{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.20492", "kind": "paper", "url": "https://arxiv.org/abs/2607.20492"}, {"id": "gh:ranausmanai/phantomfill", "kind": "repo", "url": "https://github.com/ranausmanai/phantomfill"}], "authors": ["Rana Muhammad Usman"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.20492", "source_url": "https://arxiv.org/abs/2607.20492"}, "categories": ["General AI", "Language & Knowledge", "cs.LG"], "description": {"en": "PhantomFill measures schema-coerced fabrication in language models by asking questions on unanswerable inputs under three output formats: free text, JSON with an escape option, and JSON with required fields. It reports Coerced Fabrication Rate and Escape Utilization Rate."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.20492", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.20492", "title": "PhantomFill"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.20492", "modality": null, "name": "PhantomFill", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_phantomfill_4f8c991e", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "60a8cd6ee3e4a1aa5e13bc8e024a4660c3bb90bc97fa61e1beb0040b3f1e6dce", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2607.20492"}, "publisher": null, "released": "2026-06-11", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.20492", "source_url": "https://arxiv.org/abs/2607.20492"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-20492", "source": "claire_radar", "source_benchmark_id": "2607.20492", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/ranausmanai/phantomfill", "githubScope": "benchmark_repo", "githubStars": 0, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 1, "hfPaperUrl": "https://huggingface.co/papers/2607.20492", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-01", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "dc0cfa8ee8155946c8493880bd83e29bc0d25bb525a41a8bf1625cd50dde239c", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "PhantomFill measures schema-coerced fabrication in language models by asking questions on unanswerable inputs under three output formats: free text, JSON with an escape option, and JSON with required fields. It reports Coerced Fabrication Rate and Escape Utilization Rate.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "public artifact URL"], "snippet": "We release PhantomFill, a benchmark with deterministic scoring and two reportable numbers: the Coerced Fabrication Rate and the Escape Utilization Rate."}, "familyId": "bmf_c5647cf4fd51", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/ranausmanai/phantomfill", "url": "https://github.com/ranausmanai/phantomfill"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/ranausmanai/phantomfill", "url": "https://huggingface.co/papers/2607.20492"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/ranausmanai/phantomfill", "url": "https://arxiv.org/abs/2607.20492"}], "scope": "benchmark_repo", "url": "https://github.com/ranausmanai/phantomfill", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_phantomfill_4f8c991e", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/ranausmanai/phantomfill", "data": null, "hfPaper": "https://huggingface.co/papers/2607.20492", "paper": "https://arxiv.org/abs/2607.20492", "pdf": "https://arxiv.org/pdf/2607.20492", "project": null, "report": "https://arxiv.org/abs/2607.20492"}, "metrics": [], "motivation": "Language models in production do not write prose.", "name": "PhantomFill", "oneLine": "PhantomFill measures schema-coerced fabrication in language models by asking questions on unanswerable inputs under three output formats: free text, JSON with an escape option, and JSON with required fields. It reports Coerced Fabrication Rate and Escape Utilization Rate.", "paperTitle": "PhantomFill: When the Form Demands an Answer, Language Models Invent One", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.20492", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-11", "source": {"authors": ["Rana Muhammad Usman"], "categories": ["cs.LG", "cs.AI", "cs.CL"], "id": "2607.20492", "title": "PhantomFill: When the Form Demands an Answer, Language Models Invent One", "type": "arxiv", "url": "https://arxiv.org/abs/2607.20492"}, "sourceUpdatedAt": "2026-07-27T11:27:09Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "Hallucination in form-filling contexts is under-measured and costly. PhantomFill provides deterministic, code-based metrics targeting a critical failure mode, enabling model comparison and safety evaluation in structured output settings."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
