§1 measurement keystone (instrument): emit floor concurrent-peak RSS (VmHWM) so width derives from a MEASURED peak, not a hand-grounded literal by gunbai-bot[bot] · Pull Request #5431 · gunb-ai/gunbc

gunbai-bot · 2026-06-21T04:20:09Z

What

Phase-0 measurement keystone — the instrument half (per realization-measurement-loop.md §0; scoped by bright-stag-194 to instrument only, defer the economic-tier consumer).

The floor already emits measured step timing ([measurement] discovery corpus: … resolve <ms>, evalu <ms>) and the model carriers already exist (CostAccount.basis, PerformanceReceipt, the roll-up lenses, and the floor-enrolled discriminator witness realization_measurement_keystone_test.dag). The one missing measured fact was memory: the floor executor emitted no peak-RSS at all.

This adds it: after the floor run, claim_executor reads process VmHWM (/proc/self/status) and emits

[measurement] floor peak RSS: <bytes> (VmHWM) at spawn_width=<N>

VmHWM is the concurrent peak across the whole floor run at that spawn_width — exactly the quantity a memory budget must bound. Per-shard peak ≈ this ÷ effective concurrency (a width=1 run measures one shard's peak directly).

Why (construction, not a hand-authored literal)

This is the measured fact that lets the §1-C memory-aware width derivation key on a measured peak rather than the hand-grounded 14 GB literal #5419 deleted as unwired. Same anti-pattern as the cache version-string: a pinned constant standing in for a fact that should be measured. The instrument produces the fact; the width-fold that consumes it is the deferred Phase-1 step.

ci_floor_plan.dag:256 (total: cost_account_predicted_zero()) intentionally stays predicted at plan time — there is no measurement before the run; the measured→plan feedback is the deferred consumer, not this PR.

Scope / non-goals

IN: emit the measured concurrent-peak RSS (the missing instrument output).
OUT (deferred, named): the width-fold consuming the measured peak (§1-C); the measured→plan feedback loop; the economic tier (P4). These wait on the consumer phase per manager scope.

Verification

Builds clean (cargo build -p v1-compiler --bin claim_executor --release, exit 0).
Artifact content-verified (not just exit code — dogfooding §1-A): strings target/release/claim_executor | grep 'floor peak RSS' → present.
Additive + pure: the emit is a post-run /proc read + eprintln; it does not touch eval, so witness verdicts are unperturbed (the §5 purity property holds by construction for this edit). Fail-soft: prints unavailable (never fabricates) when /proc/self/status is absent (non-linux).

🤖 Generated with Claude Code

…maining is the Phase-1 consumer Per quick-ant-298: the measurement keystone was nearly complete — model side already floor-enrolled, step timing already emitted; the only gap was peak-RSS, closed by #5431. P4's Phase-0 dependency is satisfied; remaining is the Phase-1 measured->plan feedback + width-fold (also unblocks §1-C). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ce-hawk #5427) The 3-filter allowlist was COST selection, not arbitrary gatekeeping — the v1 SEED compiler costs ~tens of CPU-sec per trivial test, so run-all-per-PR is CPU-hours (off the table). True shape: per-PR cost-bounded subset + measured #[ignore="expensive: Ns"] + completeness lens (#5427); nightly --ignored lane as the destination for expensive + the 58 currently-ignored tests (owned by §1/quick-ant, after #5431, escalate for load-bearing CI-gen). Completeness = every test runs on >=1 cadence (fail-closed). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…budgets) + plan doc (#5436) * WIP: dsl -> v2 scoping * WIP: ROADMAP planning * WIP: ROADMAP planning * WIP: ROADMAP planning * ROADMAP: scannable dependency-ordered checklist; consolidate caching plan (de-fork zesty-deer-479 owner, absorb quick-ant-298 spine) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * self-host: add bootstrap purity (no stage0 hand-edits / regen-lockstep keystone) + precise v1 cutover Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §0 fail-closed lock-down (blocks expansion) + audit doc; §4 website demo Audit: cache lossy-digest flake (resolved_graph_cache.rs:146, verified), ~inert analytical lenses (complexity/cost/etc), regen --verify unwired (#5325). Lock-down checklist gates expansion. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP: flesh §2 idea->idea compiler (medium/language axes); §0 → lock-down LANE (audits→fixes→meta), name model<->realization fork as suspected root Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * lock-down: add CI-coverage-completeness audit (rust gate runs 3 of 60 v1 suites) + axiom/syllogism lens (DESIGN open thread #1 — lock down the reasoning) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * roadmap §0/§7 + lockdown: lead with correctness-by-construction, demote lenses to residue Folds in the operator principle (relayed via quick-ant-298): a lens is validation — it concedes the bad thing is writable. Root-cause to make it unwritable (single authority / realization derived from model); reserve lenses for the genuinely- unstructurable (complexity/necessity). #5423's spec-only key lens shipped a false-green as the live proof. - ROADMAP §0: add the principle; split Fixes into tier-1 construction (dissolve model↔realization fork; cache-key derived-from-declared-inputs; self-host purity by construction) and tier-2 lens (complexity/cost; cache-redundancy; purity oracle; promote-inert). Meta-invariant → construction-justification rule. - ROADMAP §7: P1 cache-key reframed from 'realizer-key lens' to key derived from declared inputs_considered (construction). - fail-closed-lockdown.md: construction principle in the thesis; §4 checklist re-ordered construction-first / lens-residue; meta = construction-justification. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * roadmap §0: add Disposition carrier + 'confront skipped modeling decisions' Captures the lens/coproduct disposition decision (operator). One typed carrier (Terminal{reason} | Scaffold{dissolves_to}) for BOTH lens-lifecycle tags AND coproduct dissolve-markers — today freeform 🟡 comments, unreadable by lens since comments aren't Nodes. Decision: middle path (construction-capable carrier + selectively-enforcing lens that ratchets coverage) now, #1 (substrate can't-define-untagged) as the named end-state. The lens is itself a Scaffold{dissolves_to: substrate-mandatory-tag} — self-dissolving when coverage = whole tree. Rejected jumping to #1 on sequencing (load-bearing §4 substrate change → escalate; flag-day migration; derived coproducts need disposition derived not authored), not on principle. Enforceability split: presence = construction (non-optional field, no meta-lens); redundancy (scaffold + successor both present) = hard gate; Terminal-vs-Scaffold correctness = retro/judgment (synthesis-feasibility limit). - docs/plans/disposition-carrier.md (new) - ROADMAP §0 tier-1 + meta 'confront skipped decisions' standing practice Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * DESIGN §5/§6: promote construction-over-validation; roadmap: scope-partition §0, testgen §1, shelve dashboard Addresses the review's four flags + sequencing nuance. - DESIGN.md §5: 'correctness by construction, not validation' is now an axiom (a check re-stating a model constraint is a 2nd representation §2/§3; prefer realization derived from a single authority; reserve checks for the unstructurable residue). §6 'enforce with lenses' reconciled: construction first, lens = residue mechanism, AND the executable inert-lens backstop is NOT superseded by the authoring-time construction-justification judgment. (flag 4 home + flag 3) - ROADMAP §0 partitioned: In-scope this window (numeric-tower grounding; cache trustworthy + warm==cold oracle shipped NOW as detective; widen rust gate; promote inert lenses) vs Fenced-OUT fan-out (Value::Null 131-site split; self-host purity gate; cross-tree import activation; Disposition carrier). Honest framing: window reduces fail-open surface, does NOT 'lock' the class — Null split stays open. (flags 1, 2, sequencing nuance) - ROADMAP §0 meta: restored executable inert-lens hygiene backstop, construction- justification layered on top (not 'supersedes'). (flag 3) - De-dup: principle no longer restated in ROADMAP/lockdown §0; both point to DESIGN §5. cache-key construction homed in §7, §0 references it. (flag 4) - ROADMAP §1 = testgen as bug-class oracle (+ affected-set completeness half + parked anemia lens); dashboard shelved to §8. - docs/plans/testgen-oracle.md (new), fail-closed-lockdown.md realigned. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * DESIGN §5/§6/§7: wall-vs-ratchet decidability, displaced-pain denominator, open language design Folds in the operator's product thesis + the two bounds that keep it honest. - §5: construction makes a class unwritable only when membership is DECIDABLE — trichotomy (wall now / wall after grounding / ratchet forever); 'never' is the trap (lets an undecidable ratchet masquerade as a wall — optimality by Rice). - §6: denominate the benefit — the deliverable is a displaced cost (§1 time / a paid-for pain), the lens/substrate is the moat not the product; priced in elegance the work is unbounded (the economic twin of 'never'). - §7: the recursion's payoff — language design itself opens up. It's locked by cost (a check = a compiler fork; a language = an adoption problem); both dissolve here (a wall is a row §2, applied over a medium-agnostic substrate §4), so (compiler-fork × language) → (row + medium). Sound where ingest is Lossless, fail-closed where not (DecodeFidelity §4). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * ROADMAP: fix doc-graph violations from bright-eagle-46 review of #5424 Apply the apex axiom/syllogism lens (single authority / no orphan / no cycle) to the roadmap itself — manual acyclicity pass: - orphan: testgen-oracle.md backlinked §1 → repoint §4 (its own lane) - single authority: §0 cache-key now a pure pointer (= §2 F2/F3/P1); §0 numeric-tower marked the authoritative home (§5 de-fork / fork plan point here, no second checkbox) - §0↔§5 cycle: self-host purity reframed as a §5 deliverable §0's expansion-gate depends on (edge §5 → §0-gate → products), not §0-owned - undeclared edge: §7 react/html declares its dependency on §6 media - backlink sweep: the reorg had broken every numeric backlink across 7 plan docs; re-point all and anchor each to the stable section TITLE so a future renumber can't silently break them again Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §3 + plan: algorithmic-cost reduction by construction (rewrite, not budget) Reframe §3 from per-fn complexity budgets to the actual intent: rewrite common suboptimal patterns (O(n²)→O(n), O(2ⁿ)→O(n), O(n)→O(log n)) to the cheaper equivalent — construction on the cost axis, not a warning. New plan doc docs/plans/algebraic-rewrite-optimization.md captures the up-front design: the decidability split (modeled EffectShape makes the preconditions structural; equivalence stays undecidable so no optimality oracle), rewrite-rule-as-row + once-proven soundness, the common-case catalog tiered by precondition, D1 canonical-form-is-truth / D2 two seed rules / D4 constant-factor deferred, the four-witness DONE bar (incl. the non-firing control half-done versions skip), and a corpus hit-rate acceptance gate. complexity.dag is the cost oracle; synthesis.dag stays the advisory undecidable residue. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §2: Phase-0 measurement instrument done (#5431 peak-RSS) — remaining is the Phase-1 consumer Per quick-ant-298: the measurement keystone was nearly complete — model side already floor-enrolled, step timing already emitted; the only gap was peak-RSS, closed by #5431. P4's Phase-0 dependency is satisfied; remaining is the Phase-1 measured->plan feedback + width-fold (also unblocks §1-C). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * plan/§3: detection-vs-enforcement containment (E⊆D′⊆D) + explicit seed-rule I/O up front Grounded in cost.dag U2 + complexity.dag (investigated, not theorized): - detection is TOTAL by construction (kernel-level cost fold; arbitrary fns detectable); boundary is precision (ClassUnknown), not coverage - enforced rewrites are a strict subset structurally guaranteed by the class-drop witness: E ⊆ D′(precise) ⊆ D(all) - n√n excluded for a MODEL reason (PolynomialDegree is integer-only, n^1.5 unrepresentable); ternary search excluded (log base is not a class) - today's small gate roster = subject-production limit (fn-body reflection), NOT a detection limit - new §3a fully specifies the two seed rules up front: input→output→ precondition→non-firing control→discriminating equivalence input, so the worker builds to spec and the project can actually finish Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §0/§1: rust-gate is cadence-decoupling, not run-all (per fierce-hawk #5427) The 3-filter allowlist was COST selection, not arbitrary gatekeeping — the v1 SEED compiler costs ~tens of CPU-sec per trivial test, so run-all-per-PR is CPU-hours (off the table). True shape: per-PR cost-bounded subset + measured #[ignore="expensive: Ns"] + completeness lens (#5427); nightly --ignored lane as the destination for expensive + the 58 currently-ignored tests (owned by §1/quick-ant, after #5431, escalate for load-bearing CI-gen). Completeness = every test runs on >=1 cadence (fail-closed). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * plan §2a: generalize by structural-redundancy keying (O(n^x)->O(n^(x-1)) free); flag n-log-n as substitution Per operator: catalog must generalize polynomial-degree reduction without edge cases. Resolution: rules key on the structural redundancy, never on degree — degree is not evidence of redundancy (would fire on genuine O(n^x)). A structurally-keyed nested-membership->set peels one level wherever it matches; fold-to-fixpoint gives O(n^3)->O(n^2)->O(n). Cost model supports arbitrary integer degree, so witness (b) holds at every peel. Flagged OPEN (operator input invited): O(n^x)->O(n log n) is algorithmic SUBSTITUTION (different algorithms, same I/O) not redundancy elimination — verges on undecidable equivalence; tractable form is per-idiom rules (sort-based dedup, repeated-min->heap), not a parameterized rule. Seed Rule 1 now authored structurally + carries a depth-2 generalization witness. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * plan §1b: Unknown is an anemic atom — dissolve over time (reuse Disposition), never a false pass Per operator: classifying Unknown isn't a fixed up-front split — it's the standing anemic-leaf dissolution practice (DESIGN §2 decompress->map->reduce) applied to the cost lens. UnknownCost{diagnostic} already carries its reason; the anemia is the free-form reason. Each decomposition resolves an Unknown to construction (now-precise class -> new D′) or a grounded Terminal (genuinely undecidable, positively recognized -> advisory comment). DFS-first: this IS the Disposition carrier (resolves to construction-or-justified-Terminal), so reuse it, don't fork an unknown-reason enum. Supersedes the static Undecidable|Undetermined split. Two invariants fixed up front: never a false pass (Unknown=>Violates, already holds); every Unknown on the dissolution frontier. cost.dag enrichment + un-parking Disposition are operator-gated. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP: §3 reverts to budget-gate validation (stability); rewrite-engine relocated to §5 (post-stability) Operator decision 2026-06-21: budget-gate validation is fine for the stability window; the algorithmic-cost REWRITE construction design is expansion, homed with self-hosting (§5) — IR-rewrite/canonicalization is most natural once .dag is the self-hosted truth. - §3 = complexity budget gate (validation): cost-lens symbolic_max fix (#5437) + per-fn subject + budget-gates-whole-codebase (gated on fn-body reflection) + synthesis advisory. #5437 foundation stays in-window. - §5 gains an 'adjacent expansion lane' = the rewrite engine, pointing at the preserved plan doc; marked post-stability. - plan doc status -> POST-STABILITY EXPANSION, relocated to §5. Nothing deleted — the rewrite design is preserved, just fenced out of the stability window (same as Disposition / Value::Null-split). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…pinned 4) Wire the measurement->plan loop #5431 opened: gunbc_ci_floor_spawn_width now returns min(shard_count, hardware_threads, floor(0.8*live_budget / measured_peak)) instead of hardware_thread_count(4). - gunbc.ci_floor_measurement (new): committed MEASURED per-shard peak, stamped from #5431's VmHWM emit (run 27893441091: 8400113664 bytes at width 4 => /4 = ~2.1GB/shard), NOT the ~14GB hand-grounded literal #5419 deleted. + dissolve-on drift-gate marker. - std.realization_width: memory_aware_spawn_width fold (pure, v2-inheriting) with a 0.8 safety fraction; fail-closed to a conservative committed width when budget unreadable. - claim_executor: reads the LIVE cgroup memory.max (meminfo fallback) and threads it into the .dag fold via run_in_context_with_args (DESIGN §3 measured=>peripheral: we commit our measurement, read the external-authored budget live). - witnesses: non-binding (4->7), tight-budget backoff, unreadable->fallback, fraction load-bearing, sub-unit->1. Proven by execution: derived spawn_width=7. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ystone):

…gestion⁻¹ past syntax (#5442) * WIP: dsl -> v2 scoping * WIP: ROADMAP planning * WIP: ROADMAP planning * WIP: ROADMAP planning * ROADMAP: scannable dependency-ordered checklist; consolidate caching plan (de-fork zesty-deer-479 owner, absorb quick-ant-298 spine) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * self-host: add bootstrap purity (no stage0 hand-edits / regen-lockstep keystone) + precise v1 cutover Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §0 fail-closed lock-down (blocks expansion) + audit doc; §4 website demo Audit: cache lossy-digest flake (resolved_graph_cache.rs:146, verified), ~inert analytical lenses (complexity/cost/etc), regen --verify unwired (#5325). Lock-down checklist gates expansion. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP: flesh §2 idea->idea compiler (medium/language axes); §0 → lock-down LANE (audits→fixes→meta), name model<->realization fork as suspected root Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * lock-down: add CI-coverage-completeness audit (rust gate runs 3 of 60 v1 suites) + axiom/syllogism lens (DESIGN open thread #1 — lock down the reasoning) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * roadmap §0/§7 + lockdown: lead with correctness-by-construction, demote lenses to residue Folds in the operator principle (relayed via quick-ant-298): a lens is validation — it concedes the bad thing is writable. Root-cause to make it unwritable (single authority / realization derived from model); reserve lenses for the genuinely- unstructurable (complexity/necessity). #5423's spec-only key lens shipped a false-green as the live proof. - ROADMAP §0: add the principle; split Fixes into tier-1 construction (dissolve model↔realization fork; cache-key derived-from-declared-inputs; self-host purity by construction) and tier-2 lens (complexity/cost; cache-redundancy; purity oracle; promote-inert). Meta-invariant → construction-justification rule. - ROADMAP §7: P1 cache-key reframed from 'realizer-key lens' to key derived from declared inputs_considered (construction). - fail-closed-lockdown.md: construction principle in the thesis; §4 checklist re-ordered construction-first / lens-residue; meta = construction-justification. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * roadmap §0: add Disposition carrier + 'confront skipped modeling decisions' Captures the lens/coproduct disposition decision (operator). One typed carrier (Terminal{reason} | Scaffold{dissolves_to}) for BOTH lens-lifecycle tags AND coproduct dissolve-markers — today freeform 🟡 comments, unreadable by lens since comments aren't Nodes. Decision: middle path (construction-capable carrier + selectively-enforcing lens that ratchets coverage) now, #1 (substrate can't-define-untagged) as the named end-state. The lens is itself a Scaffold{dissolves_to: substrate-mandatory-tag} — self-dissolving when coverage = whole tree. Rejected jumping to #1 on sequencing (load-bearing §4 substrate change → escalate; flag-day migration; derived coproducts need disposition derived not authored), not on principle. Enforceability split: presence = construction (non-optional field, no meta-lens); redundancy (scaffold + successor both present) = hard gate; Terminal-vs-Scaffold correctness = retro/judgment (synthesis-feasibility limit). - docs/plans/disposition-carrier.md (new) - ROADMAP §0 tier-1 + meta 'confront skipped decisions' standing practice Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * DESIGN §5/§6: promote construction-over-validation; roadmap: scope-partition §0, testgen §1, shelve dashboard Addresses the review's four flags + sequencing nuance. - DESIGN.md §5: 'correctness by construction, not validation' is now an axiom (a check re-stating a model constraint is a 2nd representation §2/§3; prefer realization derived from a single authority; reserve checks for the unstructurable residue). §6 'enforce with lenses' reconciled: construction first, lens = residue mechanism, AND the executable inert-lens backstop is NOT superseded by the authoring-time construction-justification judgment. (flag 4 home + flag 3) - ROADMAP §0 partitioned: In-scope this window (numeric-tower grounding; cache trustworthy + warm==cold oracle shipped NOW as detective; widen rust gate; promote inert lenses) vs Fenced-OUT fan-out (Value::Null 131-site split; self-host purity gate; cross-tree import activation; Disposition carrier). Honest framing: window reduces fail-open surface, does NOT 'lock' the class — Null split stays open. (flags 1, 2, sequencing nuance) - ROADMAP §0 meta: restored executable inert-lens hygiene backstop, construction- justification layered on top (not 'supersedes'). (flag 3) - De-dup: principle no longer restated in ROADMAP/lockdown §0; both point to DESIGN §5. cache-key construction homed in §7, §0 references it. (flag 4) - ROADMAP §1 = testgen as bug-class oracle (+ affected-set completeness half + parked anemia lens); dashboard shelved to §8. - docs/plans/testgen-oracle.md (new), fail-closed-lockdown.md realigned. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * DESIGN §5/§6/§7: wall-vs-ratchet decidability, displaced-pain denominator, open language design Folds in the operator's product thesis + the two bounds that keep it honest. - §5: construction makes a class unwritable only when membership is DECIDABLE — trichotomy (wall now / wall after grounding / ratchet forever); 'never' is the trap (lets an undecidable ratchet masquerade as a wall — optimality by Rice). - §6: denominate the benefit — the deliverable is a displaced cost (§1 time / a paid-for pain), the lens/substrate is the moat not the product; priced in elegance the work is unbounded (the economic twin of 'never'). - §7: the recursion's payoff — language design itself opens up. It's locked by cost (a check = a compiler fork; a language = an adoption problem); both dissolve here (a wall is a row §2, applied over a medium-agnostic substrate §4), so (compiler-fork × language) → (row + medium). Sound where ingest is Lossless, fail-closed where not (DecodeFidelity §4). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * ROADMAP: fix doc-graph violations from bright-eagle-46 review of #5424 Apply the apex axiom/syllogism lens (single authority / no orphan / no cycle) to the roadmap itself — manual acyclicity pass: - orphan: testgen-oracle.md backlinked §1 → repoint §4 (its own lane) - single authority: §0 cache-key now a pure pointer (= §2 F2/F3/P1); §0 numeric-tower marked the authoritative home (§5 de-fork / fork plan point here, no second checkbox) - §0↔§5 cycle: self-host purity reframed as a §5 deliverable §0's expansion-gate depends on (edge §5 → §0-gate → products), not §0-owned - undeclared edge: §7 react/html declares its dependency on §6 media - backlink sweep: the reorg had broken every numeric backlink across 7 plan docs; re-point all and anchor each to the stable section TITLE so a future renumber can't silently break them again Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §3 + plan: algorithmic-cost reduction by construction (rewrite, not budget) Reframe §3 from per-fn complexity budgets to the actual intent: rewrite common suboptimal patterns (O(n²)→O(n), O(2ⁿ)→O(n), O(n)→O(log n)) to the cheaper equivalent — construction on the cost axis, not a warning. New plan doc docs/plans/algebraic-rewrite-optimization.md captures the up-front design: the decidability split (modeled EffectShape makes the preconditions structural; equivalence stays undecidable so no optimality oracle), rewrite-rule-as-row + once-proven soundness, the common-case catalog tiered by precondition, D1 canonical-form-is-truth / D2 two seed rules / D4 constant-factor deferred, the four-witness DONE bar (incl. the non-firing control half-done versions skip), and a corpus hit-rate acceptance gate. complexity.dag is the cost oracle; synthesis.dag stays the advisory undecidable residue. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §2: Phase-0 measurement instrument done (#5431 peak-RSS) — remaining is the Phase-1 consumer Per quick-ant-298: the measurement keystone was nearly complete — model side already floor-enrolled, step timing already emitted; the only gap was peak-RSS, closed by #5431. P4's Phase-0 dependency is satisfied; remaining is the Phase-1 measured->plan feedback + width-fold (also unblocks §1-C). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * plan/§3: detection-vs-enforcement containment (E⊆D′⊆D) + explicit seed-rule I/O up front Grounded in cost.dag U2 + complexity.dag (investigated, not theorized): - detection is TOTAL by construction (kernel-level cost fold; arbitrary fns detectable); boundary is precision (ClassUnknown), not coverage - enforced rewrites are a strict subset structurally guaranteed by the class-drop witness: E ⊆ D′(precise) ⊆ D(all) - n√n excluded for a MODEL reason (PolynomialDegree is integer-only, n^1.5 unrepresentable); ternary search excluded (log base is not a class) - today's small gate roster = subject-production limit (fn-body reflection), NOT a detection limit - new §3a fully specifies the two seed rules up front: input→output→ precondition→non-firing control→discriminating equivalence input, so the worker builds to spec and the project can actually finish Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §0/§1: rust-gate is cadence-decoupling, not run-all (per fierce-hawk #5427) The 3-filter allowlist was COST selection, not arbitrary gatekeeping — the v1 SEED compiler costs ~tens of CPU-sec per trivial test, so run-all-per-PR is CPU-hours (off the table). True shape: per-PR cost-bounded subset + measured #[ignore="expensive: Ns"] + completeness lens (#5427); nightly --ignored lane as the destination for expensive + the 58 currently-ignored tests (owned by §1/quick-ant, after #5431, escalate for load-bearing CI-gen). Completeness = every test runs on >=1 cadence (fail-closed). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * plan §2a: generalize by structural-redundancy keying (O(n^x)->O(n^(x-1)) free); flag n-log-n as substitution Per operator: catalog must generalize polynomial-degree reduction without edge cases. Resolution: rules key on the structural redundancy, never on degree — degree is not evidence of redundancy (would fire on genuine O(n^x)). A structurally-keyed nested-membership->set peels one level wherever it matches; fold-to-fixpoint gives O(n^3)->O(n^2)->O(n). Cost model supports arbitrary integer degree, so witness (b) holds at every peel. Flagged OPEN (operator input invited): O(n^x)->O(n log n) is algorithmic SUBSTITUTION (different algorithms, same I/O) not redundancy elimination — verges on undecidable equivalence; tractable form is per-idiom rules (sort-based dedup, repeated-min->heap), not a parameterized rule. Seed Rule 1 now authored structurally + carries a depth-2 generalization witness. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * plan §1b: Unknown is an anemic atom — dissolve over time (reuse Disposition), never a false pass Per operator: classifying Unknown isn't a fixed up-front split — it's the standing anemic-leaf dissolution practice (DESIGN §2 decompress->map->reduce) applied to the cost lens. UnknownCost{diagnostic} already carries its reason; the anemia is the free-form reason. Each decomposition resolves an Unknown to construction (now-precise class -> new D′) or a grounded Terminal (genuinely undecidable, positively recognized -> advisory comment). DFS-first: this IS the Disposition carrier (resolves to construction-or-justified-Terminal), so reuse it, don't fork an unknown-reason enum. Supersedes the static Undecidable|Undetermined split. Two invariants fixed up front: never a false pass (Unknown=>Violates, already holds); every Unknown on the dissolution frontier. cost.dag enrichment + un-parking Disposition are operator-gated. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP: §3 reverts to budget-gate validation (stability); rewrite-engine relocated to §5 (post-stability) Operator decision 2026-06-21: budget-gate validation is fine for the stability window; the algorithmic-cost REWRITE construction design is expansion, homed with self-hosting (§5) — IR-rewrite/canonicalization is most natural once .dag is the self-hosted truth. - §3 = complexity budget gate (validation): cost-lens symbolic_max fix (#5437) + per-fn subject + budget-gates-whole-codebase (gated on fn-body reflection) + synthesis advisory. #5437 foundation stays in-window. - §5 gains an 'adjacent expansion lane' = the rewrite engine, pointing at the preserved plan doc; marked post-stability. - plan doc status -> POST-STABILITY EXPANSION, relocated to §5. Nothing deleted — the rewrite design is preserved, just fenced out of the stability window (same as Disposition / Value::Null-split). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * #5442 review fix (warm-lark-306): grep-as-authority for the sidecar roster (10 not 9) The §0-guard impl PR (#5445) grepped current main and found 10 importers of extdeps.languages.bash.program, not 9 — the 10th (dsl/gunbc/ci_spec.dag) landed via #5432 after the original pre-merge grep. Rather than bump the frozen count, make the live grep the authority (the roster shrinks to 0 as the bash-sidecar arc migrates consumers, so any frozen number rots — the single-authority point). Also note the two *_test importers are intentionally not walled (guard scans consumer-source roots only). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP: check off 6 merged items (catch-up sweep) Flip [ ]→[x] for unambiguously-merged work (PR-ref'd for traceability): - §0 numeric-tower grounding (#5428 — == straddle guard dead-in-corpus) - §0 inert-lens hygiene executable backstop (#5433) - §2 F2/F3 resolved_graph key derived from inputs_considered (#5425) - §3 cost-lens symbolic_max zero-absorption fix (#5437) - §4 gate existing generated testgen output (#5434) - §4 affected-set completeness (#5430) Partial/compound items left for their lane managers to flip in the PR that completes them. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §1: add compile-clean-gate force-checks-every-fn-body box (2(ii) fail-open) New floor-coverage item: the compile-clean gate is fail-open — unreached fn bodies escape typecheck, so undefined symbols in dead code pass green (execution-proven on utf8_decode_bytes). Construction fix = typecheck total over every declared body. Owned by §1 (quick-ant); measure-first, operator-gated enforce-flip. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * ROADMAP §0: correct the 2(ii) box — registry-leak mechanism, not unreached-bodies snappy-gull's deeper diagnosis: the fail-open is NOT unreached bodies (bodies ARE visited). utf8_decode_bytes resolves because it's a global builtin_function_registry entry (04_method.dag, a marked bridge scaffold) not scoped to the compiled tree. Reframe the box to tree-scoped builtin availability / registry partition; instance fix = real std fn + remove the registry bridge entry. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * plan doc: enforcement roster is a FROZEN grandfather set, not derived (warm-lark correction) Deriving the realization-vocab exception roster from a live grep would make the guard vacuous (leak = non-edge importer AND NOT-in-roster; derived roster ⇒ every importer always in it ⇒ leak_count always 0 ⇒ never fires). Distinguish the informational prose count (rots, re-grep) from the lens's enforcement roster (frozen, so a new unrostered importer goes RED = the teeth). Add the 11th importer (extdeps_external_authority_transport, the #5418→#5445 race, fixed by #5453) and the roster-completeness assertion as the steady-state race-hardening. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: ROADMAP planning * ROADMAP refresh: §1 nightly→Pop-B (opt-level won), §2 resolve-cache GO + P2 de-fork-dependent, §0 census regression + gate-hygiene Reflects decisions/findings that landed 2026-06-21: - §1: the "expensive" tests were debug-build amplification, not intrinsic seed cost (proud-deer cause-table); opt-level=3 (#5456) restores Pop-A to per-PR; nightly lane reduced to Pop-B wet-captures only. Mirror corrected in §0. - §2: resolve-cache enable = GO (~18% floor-wall, purity-proven, #5429-gated); P2 ParseTable dissolution reclassified as a downstream consumer of the dsl→v2 de-fork (keen-otter: v2-local rewire is cosmetic); #5446 realize kernel green. - §0: stage0 clone-census ratchet went inert + the seed regressed 1138 over budget (rust-side coverage-by-illusion + thesis regression; #5427 surfaced it); gate-hygiene rule (floor-enrolled gate must be green-on-main at merge) + roster-completeness assertion promoted to should-land (the #5445 floor-skew). - §1: registry-partition instance fix = #5452 (verified sound). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ystone):

… peak + live budget (not pinned 4) (#5444) * WIP: §1-C memory-aware width (consume the merged #5431 measurement keystone): * §1-C: derive floor spawn_width from measured peak + live budget (not pinned 4) Wire the measurement->plan loop #5431 opened: gunbc_ci_floor_spawn_width now returns min(shard_count, hardware_threads, floor(0.8*live_budget / measured_peak)) instead of hardware_thread_count(4). - gunbc.ci_floor_measurement (new): committed MEASURED per-shard peak, stamped from #5431's VmHWM emit (run 27893441091: 8400113664 bytes at width 4 => /4 = ~2.1GB/shard), NOT the ~14GB hand-grounded literal #5419 deleted. + dissolve-on drift-gate marker. - std.realization_width: memory_aware_spawn_width fold (pure, v2-inheriting) with a 0.8 safety fraction; fail-closed to a conservative committed width when budget unreadable. - claim_executor: reads the LIVE cgroup memory.max (meminfo fallback) and threads it into the .dag fold via run_in_context_with_args (DESIGN §3 measured=>peripheral: we commit our measurement, read the external-authored budget live). - witnesses: non-binding (4->7), tight-budget backoff, unreadable->fallback, fraction load-bearing, sub-unit->1. Proven by execution: derived spawn_width=7. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fmt: rustfmt claim_executor §1-C changes Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: §1-C memory-aware width (consume the merged #5431 measurement keystone): * §1-C review fix: type byte fields as std.measure ByteSize (REQUEST_CHANGES) Addresses claude-opus-4-7 REQUEST_CHANGES: the public byte-carrying fields in dsl/std/realization_width.dag (the authority layer) must consume std.measure ByteSize, not flat Int (DESIGN §2/§3 unit modeling). - memory_aware_spawn_width + memory_aware_width_value: memory_budget / per_shard_peak are now ByteSize (the canonical Memory carrier). Counts projected to Int at the boundary via byte_size_count_int (Nat subset Int, the int_min/cores widening idiom). - gunbc.ci_floor_measurement.per_shard_peak_rss_bytes returns ByteSize. - ci_floor_plan grounds the raw host Int budget into byte_size() at the host seam. - internal arithmetic helpers stay Int on extracted counts (not flagged; the placement_supply idiom). Fixes the prior WIP's Int/Nat if-branch type error. Proven by execution: realization_width_witnesses + ci_floor_plan_witnesses green; end-to-end claim_executor still derives spawn_width=7. compile --target rust clean (only the pre-existing unrelated gcp utf8_decode_bytes error remains). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: §1-C memory-aware width (consume the merged #5431 measurement keystone): * WIP: §1-C memory-aware width (consume the merged #5431 measurement keystone): * §1-C: de-anemic measured peak — typed ConcurrentMemoryPeakSample + measure-op per-share derivation Operator review: the floor's measured peak was modeled as two loose data rows (peak: ByteSize + spawn_width: Int) with the per-shard division inlined in the consumer (byte_size(count: byte_size_count(...) / int)) — anemic on two axes: - the observation is a FACT BUNDLE (a peak RSS only meaningful AT the concurrency it was sampled at: concurrent peak ≈ per-share × concurrency) held as two rows that can drift apart and neither reads as 'a measurement' (DESIGN §2/§3); - the per-share derivation was a one-off unwrap/divide/re-ground expression, not a grounded measure operation (DESIGN §4 — operations come from inhabitance). Fix, modeled on the existing std measure-algebra (time_measure_seq/_list_total in std.realization_measurement, which project via measure_count → magnitude arithmetic → reconstruct): - std.realization_measurement: add the agnostic SHAPE — type ConcurrentMemoryPeakSample { peak: ByteSize, concurrency: Int } + concurrent_memory_peak_per_share(sample) that divides the Memory magnitude by the sample concurrency and re-grounds at the same carrier; fail-closed (§5) on non-positive concurrency → whole peak (share count 1). + discriminating witness witness_concurrent_memory_peak_per_share_divides, wired into the keystone test aggregator. - gunbc.ci_floor_measurement: the measured VALUES now live as ONE typed sample (peak 8400113664 @ concurrency 4, §3 measured⇒peripheral); per-shard fn delegates to the std measure op — no inline arithmetic. Green by execution: keystone witnesses true; gunbc_ci_floor_per_shard_peak_rss_bytes() = ByteSize { count: 2100028416 } (= 8400113664 / 4, unchanged); realization_width_witnesses true. Consumer signature (ci_floor_plan) unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

WIP: ci is slow investiation

2985b84

gunbai-bot Bot changed the title ~~ci is slow investiation~~ §1 measurement keystone (instrument): emit floor concurrent-peak RSS (VmHWM) so width derives from a MEASURED peak, not a hand-grounded literal Jun 21, 2026

gunbai-bot Bot marked this pull request as ready for review June 21, 2026 04:26

briansrls merged commit cb71163 into main Jun 21, 2026
1 of 2 checks passed

briansrls deleted the session/quick-ant-298 branch June 21, 2026 05:54

gunbai-bot Bot mentioned this pull request Jun 21, 2026

§1-C: memory-aware CI floor spawn_width — derived from measured VmHWM peak + live budget (not pinned 4) #5444

Merged

briansrls added a commit that referenced this pull request Jun 21, 2026

WIP: §1-C memory-aware width (consume the merged #5431 measurement ke…

a5664df

…ystone):

briansrls added a commit that referenced this pull request Jun 21, 2026

WIP: §1-C memory-aware width (consume the merged #5431 measurement ke…

4cbbd69

…ystone):

briansrls added a commit that referenced this pull request Jun 21, 2026

WIP: §1-C memory-aware width (consume the merged #5431 measurement ke…

4b10dbd

…ystone):

Navigation Menu

Repository navigation

§1 measurement keystone (instrument): emit floor concurrent-peak RSS (VmHWM) so width derives from a MEASURED peak, not a hand-grounded literal - #5431

§1 measurement keystone (instrument): emit floor concurrent-peak RSS (VmHWM) so width derives from a MEASURED peak, not a hand-grounded literal#5431
briansrls merged 1 commit into
mainfrom
session/quick-ant-298

gunbai-bot Bot commented Jun 21, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

gunbai-bot Bot commented Jun 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What

Why (construction, not a hand-authored literal)

Scope / non-goals

Verification

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

gunbai-bot Bot commented Jun 21, 2026 •

edited

Loading