{ "schema": "rook-ai-evaluation-run-protocol-v1", "version": "phase3f-20261005-v2", "status": "frozen_before_predictions", "catalog_sha256": "55ba3830b1ce292e88a283947790b056450629ab31fed110b9379574350f6e68", "dataset_sha256": "86a71f042c848ebfb19791053a797c6b208c733eaff1104784d01ea72779b1ae", "partitions": { "validation": { "path": "data/rook-reviewed-v2/validation.jsonl", "sha256": "6b646a8fbc1dff99ff8a13668d351025da610850212df55e5f47e1f4bdde8409", "rows": 240, "role": "validation" }, "calibration": { "path": "data/rook-reviewed-v2/calibration.jsonl", "sha256": "c4c62723355c51c819ef5d885750d04e8eeba45d68a52f7e31536fa011591bcd", "rows": 260, "role": "calibration" }, "sealed_test": { "path": "data/rook-reviewed-v2/sealed_test.jsonl", "sha256": "ad721e21393aec7db6951266ff7f9f2c444d51cf7bf178a52122e095f8391462", "rows": 528, "role": "sealed_test" } }, "baseline_id": "decision-2.0-lux-9b@78bf3c03d9147aeb30b641edfe0e30ed04887ca5", "candidate_ids": [ "rook-epoch-207", "rook-epoch-414", "rook-epoch-621" ], "models": { "rook-epoch-207": { "artifact_sha256": "a308719d8ff453457fedefbf49c5501265c80676954ec6a975b5ce4754240b15", "manifest_path": "runs/phase3d-20261005/recovered-execution-r1/checkpoints/d3-experiment-r1/checkpoint-000207/rook-checkpoint.json", "step": 207, "epoch": 1, "implementation_sha256": "bf68e22faf74136dad662cb8e3b7023a0024f05a82892e1a308cbb82f7fcb7ab" }, "rook-epoch-414": { "artifact_sha256": "bd7e0a9f8285490719fd2b0c74c193ba698f2baa034f2f4d21056ee52d6d11fd", "manifest_path": "runs/phase3d-20261005/recovered-execution-r1/checkpoints/d3-experiment-r1/checkpoint-000414/rook-checkpoint.json", "step": 414, "epoch": 2, "implementation_sha256": "bf68e22faf74136dad662cb8e3b7023a0024f05a82892e1a308cbb82f7fcb7ab" }, "rook-epoch-621": { "artifact_sha256": "452a4c3cf6ac45c582ca7281879455a9517ce825fd1973ab42701c4a9da31ef3", "manifest_path": "runs/phase3d-20261005/recovered-execution-r1/checkpoints/d3-experiment-r1/checkpoint-000621/rook-checkpoint.json", "step": 621, "epoch": 3, "implementation_sha256": "bf68e22faf74136dad662cb8e3b7023a0024f05a82892e1a308cbb82f7fcb7ab" }, "decision-2.0-lux-9b@78bf3c03d9147aeb30b641edfe0e30ed04887ca5": { "artifact_sha256": "4e0f4ca0cf3c833933f9bed3021012d061910016c9f4ca43b34f5d66e8d0a260", "implementation_sha256": "bf68e22faf74136dad662cb8e3b7023a0024f05a82892e1a308cbb82f7fcb7ab", "role": "mandatory unchanged pinned Lux; no adapter; original pretrained head" } }, "formatting_version": "decision2-segmented-options-global-query-v1", "formatter_sha256": "9f4eafed534be9c0c18ca6c45ca4d5cbeafe70c1129e1ea9224c4a3a21042a24", "tokenizer_sha256": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523", "class_namespaces": { "access-channel/choice": { "renew": "Renew credentials", "security": "Security review", "access": "Normal access", "offline": "Offline verification" }, "reservation-allocation/choice": { "waitlist": "No suitable room", "B": "Room B", "A": "Room A", "C": "Room C" }, "deadline-date/noul": { "false": "The stated rule is not satisfied", "true": "The stated rule is satisfied" }, "ledger-consistency/noul": { "false": "The stated rule is not satisfied", "true": "The stated rule is satisfied" }, "cold-chain/score": { "0": "No excursion", "1": "Short excursion", "2": "Sustained excursion", "3": "Long excursion" }, "response-sla/score": { "0": "Within SLA", "1": "Up to 30 minutes late", "2": "31 to 120 minutes late", "3": "More than 120 minutes late" }, "appointment-resolution/choice": { "attend": "Attend current booking", "rebook": "Rebook after lateness", "reschedule": "Reschedule for staff absence", "cancel": "Cancel booking" }, "repair-resolution/choice": { "isolate": "Isolate equipment", "diagnose": "Obtain diagnosis", "repair": "Repair equipment", "replace": "Replace equipment" }, "factual-evidence/noul": { "false": "The stated rule is not satisfied", "true": "The stated rule is satisfied" }, "workflow-release/noul": { "false": "The stated rule is not satisfied", "true": "The stated rule is satisfied" }, "capacity-pressure/score": { "0": "Spare capacity", "1": "Exactly full", "2": "Over capacity", "3": "Demand with zero capacity" }, "dependency-readiness/score": { "0": "Blocked", "1": "Checks incomplete", "2": "Checks complete, approval missing", "3": "Ready with approval" }, "dispatch-window/choice": { "today": "Ship today", "compliance": "Compliance review", "tomorrow": "Next-day dispatch", "address": "Request address", "backorder": "Backorder" }, "public:clinc150/choice": { "calendar": "Intent category: calendar", "smart_home": "Intent category: smart home", "reminder_update": "Intent category: reminder update", "next_song": "Intent category: next song", "what_song": "Intent category: what song", "todo_list_update": "Intent category: todo list update", "restaurant_suggestion": "Intent category: restaurant suggestion", "ingredient_substitution": "Intent category: ingredient substitution", "ingredients_list": "Intent category: ingredients list", "restaurant_reviews": "Intent category: restaurant reviews", "meal_suggestion": "Intent category: meal suggestion", "restaurant_reservation": "Intent category: restaurant reservation", "card_declined": "Intent category: card declined", "report_lost_card": "Intent category: report lost card", "credit_limit": "Intent category: credit limit", "rewards_balance": "Intent category: rewards balance", "credit_score": "Intent category: credit score", "redeem_rewards": "Intent category: redeem rewards", "pin_change": "Intent category: pin change", "bill_balance": "Intent category: bill balance", "transactions": "Intent category: transactions", "routing": "Intent category: routing", "bill_due": "Intent category: bill due", "pay_bill": "Intent category: pay bill", "spending_history": "Intent category: spending history", "transfer": "Intent category: transfer", "order_checks": "Intent category: order checks", "min_payment": "Intent category: min payment", "pto_request_status": "Intent category: pto request status", "pto_balance": "Intent category: pto balance", "pto_used": "Intent category: pto used", "schedule_meeting": "Intent category: schedule meeting", "next_holiday": "Intent category: next holiday", "pto_request": "Intent category: pto request", "apr": "Intent category: apr", "credit_limit_change": "Intent category: credit limit change", "improve_credit_score": "Intent category: improve credit score", "damaged_card": "Intent category: damaged card", "replacement_card_duration": "Intent category: replacement card duration", "new_card": "Intent category: new card", "vaccines": "Intent category: vaccines", "travel_alert": "Intent category: travel alert", "book_hotel": "Intent category: book hotel", "lost_luggage": "Intent category: lost luggage", "international_visa": "Intent category: international visa", "book_flight": "Intent category: book flight", "sync_device": "Intent category: sync device", "repeat": "Intent category: repeat", "no": "Intent category: no", "whisper_mode": "Intent category: whisper mode", "change_volume": "Intent category: change volume", "change_speed": "Intent category: change speed", "schedule_maintenance": "Intent category: schedule maintenance", "jump_start": "Intent category: jump start", "mpg": "Intent category: mpg", "gas_type": "Intent category: gas type", "gas": "Intent category: gas", "distance": "Intent category: distance", "traffic": "Intent category: traffic", "oil_change_when": "Intent category: oil change when", "current_location": "Intent category: current location", "oil_change_how": "Intent category: oil change how", "tire_change": "Intent category: tire change", "are_you_a_bot": "Intent category: are you a bot", "thank_you": "Intent category: thank you", "tell_joke": "Intent category: tell joke", "greeting": "Intent category: greeting", "who_made_you": "Intent category: who made you", "meaning_of_life": "Intent category: meaning of life", "insurance": "Intent category: insurance", "w2": "Intent category: w2", "direct_deposit": "Intent category: direct deposit", "insurance_change": "Intent category: insurance change", "reminder": "Intent category: reminder", "order_status": "Intent category: order status", "update_playlist": "Intent category: update playlist", "change_accent": "Intent category: change accent", "change_user_name": "Intent category: change user name", "change_language": "Intent category: change language", "change_ai_name": "Intent category: change ai name", "taxes": "Intent category: taxes", "income": "Intent category: income", "recipe": "Intent category: recipe", "cancel_reservation": "Intent category: cancel reservation", "nutrition_info": "Intent category: nutrition info", "calendar_update": "Intent category: calendar update", "measurement_conversion": "Intent category: measurement conversion", "timer": "Intent category: timer", "share_location": "Intent category: share location", "text": "Intent category: text", "flip_coin": "Intent category: flip coin", "calculator": "Intent category: calculator", "time": "Intent category: time", "date": "Intent category: date", "find_phone": "Intent category: find phone", "spelling": "Intent category: spelling", "flight_status": "Intent category: flight status", "carry_on": "Intent category: carry on", "plug_type": "Intent category: plug type", "translate": "Intent category: translate", "payday": "Intent category: payday", "directions": "Intent category: directions", "uber": "Intent category: uber", "last_maintenance": "Intent category: last maintenance", "cancel": "Intent category: cancel", "roll_dice": "Intent category: roll dice", "weather": "Intent category: weather", "food_last": "Intent category: food last", "confirm_reservation": "Intent category: confirm reservation", "how_busy": "Intent category: how busy", "cook_time": "Intent category: cook time", "shopping_list_update": "Intent category: shopping list update", "calories": "Intent category: calories", "make_call": "Intent category: make call", "alarm": "Intent category: alarm", "report_fraud": "Intent category: report fraud", "expiration_date": "Intent category: expiration date", "freeze_account": "Intent category: freeze account", "interest_rate": "Intent category: interest rate", "how_old_are_you": "Intent category: how old are you", "what_can_i_ask_you": "Intent category: what can i ask you", "who_do_you_work_for": "Intent category: who do you work for", "do_you_have_pets": "Intent category: do you have pets", "meeting_schedule": "Intent category: meeting schedule", "travel_notification": "Intent category: travel notification", "exchange_rate": "Intent category: exchange rate", "accept_reservations": "Intent category: accept reservations", "play_music": "Intent category: play music", "order": "Intent category: order", "what_are_your_hobbies": "Intent category: what are your hobbies", "goodbye": "Intent category: goodbye", "fun_fact": "Intent category: fun fact", "account_blocked": "Intent category: account blocked", "international_fees": "Intent category: international fees", "balance": "Intent category: balance", "definition": "Intent category: definition", "timezone": "Intent category: timezone", "tire_pressure": "Intent category: tire pressure", "yes": "Intent category: yes", "car_rental": "Intent category: car rental", "where_are_you_from": "Intent category: where are you from", "reset_settings": "Intent category: reset settings", "rollover_401k": "Intent category: rollover 401k", "todo_list": "Intent category: todo list", "shopping_list": "Intent category: shopping list", "user_name": "Intent category: user name", "application_status": "Intent category: application status", "maybe": "Intent category: maybe" }, "return-method/choice": { "accessible": "Accessible pickup", "courier": "Courier pickup", "dropoff": "Local dropoff", "hazardous": "Specialist hazardous pickup" }, "backup-recoverability/noul": { "false": "The stated rule is not satisfied", "true": "The stated rule is satisfied" }, "consent-validity/noul": { "false": "The stated rule is not satisfied", "true": "The stated rule is satisfied" }, "deadline-pressure/score": { "0": "Complete or ample time", "1": "Within a day", "2": "Blocked and within four hours", "3": "Overdue" }, "inventory-exposure/score": { "0": "No current exposure", "1": "At reorder threshold", "2": "Below reorder threshold", "3": "Backorders with zero stock" } }, "procedure": { "selection": { "partition": "validation", "probabilities": "raw uncalibrated", "score": "mean over three tasks (weight 1/3 each), then equal weight per validation family within task; choice/noul utility=accuracy, score utility=1-MAE/(K-1)", "tie_break": "exact score ties: lower family-macro NLL, then earlier step (207 < 414 < 621)", "non_regression": "relative to unchanged Lux: each task full-coverage accuracy must not decrease; no family accuracy drop >0.02 absolute; overall row-mean NLL must not increase; family-macro ordinal MAE must not increase; undefined required metrics fail eligibility", "no_eligible_candidate": "retain baseline, report failure; do not relax gates" }, "calibration": { "partition": "calibration", "temperature": "one finite positive scalar per selected model and baseline, minimize row-mean multiclass NLL on original logits; no smoothing", "threshold": "single maximum-probability threshold per model from grid [0,.5,.6,.7,.8,.9,.95,.99,1]; maximize coverage with empirical error risk <=0.05; tie lower threshold; if none has nonzero coverage, abstain from all; freeze before sealed run" }, "sealed": "after complete adjudicated gold, validation selection, calibration and procedure execution freeze; one paired run only; no tuning from sealed outcomes", "classification": "family/task namespaces; semantic supplied keys validated against descriptions; per-class precision/recall/F1/support; balanced accuracy and macro-F1 over gold-supported classes; absent classes explicitly listed, recall undefined; precision undefined if no predictions; supported F1 zero when no true positives; no global option-position F1", "probabilities": "NLL row mean; zero gold probability => +infinity, no clipping; Brier=sum_k(p_k-y_k)^2 row mean; fixed ECE ten bins [j/10,(j+1)/10) final inclusive 1; adaptive ECE ten deterministic equal-rank bins floor(10*i/N), sort by (confidence,row_id); report reliability/counts", "ordinal": "family/rubric-specific MAE of argmax level, QWK=1-observed_squared_error/expected_squared_error from marginal counts (undefined if expected=0), RPS=sum cumulative squared errors/(K-1); namespace means explicitly distinguished from row means", "breakdowns": [ "task_type", "family", "source_id", "language", "cardinality" ], "selective": "full coverage first; threshold accepts max probability >= threshold; report coverage/count/error risk; empty accepted set => null risk; no selective benefit from omitting prediction failures", "bootstrap": { "method": "paired family-cluster resampling within task, preserving 1/3 task weighting; resample all rows in each drawn whole family with multiplicity", "replicates": 2000, "seed": 20261004, "confidence": 0.95, "interval": "percentile .025/.975 with linear interpolation; candidate minus baseline; if any undefined/nonfinite replicate, interval null with count; do not clip infinity", "scope": "accuracy, declared utility score, family-macro balanced accuracy/F1, row NLL/Brier/fixed/adaptive ECE, family-macro ordinal MAE/QWK/RPS", "limitations": "6 validation, 6 calibration, 7 sealed independent families; public CLINC is one family; narrow intervals do not establish broad generalization" }, "missing_metrics": "null with supports for undefined or absent task/rubric; empty partition invalid; missing predictions invalidate comparison; positive infinity serialized explicitly", "prediction_ties": "first supplied candidate in original ordered mapping", "upstream_exposure": "unknown; public intent adaptation is not original benchmark evaluation; English only; no language or superiority claims from training loss" }, "procedure_sha256": "049bd1ce127652f3cd01bef70040366398c542ee463cf5ad565423e96caf6836", "inherited_protocol_file_sha256": "299162ed0ad7ea165226b6d1e67e914ba039dba1dbe723ff7585ddaeef261a23", "ai_overlay_sha256": "e120bf6694cdecda85409561bcc400ea9077eb3864709495ac7cdf43f0823875", "human_reviewed": false, "human_attested": false, "label_provenance": "InferHub cx/gpt-6.1-sol blind pass then same-model reference-visible reconsideration", "external_exposure": { "validation": 240, "calibration": 260, "sealed_test": 528 }, "eligible_counts": { "validation": 240, "calibration": 260, "sealed_test": 525 }, "excluded_sealed_rows": [ { "row_id": "clinc150:test:2272", "target": null, "note": "The request is to change the voice’s gender, but none of the supplied categories covers voice gender or voice type. “change_accent” is not an adequate match." }, { "row_id": "clinc150:test:2997", "target": null, "note": "The utterance asks about a card-use charge in “x,” but does not establish that x is an international location. International fees is plausible, but the supplied information does not determine that category." }, { "row_id": "clinc150:test:3541", "target": null, "note": "“Are you okay” is a wellbeing check, and none of the supplied categories represents that intent. “Greeting” is not a sufficiently supported match." } ], "numerical_fit": { "temperature_bounds": [ 0.01, 100.0 ], "log_temperature_tolerance": 1e-09, "max_iterations": 256, "optimizer": "golden-section in log(T), include endpoints and T=1", "objective": "stable logsumexp row-mean NLL of original logits", "fit_split": "calibration", "thresholds": [ 0.0, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95, 0.99, 1.0 ], "maximum_empirical_risk": 0.05, "threshold_tie": "lower threshold", "empty_feasible_policy": "abstain_all" }, "code_files_sha256": { "tools/phase3f_evaluate.py": "001a12fee157aaf346162c9e22d9a18c70bcad74bc72804b8336c2f3535f90d0", "tools/phase3f_worker.py": "1d205fcffa092625c12f89e90eaf10ea3cbf64ee69dc8212be8410a68d6967cd", "tools/ai_review_inferhub.py": "7cb50dfc6ce4e140671d1f30313250fada061a17394a93db29630f35ed5aa238", "tools/ai_adjudicate_inferhub.py": "e6af7a42d78242c347ddd11296caccf79911cf41fe1a13ac50c708167d4eba1a", "src/rook/__init__.py": "8a5f2cc499698019b210460e04a2f00a61e662d6aa4e56d7a5197e08f9f416a7", "src/rook/api.py": "3e06325591d18b8df644f3c79b0453d06de8f57cbc1e9e8e25a9cca381165284", "src/rook/backend.py": "3b0402bc2942cb42638cf61d57316c25ac48582421d4bdda6848af2a582c4837", "src/rook/cli.py": "43ff2ac6dec0845704c5f979882156d8843d288a5c00dbe4b120180848233807", "src/rook/contracts.py": "77bf5c4f0bee64eef0871368186c5b2bb963f5ae2cc82dc317199ed4736bd30e", "src/rook/curation/__init__.py": "5a430f15d2386a51a96a6e550887367807a2abed08b6dea2e9df94c6f30f0eeb", "src/rook/curation/audit.py": "ae4b8c81f9fb3d8d69054473cb9ecd08bac5a2fc152dd783640f87d67b4a868e", "src/rook/curation/boundaries.py": "abf2400236f33b1fabe6e31daa1afb7237cdf56b607e67345815847d321a3478", "src/rook/curation/build.py": "9959cbd035faf8d6e125fa65146e79563c745e1a4e3043ecbd25797936b67ee6", "src/rook/curation/commands.py": "ecc050ac2f57ede72b2bcf1b03f78f50267fb82d11c4a7dd621f33b8c271dd37", "src/rook/curation/common.py": "be7639ff7eb786141f369c30b5c64c5da504e000d4626d7700e0c823595d7739", "src/rook/curation/families.py": "487c6b6b8bd0e664f0ec96491033eb8295c27a09327515677b558acb7df82550", "src/rook/curation/family_audit.py": "d64d9f22bbed26cc04701db4cfa49d30c8adb4e23ad5a122d77044eb23ac1d4d", "src/rook/curation/finalize.py": "c6fcd6ff6ee4def8b4f95decc57876bdfc7805b011d667a654898b8d5247fcf5", "src/rook/curation/importers.py": "f029c87cee65c05e189e5cda0f4e029395e8fac0b1ed346ac2e63c98ab3dde51", "src/rook/curation/independent.py": "7cd5a545be85f78dd9823dcc609f4588dd3f7f6577d97b7753ad4f489dfde4a2", "src/rook/curation/pilot.py": "b8c61b3aaecfb397c3c986d0a27ee13230cd38e66470a8d338da4888dfa87fcd", "src/rook/curation/review.py": "a72760e01987644c1a031f693923f12d6efa5903c91968d433bb328e450abd76", "src/rook/curation/versioning.py": "4120579e681759f1e2f9fac56959a295413616984025405e9ac0e82b41b48b3d", "src/rook/data.py": "c61a6e386059f8a50566f574b4f3bb43d90940f655ffb1f83760f66c8934fdc1", "src/rook/deepseek.py": "d8aa6b01bfefe557abb36092577a90ee82cefeb7f8fc92f93bbd6e6b7df9f186", "src/rook/evaluation.py": "5716bca25f92b05072c0b4691d5eabbfbd08e7749dcafe3e0515829a46c384d4", "src/rook/evaluation_commands.py": "177032bf78d0b7524b4cd1f7734bf28f006f63e6e419c2260503c8a73dc9a1a9", "src/rook/heldout_app.py": "41d5cbb0054250efad4606a94428e1410bc2923c796e34f28ec7783b7a4f8351", "src/rook/heldout_catalog.py": "edf65df94e2590312b5e3d78858db2fdfa184128199097354fc2c60ec3798f54", "src/rook/heldout_review.py": "d4962810ef7e523eed93e1f8b37449e511cb02f2565cb9ba78847ca5e0244b05", "src/rook/native/__init__.py": "a903ce28822e6c11a37156c65d0d6584c210bf23a1024b06165609e5541bda56", "src/rook/native/batch.py": "bd81dcb6b394e9cbab9d1ba33db8ad3d8107c73cf1aedce0747ce110eb7ee676", "src/rook/native/checkpoint.py": "607562e0e40b961a71e6fcaf4925dd3a2ae524c3cf89f95e8fe586532c6d30e1", "src/rook/native/commands.py": "e7b38759dc144c1284fc9f71cc3e66ec7165b42464c23f149c13f8831f008680", "src/rook/native/epoch.py": "537529228c656caea9f0de9e289d6cb65bf34e4185fa0dd8bdbd403b4c712b22", "src/rook/native/full_checkpoint.py": "4d0922df25d5debe49573c6e5508e1382cd601c6c71c29f9c0bd10bca04d26fa", "src/rook/native/full_commands.py": "31629345d10dbad33ef907ec4bd6f054d1de741ff633f25668c50f4e5e97ffbe", "src/rook/native/full_train.py": "21c0f011b04ea38dc9c16070b3ab1bf5154c5164e9875f1e140924e50668ce19", "src/rook/native/guards.py": "38d97e4da40fc4de8fcb78733006038760206ed7b44bcbd34a08f8ffd87675dc", "src/rook/native/probes.py": "4e5396e26b27fb5faf138edaaeea954b63f23008d8f7f6060254c8bf5c31912c", "src/rook/native/runtime.py": "881ea83fee816edc863909d4479b0138c718735e809551715a7ddcf4740b53e9", "src/rook/native/session_budget.py": "797bf31a7ebab48b94ad62fff2f65a1f2829f2117c02dbda09890a19bd47ade8", "src/rook/native/storage.py": "6eaccfe7556cedabc6169438189a2ff09bdc54741033c8cf13d83b691d9c5772", "src/rook/native/train.py": "b499a57e1cf2c95ec1d52b318cb2527a4aaf2693aa44afb1dc06fcf235138bc7", "src/rook/oracles.py": "1f2e61ef519f2bc9036dba45615d5ff64f0dcf3e4d739ddb902250767eb18877", "src/rook/render.py": "9f4eafed534be9c0c18ca6c45ca4d5cbeafe70c1129e1ea9224c4a3a21042a24", "src/rook/review_app.py": "2b559c29ca306259e9cb437810f7ed28b4b64f403931a02b2beb7f23b9e5b04a", "src/rook/review_catalog.py": "55efcd27fe408e0533273a256194f4f88b8c3e01ce9154b07ac1b12f93b39c9d", "src/rook/review_store.py": "2248ae62fe588a688031e080199b8eeeb31f5f164fc7bfc025a839ad5153f07a", "src/rook/teacher.py": "b95aa3a8789525eef70653352a9ac19b96c8e5e03cfd7917d343999d64a71a75", "src/rook/teacher_ledger.py": "3abaf618b1ef6687f09b42b62aa68cbbfa7f91249bce00585a7dc6d0c26158f8", "src/rook/teacher_tokens.py": "d4bfae50dc43eacc4139af934e27740d59a1292262373e74e0f91124f292dd8d" }, "worker_context_sha256": "6108c4d50c3a312411c8e02c27d383da7dc624a36d319e0246efe75ba3c745d3", "input_rows_sha256": { "validation": "46bccb89d2eb107a398006f2f2a01e8b804e6f6cd6762b8fa9fe4ffc06c95883", "calibration": "f81fdc079fd5347ec1c454753ad49472b5e413742928322887d4629d04f6af16", "sealed_test": "1790ca822d9fcf0177fd56d784669d71e1ac851108c423327d7fc75a2ac2e579" }, "execution_order": [ "validation", "selection", "calibration", "sealed_freeze", "sealed_test" ], "inference": { "batch_size": 1, "max_length": 16384, "pilot_rows_per_model": 8, "logits_required": true, "probability_logit_absolute_tolerance": 1e-06, "historical_snapshot_required": true }, "evidence_sha256": { "audit/ai-review-inferhub/20261005/execution-verification.json": "69ab3b32237ac581efc0c3300c397eaf9fa2e1958883493fcc6ba2a2d8822404", "audit/ai-adjudication-inferhub/20261005/execution-verification.json": "7e4023711971e0389311ff823bcddbd8136e795dce4cda7e5cc5591b94e90bc6" } }