From d045f309fa14b36fb5f4a67bf1efbb6167bb0c84 Mon Sep 17 00:00:00 2001 From: Brian Searls Date: Fri, 28 Aug 2026 18:32:31 +0000 Subject: [PATCH] The censored figure is not a cost: 502ms and 2902ms render as the same number `floor_cost_debt` already states that an interrupted claim's reported figure is a lower bound. That sentence is true and it was not enough: three readers independently reasoned from the cluster's SHAPE as if the figures were point estimates, and the roster and ceiling decisions on this axis were argued from that reading. Measured with a controlled pair of floor runs differing only in the ceiling -- 33191118450 (500ms CPU) and 33190645319 (raised), same tree, both artifacts' own `required_floor_claim_cost.tsv` joined on identity -- the bound and the true cost are uncorrelated over the 12 interrupted rows. Seven rows reported at 502-527ms measure 1596-2907ms uncensored; four measure UNDER the ceiling. `502 -> 2902` and `502 -> 355` are the same reported number. That refutes two live arguments by name: this roster's own "membership is decided by run-to-run variance" (true of the four `bootstrap_footprint` siblings it was applied to, false of the seven rows beside them) and the 5000ms-era handoff's "26 of 44 sit within 2% of the ceiling, so scheduling noise decides them" (those 26 true costs are unmeasured and unbounded above). Agreement among the three readers was not corroboration: all three were reading one column that cannot express the distinction they were drawing. No ceiling moves and no roster row is added or removed. The construction this wants -- a censored figure that cannot inhabit the measured-cost slot in `write_required_floor_claim_cost_tsv` -- is declared with its trigger and deliberately not built, under the 2026-08-28 forward-implementation freeze. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_01QsmcgXPM8LdvEhUzadHkuL --- src/v2/workflow/floor_cost_debt.dag | 70 +++++++++++++++++++++++++++++ 1 file changed, 70 insertions(+) diff --git a/src/v2/workflow/floor_cost_debt.dag b/src/v2/workflow/floor_cost_debt.dag index e0efaec160d..b2beaf17b40 100644 --- a/src/v2/workflow/floor_cost_debt.dag +++ b/src/v2/workflow/floor_cost_debt.dag @@ -196,6 +196,76 @@ import v2.std.text { String } // Verified against the same cost receipt before enrolment: all 46 are `budget_interrupted` or // `completed_over_budget` and NONE has outcome `fail`. // +// HOW UNINFORMATIVE THAT LOWER BOUND IS, MEASURED RATHER THAN ASSERTED -- and the number is the +// point, because THREE READERS INDEPENDENTLY TREATED IT AS A POINT ESTIMATE. This carrier says +// above that a censored figure is a lower bound. It is, and the sentence is not enough: the bound +// and the true cost are essentially UNCORRELATED in this population, so any reasoning that sorts, +// buckets or thresholds on the reported figure is unsound even when every individual figure is +// honest. +// +// THE INSTRUMENT IS A PAIR OF FLOOR RUNS DIFFERING ONLY IN THE CEILING, so the join is a +// controlled measurement and not two observations of two trees. Censored: run 33191118450 +// (`session/witty-wren-148`), 500ms CPU. Uncensored: run 33190645319 (`probe/ceiling-uncensor-9517`), +// ceiling raised, same tree. Both emit `required_floor_claim_cost.tsv`; the rows below are that +// artifact's own `cpu_ms` column joined on `identity`. The figures are transcribed here, as a +// declared exception to "name the instrument, never transcribe its output", for the reason this +// document grants that exception elsewhere: the probe branch is DELETED, so no producer on this +// side of the boundary can re-derive the uncensored half, and the subject of the row IS the +// divergence between the two runs rather than either run's state. +// +// identity censored -> true (cpu_ms) +// ..cost_coverage_witness.cost_coverage_dag_logic_has_three_fn_bodies 506 -> 2897 +// ..cost_coverage_witness.cost_coverage_file_row_totality_holds_on_smoke 506 -> 2907 +// ..cost_coverage_witness.cost_coverage_v2_ingested_body_is_not_decl.. 502 -> 2902 +// ..identity_captured_navigation_roster_gate.green_control_sanctioned.. 507 -> 2227 +// ..emit_on_demand_classical_not_ingested_family_witness.emit_on_demand.. 502 -> 2374 +// ..emit_ingest_grammar_relation_round_trip.same_grammar_parse_ingest.. 527 -> 1658 +// ..nominal_distinctness_cross_call.nominal_distinct_control_compiles_ok 505 -> 1596 +// ..lens_module_gate_witness.lens_closure_question_zero_holds_live 1655 -> 1713 +// ..external_model_scope_live_cover_witness.frontier_cover_of_live.. 501 -> 559 +// ..operation_argv_corpus_witness.operation_argv_corpus_every_row_is.. 502 -> 520 +// ..roadmap_belt_witness.witness_belt_plan_equals_step_composition 502 -> 355 +// ..roadmap_belt_witness.witness_belt_plan_paused_holds_on_authority 502 -> 380 +// +// READ THE FIRST AND THE ELEVENTH LINES TOGETHER. `502 -> 2902` and `502 -> 355` are THE SAME +// REPORTED NUMBER. One row costs nearly six times the ceiling; the other is comfortably under it +// and was interrupted only because the fold's charged clock crossed the line during it. The +// reported figure does not distinguish them, and no amount of care in reading it can, because the +// information is not in the column. +// +// WHAT THAT REFUTES, NAMED SO THE READING DOES NOT RETURN. Two decisions on this axis were argued +// from the shape of this cluster, and both arguments are void: +// - "membership is decided by run-to-run variance" (this branch's own PR body, over four +// `bootstrap_footprint` siblings at 492/502/502/506ms). TRUE OF THOSE FOUR -- they measure +// 460ms uncensored -- and FALSE of the seven rows above them, which measure 1596-2907ms. The +// reasoning could not tell the two groups apart, so it was right by luck on the group it was +// applied to. +// - "26 of 44 sit within 2% of the 5000ms ceiling, so scheduling noise decides them" (the +// handoff for the 5000ms-era population, `calm-ram-380`). Same shape one ceiling up, and +// the same refutation: those 26 rows' true costs are UNMEASURED and unbounded above. +// +// AND THE AGREEMENT WAS NOT CORROBORATION. Three readers reached the marginality reading +// independently, which reads as confirmation and is not: a shared input is a shared potential +// defect, and all three were reading one column that cannot express the distinction they were +// drawing. That is this document's own recognition rule applied to its own authors. +// +// WHAT THIS DOES NOT LICENSE. It is not an argument to move the ceiling -- if anything it is the +// opposite, since the rows nearest the line are the ones most likely to be far over it. It is not +// an argument to enrol or release any row here; the four `bootstrap_footprint` siblings measure +// under the ceiling uncensored and are NOT on this roster, which is correct. It settles one thing +// only: a censored figure may not be used as a cost, and a decision that needs a cost for an +// interrupted row must uncensor and re-measure, as the pair above did. +// +// THE CONSTRUCTION THIS WANTS IS NOT BUILT, AND THAT IS A DECLARED OMISSION RATHER THAN AN +// OVERSIGHT. `write_required_floor_claim_cost_tsv` renders a censored lower bound into the same +// `cpu_ms`/`wall_ms` columns a completed row's exact measurement occupies, distinguished only by +// the adjacent `verdict_reached` column -- which is an exact discriminator (false for exactly the +// 12 `budget_interrupted` rows and nothing else) that every reader above failed to join. Making +// the bound unable to inhabit the measured-cost slot is the DESIGN section 5 construction move, +// and it was deliberately not landed here: forward implementation is frozen under the 2026-08-28 +// wind-down. TRIGGER: when that freeze lifts, a censored figure gets a rendering a numeric reader +// cannot silently consume, and this paragraph's warning becomes the regression control for it. +// // THE ONE THING NEITHER POPULATION MAY CONTAIN IS A KNOWN FAILURE, and the reason is a live // receipt rather than a principle. A first attempt at this roster was measured on THIS branch, // with the 500ms ceiling already in place, and enrolled 276 interrupted rows. The tightened