DRIFT CONTROL (unmutated baseline, complete pass, exit 5): the corpus moved under the baseline at 75 id(s) -- {"accepted_text_moved": [], "diff:appeared": [1482, 1483, 1484, 1486, 1487, 1489, 1490, 1491, 1492, 1494, 1495, 1496, 1498, 1499, 1501, 1502, 1503, 1504, 1506, 1507, 1508, 1509, 1511, 1516, 1517, 1518, 1519, 1521, 1522, 1523, 1524, 1525, 1526, 1527, 1528, 1529, 1530, 1531, 1532, 1533, 1534, 1535, 1536, 1537, 1538, 1539, 1540], "diff:changed": [356, 360, 361, 370, 415, 424, 426, 428, 435, 437, 439, 442, 443, 444, 450, 459, 567, 629, 664, 693, 901, 916, 918, 919, 921, 926, 927, 929], "diff:vanished": [], "status_moved": [356, 360, 361, 370, 415, 424, 426, 428, 435, 437, 439, 442, 443, 444, 450, 459, 567, 629, 664, 693, 901, 916, 918, 919, 921, 926, 927, 929], "vanished": []} ...the five injected victims are chosen OUTSIDE this set, so each injected movement has exactly one cause and the equality checks below are not cofounded the NAIVE text scope -- ids whose BASELINE already carries a text field -- holds 1139 of 1293 projected ids; it excludes the first-ascent victim 11, which is the falsifier below falsified baseline: dropped 6 (must APPEAR), added 1550 (must VANISH+ALERT), corrupted the fingerprint of 9 (must CHANGE), the status of 7, the patch_sha of 8, and -- SAMPLE-BLIND -- the patch_status of 709 ok A: the dropped id comes back as APPEARED [6, 1540] ok A: an APPEARED return raises NO alert (a new return is not a changed one) ok A: the ghost id comes back as VANISHED and alerts [1550] ok A: the corrupted fingerprint comes back as CHANGED [7, 8, 9, 360] ok A: the falsified STATUS alerts as a status move [7, 360] ok A: the falsified accepted TEXT alerts as a text move [8] ok A: the SAMPLE-BLIND return is NOT seen -- the sample's blind spot, demonstrated blind id 709 ok A: every member of every set is an INJECTED id, measured drift, or a movement the baseline corroborates -- no alert is raised that no artefact explains {'unexplained': {}, 'late-but-corroborated': {'appeared': [], 'vanished': [], 'status_moved': [], 'accepted_text_moved': [], 'changed': []}} ok A: the scope says SAMPLED, not complete {'complete': False, 'escalated': False, 'probe_range': 1700, 'queried': 399, 'sample_step': 20} ok A: the pass exits 5, so a scheduler can see the alarm 5 ok B: the SAMPLE-BLIND alert is FOUND once the whole range is re-measured [7, 356, 360, 361, 370, 415, 424, 426, 428, 435, 437, 439, 442, 443, 444, 450, 459, 567, 629, 664, 693, 709, 901, 916, 918, 919, 921, 926, 927, 929] ok B: the complete pass is a SUPERSET of the sampled one {'sampled': {'accepted_text_moved': [8], 'status_moved': [7, 360], 'vanished': [1550]}, 'complete': {'accepted_text_moved': [8], 'status_moved': [7, 356, 360, 361, 370, 4 ok B: the scope says COMPLETE OVER THE ALERT-CAPABLE POPULATION and names what it escalated from {'alert_capable': 1294, 'forced_by': '4 alert(s) rang: vanished, status, text', 'ids_added': 1055, 'ids_reused': 390, 'kind_scope': {'kinds': ['status', 'text', 'vanished ok B: the confirmation REUSED this pass's measurements instead of paying for them twice 390 alert-capable ids reused of 399 measured, 1055 added, total 1454 ok B: the measured ids that can carry NO alert are stated, not hidden (the arithmetic adds up) 9 ok B: it cost LESS than the id space it could have walked (the pre-scoping is the saving) 1454 measured vs 1700 ids in 1..1700 (the sampled pass was 399) ok B: on a COMPLETE pass all four typed sets are exactly the injected ids plus what the corpus did -- the drift control measured, and anything later than it verified against the baseline file -- with nothing left over {'unexplained': {}, 'sets': {'changed': [7, 8, 9, 356, 360, 361, 370, 415, 424, 426, 428, 435, 437, 439, 442, 443, 444, 450, 459, 567, 629, 664, 693, 709, 901, 916, 918, ok B: the four typed sets are exact -- drift | injected -- or differ from exact ONLY by ids the corpus moved after the drift control and whose movement the baseline file corroborates {'beyond-drift': {}, 'late-but-corroborated': {}} ok B: the dropped id is reported as APPEARED, and no arrival is INVENTED beyond the measured drift (the arrival set is the one a complete pass cannot make exact -- whether an ARRIVAL is seen at all is the selection rule's coverage, disclosed as `not_queried_existing`) {'invented': [], 'late-but-corroborated': [], 'arrivals the control held': 47} ok B: the pass still exits 5 5 ok B: the gate admitted the claims ok B: the population moves by the ghost's removal (and any return filed since the baseline) 1294 now vs 1293 in the falsified baseline (+47 arrivals measured by the drift control) ok C: a return whose BASELINE projection carried no text field still raises an ACCEPTED TEXT move (the first ascent) [8, 11] ok C: ...and the NAIVE scope -- ids already carrying a text field -- excludes exactly that id, so the narrowing is falsified by a real alert naive scope holds 1139 ids, victim 11 ok C: the scope that DID find it is the projected population, not the text-bearing subset {'alert_capable': 1294, 'complete': True, 'complete_over': 'every alert-capable id, plus the frontier', 'escalated': True, 'excluded_never_existed': 246, 'ids_added': 105 ok C: the per-kind report says how much the narrowing would have saved, and that it is 0 once every kind has rung {'kinds': ['status', 'text', 'vanished'], 'narrowed_by_kind': 0, 'per_kind': {'status': 1293, 'text': 1293, 'vanished': 1293}, 'union': 1294, 'union_all_kinds': 1294} ok C: the text set is accounted for by both injected text movements (the strip and the patch_sha), the measured drift, and nothing else {'unexplained': [], 'late-but-corroborated': []} ok D1: the sample-blind status move raises NO alert on the sampled pass -- a zero at the sample's probability, not a finding {'accepted_text_moved': [], 'status_moved': [360], 'vanished': []} ok D1: every alert the sampled pass DOES raise is explained by an artefact (measured drift or a baseline-corroborated movement), so the injected miss is the only unexplained absence {'unexplained': [], 'raised': [360], 'drift': [356, 360, 361, 370, 415, 424, 426, 428, 435, 437, 439, 442, 443, 444, 450, 459, 567, 629, 664, 693, 901, 916, 918, 919, 921 ok D1: the exit code AGREES with the pass's own alert count (5 iff >0), so the alarm cannot contradict the verdict it reports 1 alerts, exit 5 ok D2: --complete finds it with NO alert having rung first {'accepted_text_moved': [], 'status_moved': [356, 361, 370, 415, 424, 426, 428, 435, 437, 439, 442, 443, 444, 450, 459, 567, 629, 664, 693, 709, 901, 916, 918, 919, 921, ok D2: the reason is NAMED (a corpus-wide reading was claimed), not left implicit --complete/--publish-check: a corpus-wide reading is claimed ok D2: the scope reads COMPLETE and the pass exits 5 5 ok D2: the status set is the blind mutation, plus what the corpus did and the baseline corroborates, and nothing else {'unexplained': [], 'late-but-corroborated': []} ok each of the five passes appended exactly one dated line 41 -> 46 ok the sampled line is DATED, names its basis and its scope ('2026-09-23T16:49:02Z', 'diff') ok the escalated line records the COMPLETE scope {'alert_capable': 1294, 'complete': True, 'complete_over': 'every alert-capable id, plus the frontier', 'escalated': True, 'excluded_never_existed': 246, 'ids_added': 105 ok the escalated line carries the same alert sets as its run series and stdout agree ok the FORCED-COMPLETE line records its reason and the same alerts as its run --complete/--publish-check: a corpus-wide reading is claimed ok the drift control's own measurement is preserved as a DATED series line 2026-09-23T16:48:53Z census diff + watch + escalation selftest: PASS