Record unread-supply as a non-member of absent_reads, not a face

dag/gunbc/recurring_failure_mode/absent_reads_identically_to_never_looked.dag

@@ -41,13 +41,13 @@ data absent_reads_identically_to_never_looked: RecurringFailureMode = RecurringF
 
     "**THE SAME AUDIT MEASURED WHY THE FAMILY IS NOT CONSULTED, AND THE ANSWER IS NOT THAT THE ROSTER SITS SOMEWHERE AUTHORS DO NOT GO.** Across those nine members, at the head audited, there was exactly ONE intra-family boundary citation in the entire family -- the vacuity row naming [[empty_observation_narrow]] -- and eight of nine rows named no sibling at all. THE FAMILY IS THEREFORE UNREACHABLE FROM ITS OWN MEMBERS AT READING TIME, not merely un-surfaced at authoring time, and the distinction decides the remedy: a mechanism that put the RIGHT ROW in front of an author would still have put an ISLAND in front of them. The aggravating measurement is that [[empty_capture_read_as_clean_result]] DID name this row -- as the nickname `blank is not evidence` -- so the one existing edge was invisible to the identity join that found the other eight missing. A NICKNAME IS NOT A WEAKER CITATION, IT IS A DELETED EDGE, which is the DESIGN section 3 argument arriving as a measurement rather than as a principle.",
 
-    "NOT MEMBERS, RECORDED BECAUSE THE IDENTITY SPELLING WOULD ADMIT THEM AND A LATER CONSOLIDATION MUST NOT. [[identity_absent_graph_traversal]] names an absence in the SUBSTRATE -- no identity the model is allowed to name -- and produces no verdict at all, so merging it here would fold a capability gap into a misreading class. [[receipt_whose_competent_refuter_is_absent]] is an absent PARTY, not an absent observation. [[non_execution_undifferentiated_by_what_it_silenced]] reports UNDECIDED rather than pass; its defect is that undecided is undifferentiated by consequence. [[required_lane_green_over_a_population_it_never_offered]] and [[green_reported_over_a_population_the_instrument_does_not_own]] report correctly over a non-empty set they own and are defective in DENOMINATOR SCOPE, which [[roster_is_its_own_denominator]] already separates itself from. [[undecided_fraction_read_as_denominator]] collapses a THIRD disposition rather than an emptiness. [[sweep_terminated_by_process_death_rather_than_a_typed_refusal]] produces no typed refusal at all. [[executed_conjunct_discriminates_nothing]] and [[closure_subject_with_no_enumerable_membership]] are bounded away in the vacuity row own text.",
+    "NOT MEMBERS, RECORDED BECAUSE THE IDENTITY SPELLING WOULD ADMIT THEM AND A LATER CONSOLIDATION MUST NOT. [[identity_absent_graph_traversal]] names an absence in the SUBSTRATE -- no identity the model is allowed to name -- and produces no verdict at all, so merging it here would fold a capability gap into a misreading class. [[receipt_whose_competent_refuter_is_absent]] is an absent PARTY, not an absent observation. [[non_execution_undifferentiated_by_what_it_silenced]] reports UNDECIDED rather than pass; its defect is that undecided is undifferentiated by consequence. [[required_lane_green_over_a_population_it_never_offered]] and [[green_reported_over_a_population_the_instrument_does_not_own]] report correctly over a non-empty set they own and are defective in DENOMINATOR SCOPE, which [[roster_is_its_own_denominator]] already separates itself from. [[undecided_fraction_read_as_denominator]] collapses a THIRD disposition rather than an emptiness. [[sweep_terminated_by_process_death_rather_than_a_typed_refusal]] produces no typed refusal at all. [[executed_conjunct_discriminates_nothing]] and [[closure_subject_with_no_enumerable_membership]] are bounded away in the vacuity row own text. strategy.node_rental SupplyUnobserved after a published-rate-to-region join (crisp-lark-679, 2026-09-12, gunbc-private#92) is the same third-disposition shape, not this row: the constructor KEPT the unread arm, so no emptiness was consumed as an affirmative verdict. Collapsing Unobserved into not-purchasable would have made admissible_as_competitive_floor false for the same reason a sold-out observation would, and would have let cloud-overflow treat unread supply as priced-and-infeasible. That collapse would inhabit [[undecided_fraction_read_as_denominator]], whose recognition rule is a third disposition treated as decided. It is not appended there either -- no census ratio was taken; the conjunction never collapsed -- and minting a third row for a prevented form is the redundancy this ledger exists to avoid. Recorded here so the next appender does not re-file the temptation as a face of an unreached instrument.",
     "**THE SILENT-CONSOLE FACE, WHERE THE INSTRUMENT RETURNED THE SAME BYTES UNDER FOUR DIFFERENT WORLDS (still-dove-458, 2026-09-11, Mt. Collins unit 1 after the 16-to-32 DIMM change).** Four SOL captures -- idle, after a chassis power cycle, after a hand-performed AC cycle, after a chassis reset -- are BYTE-IDENTICAL at 86 bytes, sha256 3aea4368b8745ae7bdca3c7ec052a324032974f7b849bca460e7d0e14f5e0871 (artifacts/bmc/mtcollins1-32dimm/03-sol-idle.log and 99-digests.txt), and the machine's behaviour differed across those conditions: the SEL shows a rail droop after the reset, a soft-power failure and power-cycle after the cycle, and every fan and PSU presence re-asserted after the AC removal. The 86 bytes are the controller's own session banner with ipmitool's tcgetattr complaint -- what a capture contains when the session is live and the host has sent nothing. The observer asserted the strong reading early -- silence as firmware bring-up failure -- and retracted it. THIS IS THE RECOGNITION RULE VERBATIM: what would the instrument have returned had it never reached its subject? The same 86 bytes. So the capture carries no host information, and a verdict about the host built from it is built from nothing. WHAT MAKES IT A FACE OF THIS ROW AND NOT OF [[empty_capture_read_as_clean_result]]: the polarity is inverted -- the empty capture was read as FAILURE, not as clean -- and the reading was retracted by the positive control the rule prescribes, run across conditions rather than planted: four worlds, one output, therefore no measurement. THE REPAIR, in `gunbc.machine_intake_pre_os_bringup_verdict`: the console reading is typed `SolBannerOnly` by DIGEST against the platform's banner digest, the capability profile carries `SolPresentSilent` as a measured arm distinct from absent and from emitting, and the verdict's failure constructor is reachable from the standard SEL window only -- the console enters the fold once, on the empty-window arm, and only to name the cause more precisely. The witness `on_an_emitting_console_a_banner_only_capture_beside_an_empty_window_is_no_restart_not_silence` shows the arm moves when the profile says the console emits. RUNG: structurally impossible for the verdict on this platform -- there is no path from a banner-only reading to a failure arm -- and the platform question (is the console wired at all) stays a measured profile row, which is the honest ceiling for an external fact.",
     "**CORRECTION TO THE SILENT-CONSOLE FACE, SAME LANE, TWELVE HOURS LATER (still-dove-458, 2026-09-12).** The 86 bytes were not the host being quiet; they were the COLLECTOR LEAVING. eager-owl-205 re-ran the same client with a pty and stdin held open (artifacts/bmc/mtcollins1-sol-dram-training-2026-09-12.txt, sha256 e5044f71cde73fe9b2ea1e76c9b894c1ab2f007577b9cbb7e8d50650c2b78a4c): it stayed connected 645 seconds and captured the firmware's DRAM training output, including a stated refusal with a socket and controller. The eight earlier captures were ipmitool printing its banner, hitting stdin EOF, and exiting rc=0 after about three seconds while the wrapper slept out the window. So the recognition rule's question -- what would the instrument return had it never reached its subject -- had the answer `exactly these bytes` for a stronger reason than a silent console: the instrument was NOT THERE for the interval it was cited for. Two things change in the repair. The console capability arm `SolPresentSilent` is no longer what Mt. Collins inhabits and is admissible only on a held-session capture; and the banner is demoted from a proof that the console collector is running to SESSION ACTIVATION ACKNOWLEDGED, with the running proof now `ConsoleSessionHeldPastActivation` -- client alive at an instant after activation, under a deadline that terminates it, because a correctly held client never exits on its own. What does NOT change: a banner-only capture still constructs no host verdict, on any platform. The rung claim above is narrowed accordingly: what is structurally impossible is a verdict from banner-only bytes; which platforms emit is a measured profile row, and this one was measured wrong once by the instrument this row exists to describe. The collector-side defect is filed by that lane as the seventh instance of [[empty_capture_read_as_clean_result]].",
 
     "**THE CONFLATION CAN HAPPEN ENTIRELY OUTSIDE THE PRODUCER, IN A READER THE PRODUCER NEVER SEES, AND EVERY GUARD ABOVE IS BLIND TO IT** (snappy-pike-154, 2026-09-12, M2 compliance spine, gunbc#11188 and gunbc-private#94). The specimens above are instruments that failed to reach a subject, and their repairs are positive controls and denominators -- things the producer can assert about its own run. THIS ONE HAS A CORRECT INSTRUMENT, A HONEST CARRIER AND A TRUE EMPTY VALUE, and the affirmative verdict is supplied by a READER OUTSIDE THE SYSTEM. A customer projection was to derive provider, subprocessor and location disclosures from the ADMITTED ROUTE POPULATION, which was empty and correctly empty. Rendered, that is a privacy notice promising nothing and a subprocessor snapshot with no entries. A COUNTERPARTY DOES NOT READ AN EMPTY DISCLOSURE AS NOTHING WAS ASSESSED; THEY READ IT AS THERE IS NOTHING TO DISCLOSE -- the opposite meaning, supplied by them, from an artifact that lied about nothing. No control fires, because the instrument worked; no denominator helps, because the denominator really is zero. The repair is not instrumentation at all: the artifact must REFUSE TO RENDER rather than render empty, because emptiness crossing an accountability boundary acquires a meaning the producer cannot see and cannot caveat. PLACED AGAINST ITS NEAREST NEIGHBOURS, since the receipt above this one measured that eight of nine family members name no sibling and are therefore unreachable from each other at reading time: this is NOT [[predicate_vacuously_true_on_an_empty_domain]], where the domain is genuinely empty and the check ran correctly, because there the supply is a bound on the domain and the reader is inside the system; it is NOT [[required_evidence_absent_reads_as_evidence_of_pass]], where the missing thing is evidence OWED TO US and the supply is demanding it, whereas here the record is complete and correct; and it is NOT [[empty_observation_narrow]], where the observation itself was too narrow to have seen the subject. The pairwise non-substitutability holds in the direction that matters: a domain bound, a demanded record and a widened observation all leave the empty artifact renderable, and none of them stops a counterparty reading zero disclosures as zero to disclose. THE DISTINGUISHING FACT IS WHO SUPPLIES THE VERDICT AND WHETHER WE CAN REACH THEM -- which is why the repair is a refusal at the boundary rather than any improvement to the measurement, and why DESIGN section 5's externalization arm is the neighbouring law rather than its instrumentation guidance.",
 
-    "THE SAME LANE CARRIES THE PREVENTED FORM ONE LAYER DOWN, WHICH IS WHY THE CARRIER IS WORTH NAMING: product.fabric.provider_route ProviderPathStanding is Located | Absent | Unread, so on the Hetzner census (gunbc#11188) NINE data paths on two routes were answered and ProviderPathAbsent was taken ZERO times -- not because every path is populated, but because nobody had stated that telemetry or spill do not exist, and Absent is a claim rather than a silence. Both routes refuse. The contrast is the whole lesson: where the distinction is a CONSTRUCTOR the representation PRESERVES it and forces an explicit choice -- the census under pressure to look complete chose Unread -- whereas where it is a RENDERED EMPTY LIST nothing stops the collapse downstream.",
+    "THE SAME LANE CARRIES THE PREVENTED FORM ONE LAYER DOWN, WHICH IS WHY THE CARRIER IS WORTH NAMING: product.fabric.provider_route ProviderPathStanding is Located | Absent | Unread, so on the Hetzner census (gunbc#11188) NINE data paths on two routes were answered and ProviderPathAbsent was taken ZERO times -- not because every path is populated, but because nobody had stated that telemetry or spill do not exist, and Absent is a claim rather than a silence. Both routes refuse. The contrast is the whole lesson: where the distinction is a CONSTRUCTOR the representation PRESERVES it and forces an explicit choice -- the census under pressure to look complete chose Unread -- whereas where it is a RENDERED EMPTY LIST nothing stops the collapse downstream. THE NODE-RENTAL SUPPLY CELL IS THAT CONSTRUCTOR AGAIN, NOT A FACE OF THIS ROW: SupplyUnobserved survived the region join, overflow stayed SelectionNeedsEvidence, and overflow_quota_increase_authorized stayed false. The operator half (CAX purchasability, GCP/OCI supply, control-plane siting) is not discharged by naming the neighbour.",
 
     "AND THE EMPTY POPULATION HAD TWO INDEPENDENT CAUSES WHOSE REMEDIES DIFFER, WHICH A SINGLE EMPTY VALUE CANNOT CARRY: the route facts were incomplete (a data path unread, a downstream unassessed), AND no compliance posture or sanction roster had ever been declared, so even a fully Located route would have been admitted by nothing. Reporting one when the other holds names the wrong thing to go and fix. Recorded with the miss that produced it: the lane manager twice instructed the worker to ship the empty disclosures, and the worker refused both times -- once declining to infer that no processing occurs from no admitted routes, once declining to render a boundary no permit could produce. THE WORKER WAS RIGHT AND THE INSTRUCTION WAS WRONG, and it is filed that way because an author under instruction is exactly the position from which this class is hardest to refuse.",
   ],