{ "schema_version": 1, "builder_plan": { "scenario": "An agent looks good in demos, but you cannot tell which product tasks it fails.", "change": "Use the study's evaluation idea to inspect a small set of tasks with human-checked answers.", "baseline": "Your current evaluator and a manually reviewed reference set.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Agreement with human review, missed failures, false alarms, and evaluation cost per task.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2610.13790", "title": "OccludeBench: Can VLMs Find What's Missing?", "summary": "The technique replaces single-pass question answering about occluded objects with a two-step procedure in which the model first generates a candidate hidden object via inpainting and then evaluates that hypothesis against the visible physical cue. It costs an extra image-generation call and a second verification pass per query, and it is only available on models with image-generation capabilities; extended reasoning, tool use, and few-shot prompting were shown not to work reliably as alternatives. It can fail on the 50% of cases where models cannot recover the object from the projection in the full image, and the improvement is reported on only two models on a curated benchmark, so transfer to arbitrary production imagery is unverified.", "rationale": "The intervention is cheap to trial through third-party APIs but its benefit is demonstrated only on generation-capable models in a controlled benchmark, so workload-specific validation is required before production adoption.", "technique": "Inpaint-then-confront prompting for occluded-object reasoning", "family": "avaliacao_de_agente", "recommendation": "testar", "published": "2026-10-09", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.13790", "independent_implementations": 0, "total_implementations": 1, "stars": 122, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "The technique replaces single-pass question answering about occluded objects with a two-step procedure in which the model first generates a candidate hidden object via inpainting and then evaluates that hypothesis against the visible physical cue. It costs an extra image-generation call and a second verification pass per query, and it is only available on models with image-generation capabilities; extended reasoning, tool use, and few-shot prompting were shown not to work reliably as alternatives. It can fail on the 50% of cases where models cannot recover the object from the projection in the full image, and the improvement is reported on only two models on a curated benchmark, so transfer to arbitrary production imagery is unverified.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.13790", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "VLMs average 42.7% on direct physical-cue cases versus 62.7% on inferential cases (20-point gap); the inpaint-then-confront intervention reduces this gap to a statistically non-significant difference on the two models with image-generation capabilities.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.13790", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }