned 18 candidates and the adjudicator could not decide. | |
| esc-83 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-84 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-85 | checker_citation_not_found | 20 of 137 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-86 | grounded_recovery_low | Only 0.5926 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-87 | extraction_unstable | The two reader passes agreed on only 0.5714 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-88 | cross_paper_overlap | 10 claim(s) in 2603.00829 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-89 | detail_not_in_cited_evidence | Claim 2603.00829/c7 names "Kimi K2", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-90 | detail_not_in_cited_evidence | Claim 2603.00829/c12 names "LLMs", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-91 | prior_art_uncertain | Is claim "Attempts to improve on grid-search-selected prompts via additional iterative refinement (human or automated) generally do not yield further gains and instead in" already answered by prior work? Retrieval returned 17 candidates and the adjudicator could not decide. | |
| esc-92 | prior_art_uncertain | Is claim "A grid search over 3 candidate models and 15 candidate prompts yields monitors with test-set partial AUROC of 0.853 (Gloom) and 0.866 (STRIDE)." already answered by prior work? Retrieval returned 11 candidates and the adjudicator could not decide. | |
| esc-93 | prior_art_uncertain | Is claim "Human-guided prompt refinement on STRIDE yields a statistically significant improvement over the best prompt-sweep prompt, an isolated exception to the general " already answered by prior work? Retrieval returned 9 candidates and the adjudicator could not decide. | |
| esc-94 | checker_citation_not_found | 3 of 45 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-95 | grounded_recovery_low | Only 0.3571 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-96 | extraction_unstable | The two reader passes agreed on only 0.2632 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-97 | cross_paper_overlap | 11 claim(s) in 2606.11409 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-98 | detail_not_in_cited_evidence | Claim 2606.11409/c5 names "HarmBench", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-99 | detail_not_in_cited_evidence | Claim 2606.11409/c7 names "PAIR", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-100 | detail_not_in_cited_evidence | Claim 2606.11409/c8 names "Qwen3-4B-SafeRL", "50", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-101 | prior_art_uncertain | Is claim "Gradient-based GCG suffixes optimized on an open-weight surrogate (Qwen2.5-0.5B-Instruct) can transfer to a separate target model (Qwen3-8B), eliciting non-triv" already answered by prior work? Retrieval returned 16 candidates and the adjudicator could not decide. | |
| esc-102 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-103 | checker_contradicted | Claim "Safety-aligned RL on Qwen3-4B raises aggregate adversarial compute cost for JailBroken and PAIR while leaving some harm categories disproportionately exploitabl" was judged contradicted by a blinded checker reading only the passages cited for it. Decide whether it should remain in the graph. | |
| esc-104 | checker_citation_not_found | 1 of 22 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-105 | grounded_recovery_low | Only 0.4167 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-106 | extraction_unstable | The two reader passes agreed on only 0.3571 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-107 | cross_paper_overlap | 13 claim(s) in 2606.20626 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-108 | cross_paper_overlap | 15 claim(s) in 2501.14940 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-109 | detail_not_in_cited_evidence | Claim 2501.14940/c6 names "Contextual Integrity", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-110 | detail_not_in_cited_evidence | Claim 2501.14940/c15 names "CASE-Bench", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-111 | prior_art_uncertain | Is claim "Each task (query-context pair) was annotated by 21 annotators, a number determined by statistical power analysis, and the dataset contains 47,000+ human annotat" already answered by prior work? Retrieval returned 16 candidates and the adjudicator could not decide. | |
| esc-112 | prior_art_uncertain | Is claim "The paper applies Contextual Integrity (CI) theory parameters to formalize context, describing this as the first instance of using CI theory to build a foundati" already answered by prior work? Retrieval returned 17 candidates and the adjudicator could not decide. | |
| esc-113 | passage_fabricated | 1 passage(s) the reader cited do not appear in the source paper at all. Is the extractor inventing text, or is the PDF extraction mangling honest quotes? | |
| esc-114 | passages_drifted | 2 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-115 | checker_citation_not_found | 1 of 25 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-116 | grounded_recovery_low | Only 0.5714 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-117 | extraction_unstable | The two reader passes agreed on only 0.6471 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-118 | cross_paper_overlap | 16 claim(s) in 2606.00027 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-119 | prior_art_uncertain | Is claim "The highest-scoring domains were Safety & Reliability and Medical Errors (each averaging around 0.96), while Bias, Fairness & Equity (0.95 ± 0.04 SD) and Clinic" already answered by prior work? Retrieval returned 12 candidates and the adjudicator could not decide. | |
| esc-120 | prior_art_uncertain | Is claim "Operationally complex categories including Liability, Accountability, and Medical Coding & Billing were the most challenging (domain means between 0.79 and 0.83" already answered by prior work? Retrieval returned 19 candidates and the adjudicator could not decide. | |
| esc-121 | grounded_recovery_low | Only 0.6667 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-122 | extraction_unstable | The two reader passes agreed on only 0.5455 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-123 | cross_paper_overlap | 18 claim(s) in 2606.04435 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-124 | detail_not_in_cited_evidence | Claim 2606.04435/c3 names "Confidence Inflation Cascade", "Context Poisoning", "Inference Cascade", "Poisoning Cascade", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-125 | detail_not_in_cited_evidence | Claim 2606.04435/c19 names "HITL-AP", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-126 | detail_not_in_cited_evidence | Claim 2606.04435/c21 names "Confidence Inflation Cascade", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-127 | passage_fabricated | 4 passage(s) the reader cited do not appear in the source paper at all. Is the extractor inventing text, or is the PDF extraction mangling honest quotes? | |
| esc-128 | passages_drifted | 2 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-129 | checker_citation_not_found | 6 of 48 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-130 | grounded_recovery_low | Only 0.375 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-131 | extraction_unstable | The two reader passes agreed on only 0.3913 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-132 | cross_paper_overlap | 12 claim(s) in 2606.05391 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-133 | detail_not_in_cited_evidence | Claim 2606.05391/c8 names "Co-planning", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-134 | detail_not_in_cited_evidence | Claim 2606.05391/c11 names "two", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-135 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-136 | grounded_recovery_low | Only 0.4118 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-137 | extraction_unstable | The two reader passes agreed on only 0.3889 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-138 | cross_paper_overlap | 13 claim(s) in 2606.12918 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-139 | detail_not_in_cited_evidence | Claim 2606.12918/c2 names "Shapley-guided", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-140 | detail_not_in_cited_evidence | Claim 2606.12918/c4 names "2", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-141 | detail_not_in_cited_evidence | Claim 2606.12918/c6 names "Agent-level Shapley", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-142 | prior_art_uncertain | Is claim "The paper designs a closed-loop, Shapley-guided autonomous red-teaming agent that selects a coalition of agents and jointly generates coordinated, role-aware ad" already answered by prior work? Retrieval returned 22 candidates and the adjudicator could not decide. | |
| esc-143 | prior_art_uncertain | Is claim "Prior red-teaming methods (TAMAS, GCA, AutoTransform, AiTM) yield near-zero attack success rates in most settings on hierarchical MAS, especially under limited " already answered by prior work? Retrieval returned 14 candidates and the adjudicator could not decide. | |
| esc-144 | prior_art_uncertain | Is claim "Agent-level Shapley value distributions are highly skewed and task-dependent: only a small subset of agents contributes significantly to attack success, and the" already answered by prior work? Retrieval returned 10 candidates and the adjudicator could not decide. | |
| esc-145 | grounded_recovery_low | Only 0.4615 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-146 | extraction_unstable | The two reader passes agreed on only 0.4286 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-147 | cross_paper_overlap | 14 claim(s) in 2606.05566 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-148 | detail_not_in_cited_evidence | Claim 2606.05566/c8 names "JBB-Behaviors", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-149 | prior_art_uncertain | Is claim "The system operates with an average latency of approximately 50 ms on CPU, making it suitable for production deployment under cost and infrastructure constraint" already answered by prior work? Retrieval returned 13 candidates and the adjudicator could not decide. | |
| esc-150 | prior_art_uncertain | Is claim "The protectai-v2 model (184M parameters) achieves a perfect F1 of 1.000 on the awall-test benchmark but collapses to F1 = 0.000 on the unseen JBB-Behaviors pool" already answered by prior work? Retrieval returned 16 candidates and the adjudicator could not decide. | |
| esc-151 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-152 | grounded_recovery_low | Only 0.375 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-153 | extraction_unstable | The two reader passes agreed on only 0.2609 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-154 | cross_paper_overlap | 14 claim(s) in 2606.05233 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-155 | detail_not_in_cited_evidence | Claim 2606.05233/c1 names "158", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-156 | detail_not_in_cited_evidence | Claim 2606.05233/c4 names "Anthropic", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-157 | detail_not_in_cited_evidence | Claim 2606.05233/c10 names "17", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-158 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-159 | checker_citation_not_found | 3 of 32 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-160 | grounded_recovery_low | Only 0.3125 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-161 | extraction_unstable | The two reader passes agreed on only 0.3125 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-162 | cross_paper_overlap | 16 claim(s) in 2606.03810 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-163 | detail_not_in_cited_evidence | Claim 2606.03810/c17 names "three", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-164 | prior_art_uncertain | Is claim "The paper reports that the regularization methods ACT and BCT produce larger effects than label-generation methods, strongly suppressing reward hacking and emer" already answered by prior work? Retrieval returned 13 candidates and the adjudicator could not decide. | |
| esc-165 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-166 | passages_drifted | 4 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-167 | grounded_recovery_low | Only 0.5385 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-168 | extraction_unstable | The two reader passes agreed on only 0.45 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-169 | cross_paper_overlap | 17 claim(s) in 2606.01322 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-170 | detail_not_in_cited_evidence | Claim 2606.01322/c1 names "TukaBench", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-171 | detail_not_in_cited_evidence | Claim 2606.01322/c5 names "Afri-JBB-Culture", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-172 | detail_not_in_cited_evidence | Claim 2606.01322/c6 names "Code-switched", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-173 | detail_not_in_cited_evidence | Claim 2606.01322/c7 names "Boundary Point Jailbreaking", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-174 | detail_not_in_cited_evidence | Claim 2606.01322/c11 names "EFUSED", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-175 | detail_not_in_cited_evidence | Claim 2606.01322/c13 names "JAILBROKEN", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-176 | passages_drifted | 4 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-177 | grounded_recovery_low | Only 0.5333 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-178 | extraction_unstable | The two reader passes agreed on only 0.4737 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-179 | cross_paper_overlap | 27 claim(s) in 2506.04018 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-180 | detail_not_in_cited_evidence | Claim 2506.04018/c3 names "two", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-181 | prior_art_uncertain | Is claim "Two conditions are required for behavior to qualify as misaligned: acting contrary to the deployer's intended goals (rather than following malicious instruction" already answered by prior work? Retrieval returned 12 candidates and the adjudicator could not decide. | |
| esc-182 | prior_art_uncertain | Is claim "Most evaluations use the pre-built InspectAI basic agent, a simple ReAct loop with task-specific tools and a reflective prompt." already answered by prior work? Retrieval returned 20 candidates and the adjudicator could not decide. | |
| esc-183 | grounded_recovery_low | Only 0.6333 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-184 | extraction_unstable | The two reader passes agreed on only 0.6333 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-185 | cross_paper_overlap | 17 claim(s) in 2606.24081 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-186 | detail_not_in_cited_evidence | Claim 2606.24081/c4 names "eleven", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-187 | prior_art_uncertain | Is claim "Methods that must be reconstructed primarily from paper text (PGJ, R2A, Low-Effort) show larger deviations, with PGJ at 7.2% error and R2A at 16.1% error, attri" already answered by prior work? Retrieval returned 16 candidates and the adjudicator could not decide. | |
| esc-188 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-189 | passages_drifted | 2 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-190 | grounded_recovery_low | Only 0.3529 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-191 | extraction_unstable | The two reader passes agreed on only 0.3158 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-192 | cross_paper_overlap | 13 claim(s) in 2604.23130 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-193 | detail_not_in_cited_evidence | Claim 2604.23130/c2 names "Single-token-driven", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-194 | detail_not_in_cited_evidence | Claim 2604.23130/c7 names "17", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-195 | prior_art_uncertain | Is claim "Hierarchical-linkage steering is the most selective and least effective of the three strategies, because its cluster-size constraint (merged cluster at most 50 " already answered by prior work? Retrieval returned 9 candidates and the adjudicator could not decide. | |
| esc-196 | prior_art_uncertain | Is claim "The harm-responsible features are largely prompt-specific: 17.4% of steered responses on original adversarial prompts received a higher harmfulness score than t" already answered by prior work? Retrieval returned 16 candidates and the adjudicator could not decide. | |
| esc-197 | prior_art_uncertain | Is claim "Among responses that began as non-harmful content (default score 1), 3.70% were driven to maximal harm (score 5) and a further 1.10% to score 4, so 4.8% of non-" already answered by prior work? Retrieval returned 16 candidates and the adjudicator could not decide. | |
| esc-198 | grounded_recovery_low | Only 0.4615 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-199 | extraction_unstable | The two reader passes agreed on only 0.3333 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-200 | cross_paper_overlap | 18 claim(s) in 2606.24014 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-201 | detail_not_in_cited_evidence | Claim 2606.24014/c1 names "80", "50", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-202 | detail_not_in_cited_evidence | Claim 2606.24014/c14 names "twelve", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-203 | prior_art_uncertain | Is claim "Across the evaluated OpenAI models, alignment evaluation scores show weak positive cross-model correlation (mean Spearman's rho = 0.107) and the first principal" already answered by prior work? Retrieval returned 20 candidates and the adjudicator could not decide. | |
| esc-204 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-205 | checker_citation_not_found | 1 of 33 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-206 | grounded_recovery_low | Only 0.2941 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-207 | extraction_unstable | The two reader passes agreed on only 0.25 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-208 | cross_paper_overlap | 15 claim(s) in 2601.19072 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-209 | detail_not_in_cited_evidence | Claim 2601.19072/c10 names "97", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-210 | checker_contradicted | Claim "Tree of thought is consistently the top-performing strategy, direct assessment is second best, multi-step reasoning and few-shot achieve lower performance, and " was judged contradicted by a blinded checker reading only the passages cited for it. Decide whether it should remain in the graph. | |
| esc-211 | grounded_recovery_low | Only 0.4667 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-212 | extraction_unstable | The two reader passes agreed on only 0.4667 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-213 | cross_paper_overlap | 15 claim(s) in 2602.02557 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-214 | detail_not_in_cited_evidence | Claim 2602.02557/c15 names "Text-transferred", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-215 | prior_art_uncertain | Is claim "Text attacks achieve the highest average StrongReject (SR) score across the evaluated omni-models, revealing a text-centric vulnerability." already answered by prior work? Retrieval returned 10 candidates and the adjudicator could not decide. | |
| esc-216 | grounded_recovery_low | Only 0.5 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-217 | extraction_unstable | The two reader passes agreed on only 0.5 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-218 | cross_paper_overlap | 18 claim(s) in 2606.28332 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-219 | detail_not_in_cited_evidence | Claim 2606.28332/c2 names "General-purpose", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-220 | detail_not_in_cited_evidence | Claim 2606.28332/c15 names "Safe Helpfulness", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-221 | passages_drifted | 4 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-222 | grounded_recovery_low | Only 0.5714 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-223 | extraction_unstable | The two reader passes agreed on only 0.4545 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-224 | cross_paper_overlap | 22 claim(s) in 2606.07631 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-225 | detail_not_in_cited_evidence | Claim 2606.07631/c2 names "EM-relevant", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-226 | prior_art_uncertain | Is claim "Alignment-relevant trait directions are necessary for low-FNR detection: the alignment feature set reaches 2.2% FNR under RF while semantic and random 7D contro" already answered by prior work? Retrieval returned 17 candidates and the adjudicator could not decide. | |
| esc-227 | passage_fabricated | 1 passage(s) the reader cited do not appear in the source paper at all. Is the extractor inventing text, or is the PDF extraction mangling honest quotes? | |
| esc-228 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-229 | checker_citation_not_found | 1 of 44 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-230 | grounded_recovery_low | Only 0.2857 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-231 | extraction_unstable | The two reader passes agreed on only 0.3043 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-232 | cross_paper_overlap | 14 claim(s) in 2603.26846 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-233 | detail_not_in_cited_evidence | Claim 2603.26846/c2 names "two", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-234 | detail_not_in_cited_evidence | Claim 2603.26846/c8 names "Instruct", "Llama-3", "3.3", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-235 | detail_not_in_cited_evidence | Claim 2603.26846/c9 names "Gaussian", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-236 | detail_not_in_cited_evidence | Claim 2603.26846/c12 names "Lagrange", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-237 | prior_art_uncertain | Is claim "Among the four stability metrics, semantic entropy (SE) maintains the most consistent separability across CoT and Response, whereas PPL, Pmax, and Cosine Sim ar" already answered by prior work? Retrieval returned 15 candidates and the adjudicator could not decide. | |
| esc-238 | prior_art_uncertain | Is claim "CoT Monitor induces obfuscated reward hacking, paradoxically worsening Actual Deception while collapsing CoT Faithfulness." already answered by prior work? Retrieval returned 15 candidates and the adjudicator could not decide. | |
| esc-239 | prior_art_uncertain | Is claim "SAR retains general model capability, performing within normal fluctuation ranges and avoiding alignment tax or capability collapse." already answered by prior work? Retrieval returned 11 candidates and the adjudicator could not decide. | |
| esc-240 | critic_high_severity | 2 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-241 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-242 | checker_citation_not_found | 2 of 21 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-243 | grounded_recovery_low | Only 0.5385 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-244 | extraction_unstable | The two reader passes agreed on only 0.5 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-245 | cross_paper_overlap | 13 claim(s) in 2505.14289 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-246 | prior_art_uncertain | Is claim "Effective adversarial semantics form a dense, continuous 'semantic attack space' in the model's latent representation rather than isolated sparse points, which " already answered by prior work? Retrieval returned 15 candidates and the adjudicator could not decide. | |
| esc-247 | prior_art_uncertain | Is claim "An 'alignment paradox' exists: models with more extensive alignment training sometimes show increased vulnerability to EVA's attacks, because alignment training" already answered by prior work? Retrieval returned 13 candidates and the adjudicator could not decide. | |
| esc-248 | prior_art_uncertain | Is claim "Successful adversarial payloads are not uniformly distributed over persuasion dimensions but concentrate on two attractors—trust-aligned and urgency-aligned sem" already answered by prior work? Retrieval returned 13 candidates and the adjudicator could not decide. | |
| esc-249 | passages_drifted | 2 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-250 | checker_citation_not_found | 1 of 24 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-251 | grounded_recovery_low | Only 0.7273 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-252 | extraction_unstable | The two reader passes agreed on only 0.6923 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-253 | cross_paper_overlap | 10 claim(s) in 2606.15396 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-254 | detail_not_in_cited_evidence | Claim 2606.15396/c4 names "three", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-255 | prior_art_uncertain | Is claim "MDPO dynamically adjusts the KL penalty coefficient based on the policy model's real-time responsiveness to sample difficulty, using normalized reward gaps, out" already answered by prior work? Retrieval returned 15 candidates and the adjudicator could not decide. | |
| esc-256 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-257 | passages_drifted | 2 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-258 | grounded_recovery_low | Only 0.375 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-259 | extraction_unstable | The two reader passes agreed on only 0.2667 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-260 | cross_paper_overlap | 15 claim(s) in 2606.00033 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-261 | grounded_recovery_low | Only 0.5882 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-262 | extraction_unstable | The two reader passes agreed on only 0.4167 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-263 | cross_paper_overlap | 14 claim(s) in 1805.03090 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-264 | detail_not_in_cited_evidence | Claim 1805.03090/c5 names "POMDPs", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-265 | grounded_recovery_low | Only 0.2667 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-266 | extraction_unstable | The two reader passes agreed on only 0.2353 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-267 | cross_paper_overlap | 19 claim(s) in 2604.26360 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-268 | detail_not_in_cited_evidence | Claim 2604.26360/c13 names "Hopper-v4", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-269 | prior_art_uncertain | Is claim "The reciprocal reliability filter is derived from risk-sensitive mean-variance utility and replaces an unbounded linear penalty that can become negative under h" already answered by prior work? Retrieval returned 15 candidates and the adjudicator could not decide. | |
| esc-270 | prior_art_uncertain | Is claim "The reciprocal reliability filter satisfies positivity, monotonicity, boundedness, identity at zero uncertainty, and Lipschitz continuity." already answered by prior work? Retrieval returned 15 candidates and the adjudicator could not decide. | |
| esc-271 | prior_art_uncertain | Is claim "The magnitude of reward misspecification is assumed to be bounded by a monotonically increasing function of epistemic and aleatoric uncertainty." already answered by prior work? Retrieval returned 13 candidates and the adjudicator could not decide. | |
| esc-272 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-273 | checker_citation_not_found | 1 of 37 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-274 | grounded_recovery_low | Only 0.4211 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-275 | extraction_unstable | The two reader passes agreed on only 0.4 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-276 | cross_paper_overlap | 14 claim(s) in 2606.07706 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-277 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-278 | grounded_recovery_low | Only 0.7857 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-279 | extraction_unstable | The two reader passes agreed on only 0.7333 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-280 | cross_paper_overlap | 22 claim(s) in 2606.07612 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-281 | detail_not_in_cited_evidence | Claim 2606.07612/c16 names "Fine-tuning", "Instruct", "Llama-3", "3.1", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-282 | detail_not_in_cited_evidence | Claim 2606.07612/c19 names "3.15", "5.18", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-283 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-284 | checker_citation_not_found | 3 of 56 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-285 | grounded_recovery_low | Only 0.2381 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-286 | extraction_unstable | The two reader passes agreed on only 0.2273 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-287 | cross_paper_overlap | 17 claim(s) in 2502.20914 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-288 | detail_not_in_cited_evidence | Claim 2502.20914/c13 names "Gaussian", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-289 | prior_art_uncertain | Is claim "An exhaustive circuit-first pass on the illustrative example network (k = 3, n = 1, loss cutoff 10^-3) yielded 59 circuits and 114,230 interpretations." already answered by prior work? Retrieval returned 17 candidates and the adjudicator could not decide. | |
| esc-290 | grounded_recovery_low | Only 0.4737 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-291 | extraction_unstable | The two reader passes agreed on only 0.4737 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-292 | cross_paper_overlap | 12 claim(s) in 2604.24668 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-293 | detail_not_in_cited_evidence | Claim 2604.24668/c5 names "50", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-294 | detail_not_in_cited_evidence | Claim 2604.24668/c9 names "two", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-295 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-296 | grounded_recovery_low | Only 0.4545 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-297 | extraction_unstable | The two reader passes agreed on only 0.375 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-298 | cross_paper_overlap | 17 claim(s) in 2606.18988 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-299 | detail_not_in_cited_evidence | Claim 2606.18988/c4 names "VAC-GRPO", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-300 | detail_not_in_cited_evidence | Claim 2606.18988/c15 names "four", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-301 | prior_art_uncertain | Is claim "The authors construct Deception-10K, described as the first fine-grained audio-visual Chain-of-Thought dataset, comprising 10,000 video-reasoning pairs (~50 hou" already answered by prior work? Retrieval returned 18 candidates and the adjudicator could not decide. | |
| esc-302 | prior_art_uncertain | Is claim "The paper proposes Visual-Audio Consistency Group Relative Policy Optimization (VAC-GRPO) with a progressive training strategy that stratifies data into four di" already answered by prior work? Retrieval returned 17 candidates and the adjudicator could not decide. | |
| esc-303 | prior_art_uncertain | Is claim "Most existing baseline multimodal large language models perform around the random-guess baseline of 50% on deception detection despite identical prompts." already answered by prior work? Retrieval returned 13 candidates and the adjudicator could not decide. | |
| esc-304 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-305 | passage_fabricated | 1 passage(s) the reader cited do not appear in the source paper at all. Is the extractor inventing text, or is the PDF extraction mangling honest quotes? | |
| esc-306 | passages_drifted | 3 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-307 | grounded_recovery_low | Only 0.5385 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-308 | extraction_unstable | The two reader passes agreed on only 0.4706 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-309 | cross_paper_overlap | 14 claim(s) in 2606.08451 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-310 | detail_not_in_cited_evidence | Claim 2606.08451/c8 names "six", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-311 | detail_not_in_cited_evidence | Claim 2606.08451/c13 names "38", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-312 | prior_art_uncertain | Is claim "In the most severe cases, models agree with harmful prompts over 70% of the time in zero-shot languages, defaulting to explicit agreement with safety-critical p" already answered by prior work? Retrieval returned 20 candidates and the adjudicator could not decide. | |
| esc-313 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-314 | passages_drifted | 2 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-315 | grounded_recovery_low | Only 0.5833 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-316 | extraction_unstable | The two reader passes agreed on only 0.5 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-317 | cross_paper_overlap | 16 claim(s) in 2606.10106 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-318 | detail_not_in_cited_evidence | Claim 2606.10106/c8 names "four", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-319 | detail_not_in_cited_evidence | Claim 2606.10106/c9 names "four", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-320 | checker_citation_not_found | 4 of 34 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-321 | grounded_recovery_low | Only 0.625 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-322 | extraction_unstable | The two reader passes agreed on only 0.5 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-323 | cross_paper_overlap | 19 claim(s) in 2606.07532 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-324 | detail_not_in_cited_evidence | Claim 2606.07532/c2 names "Justice", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-325 | detail_not_in_cited_evidence | Claim 2606.07532/c6 names "Experiment", "200", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-326 | detail_not_in_cited_evidence | Claim 2606.07532/c10 names "ChatEval", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-327 | prior_art_uncertain | Is claim "A single-model ablation isolating identity stripping (Experiment 3 / pilot14) produces a directional accuracy gain of 6.0 percentage points over unstripped cont" already answered by prior work? Retrieval returned 13 candidates and the adjudicator could not decide. | |
| esc-328 | checker_citation_not_found | 2 of 59 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-329 | grounded_recovery_low | Only 0.6 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-330 | extraction_unstable | The two reader passes agreed on only 0.48 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-331 | cross_paper_overlap | 18 claim(s) in 2606.20814 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-332 | detail_not_in_cited_evidence | Claim 2606.20814/c2 names "Instruct", "Qwen2", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-333 | prior_art_uncertain | Is claim "The relationship between evaluation sample size and the Max Score Difference roughly follows a power law." already answered by prior work? Retrieval returned 14 candidates and the adjudicator could not decide. | |
| esc-334 | prior_art_uncertain | Is claim "The raw (non-JSON, non-template) format of the Initial EM questions is almost always the most misaligned format." already answered by prior work? Retrieval returned 14 candidates and the adjudicator could not decide. | |
| esc-335 | prior_art_uncertain | Is claim "Even with variations in learning schedules, in-domain loss has a dominant effect on the level of misalignment in the paper's experiment setting." already answered by prior work? Retrieval returned 19 candidates and the adjudicator could not decide. | |
| esc-336 | prior_art_uncertain | Is claim "Across the training-dynamics experiments summarized in Table 3, training loss still guides the level of misalignment." already answered by prior work? Retrieval returned 22 candidates and the adjudicator could not decide. | |
| esc-337 | passages_drifted | 2 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-338 | grounded_recovery_low | Only 0.5625 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-339 | extraction_unstable | The two reader passes agreed on only 0.5556 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-340 | cross_paper_overlap | 14 claim(s) in 2606.26793 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-341 | detail_not_in_cited_evidence | Claim 2606.26793/c2 names "Prior Sampling", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-342 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-343 | checker_citation_not_found | 1 of 20 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-344 | grounded_recovery_low | Only 0.625 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-345 | extraction_unstable | The two reader passes agreed on only 0.625 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-346 | cross_paper_overlap | 8 claim(s) in 2602.18008 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-347 | detail_not_in_cited_evidence | Claim 2602.18008/c2 names "Existing LLM-based", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-348 | prior_art_uncertain | Is claim "Models generated by NIMMGen can be used for counterfactual intervention simulation: increasing simulated social distancing strength produces systematic reductio" already answered by prior work? Retrieval returned 21 candidates and the adjudicator could not decide. | |
| esc-349 | checker_citation_not_found | 3 of 22 quote(s) produced by the checker could not be located in the material it was shown. Its verdicts rest on text it may not have read correctly. | |
| esc-350 | grounded_recovery_low | Only 0.75 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-351 | extraction_unstable | The two reader passes agreed on only 0.5 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-352 | cross_paper_overlap | 21 claim(s) in 2606.27188 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-353 | detail_not_in_cited_evidence | Claim 2606.27188/c1 names "Agentic Business Process", "Business Process Management", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-354 | detail_not_in_cited_evidence | Claim 2606.27188/c5 names "Model Context Protocol", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-355 | detail_not_in_cited_evidence | Claim 2606.27188/c16 names "two", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-356 | detail_not_in_cited_evidence | Claim 2606.27188/c18 names "CUGA FLO", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-357 | grounded_recovery_low | Only 0.3333 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-358 | extraction_unstable | The two reader passes agreed on only 0.3333 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-359 | cross_paper_overlap | 27 claim(s) in 2605.11047 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-360 | prior_art_uncertain | Is claim "Qwen3.5-Plus, DeepSeek-v4-Flash, and DeepSeek-v4-Pro show consistently high AGS across the six risk categories, indicating the generated traps transfer beyond t" already answered by prior work? Retrieval returned 11 candidates and the adjudicator could not decide. | |
| esc-361 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-362 | grounded_recovery_low | Only 0.4615 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-363 | extraction_unstable | The two reader passes agreed on only 0.4444 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-364 | cross_paper_overlap | 16 claim(s) in 2510.00845 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-365 | prior_art_uncertain | Is claim "Bootstrap resampling of the input dataset yields the lowest structural consistency and highest variability of discovered circuits (Jaccard µ = 0.561, CV = 0.335" already answered by prior work? Retrieval returned 16 candidates and the adjudicator could not decide. | |
| esc-366 | prior_art_uncertain | Is claim "Circuits discovered under bootstrap resampling also have the highest average circuit error (0.440), meaning they are structurally different and less faithful to" already answered by prior work? Retrieval returned 12 candidates and the adjudicator could not decide. | |
| esc-367 | prior_art_uncertain | Is claim "Shifting the meta-distribution (meta-dataset or prompt paraphrasing) yields more stable circuits than bootstrap resampling, with higher Jaccard indices (0.790 a" already answered by prior work? Retrieval returned 15 candidates and the adjudicator could not decide. | |
| esc-368 | prior_art_uncertain | Is claim "Circuit discovery methods do not scale trivially: stability degrades for larger models, with gpt2-small yielding relatively clustered results while Llama-3.2 (1" already answered by prior work? Retrieval returned 9 candidates and the adjudicator could not decide. | |
| esc-369 | passages_drifted | 1 passage(s) are neither verbatim nor invented: they open with real source text and then diverge into paraphrase. Should the extractor be required to quote contiguously, or should near-quotes be accepted with a marker? | |
| esc-370 | grounded_recovery_low | Only 0.5333 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-371 | extraction_unstable | The two reader passes agreed on only 0.5 of claims. Should extractions below this threshold be treated as unusable? | |
| esc-372 | cross_paper_overlap | 26 claim(s) in 2606.21399 closely restate a claim the society already holds from another paper. Is this an independent confirmation, a duplicate that should be merged, or a contradiction that should be recorded? | |
| esc-373 | detail_not_in_cited_evidence | Claim 2606.21399/c2 names "two", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-374 | detail_not_in_cited_evidence | Claim 2606.21399/c6 names "two", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-375 | detail_not_in_cited_evidence | Claim 2606.21399/c10 names "0.423", "0.394", "0.436", "0.417", "four", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-376 | detail_not_in_cited_evidence | Claim 2606.21399/c15 names "ScienceWorld", "23", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-377 | detail_not_in_cited_evidence | Claim 2606.21399/c17 names "Prompt-only", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-378 | detail_not_in_cited_evidence | Claim 2606.21399/c20 names "On ALFWorld", which is real paper vocabulary but appears in none of the passages cited for this claim, nor in the window shown to the checker. Is this misattribution or partial citation? | |
| esc-379 | critic_high_severity | 1 high-severity objection(s) stand against this extraction. Do any invalidate a claim that should not remain in the graph? | |
| esc-380 | grounded_recovery_low | Only 0.2692 of source-grounded claims were recovered by the second reader pass. Is this coverage diversity, or extractor unreliability? The distinction determines whether the 0.50 raw overlap is bad news. | |
| esc-381 | extraction_unstable | The two reader passes agreed on only 0.2692 of claims. Should extractions below this threshold be treated as unusable? |
Machine-readable: /society/escalations.json.