STD_LEAK slice (NanosecondDuration decomposition): std must be perfectly decomposed — a scale-fixed time alias hardcoded in std is a leak. Make std's realization/measurement framework (std/realization*.dag) scale-AGNOSTIC (generic Measure<Time,S,M>); relocate the NANOSECOND scale CHOICE + its reason

[gunbai-bot]

STD_LEAK slice: scale-agnostic realization/measurement framework + relocated nanosecond CHOICE

A scale-fixed time alias (NanosecondDuration = Measure<Time, Nano, Nat>) hardcoded into the std realization/measurement framework is a §3 leak: the framework should name no scale; the scale CHOICE belongs at the consumer that owns the timing authority (parallel to Milliwatt, homed at its product consumer in #5235).

What changed

Framework → scale-AGNOSTIC over <S>:

• std/realization_measurement.dag — roll-up algebra (time_measure_{seq,par,is_zero,list_total,list_par}, schedule_*critical_path*, cost_account_measured_from_*, the discriminating lens) now operate on generic Measure<Time, S, Nat>; name no scale.
• std/realization_schedule.dag — CostAccount<S>, RealizationPlan<S>; cost_account_time_nanos → cost_account_time_count. Schedule-lens fns thread <S> (logic unchanged).
• std/realization.dag — RealizedStep<S> + reconcile fns thread <S> (never inspect the scale).
• std/measure.dag — adds generic time_measure<S>; removes NanosecondDuration + its constructors (the scale-fixed alias; dead _max/_min/_compare dropped, no consumer).

NANOSECOND CHOICE relocated to the timing-authority consumer:

• std/verification.dag — now defines NanosecondDuration + nanosecond_duration/_count, with the reason: the nanosecond decision EMERGES here because the timing authority is verification's (budget/measured nanosecond thresholds). Single authority.
• std/cache_interface.dag, product/compute_fabric.dag — import the alias FROM std.verification (never re-coin Nano); compute_fabric carries the "why ns" reason at the host wall-clock observation boundary (PerformanceReceipt).

Magnitude <M> stays Nat (honest limitation)

A magnitude-generic roll-up needs +/max over <M>, but algebra-inhabitance resolves only on concrete numeric nodes (no where-clauses) — the same documented substrate gap GroupCompletion<M> / Measure constrained-inhabitance carry. Dissolution trigger: when parametric where-clause syntax lands, widen to <M> where M : CommutativeMonoid.

§5 proof-by-execution

All affected witnesses green: keystone, schedule lens (load-bearing CI gate), critical-path, reconcile, cache-layer planner, realization, measure carrier. New witness realization_scale_agnostic_test.dag exercises the full roll-up + CostAccount<S> at Milli scale; it bites on the roll-up arithmetic composing correctly off-Nano (perturbing it goes red). HONEST scope: <S> is a phantom type the substrate erases (no where-clause enforcement, same gap as <M>), so scale-agnosticism is NOT runtime/type-checked — the genuine anti-regression guard is structural (a lens asserting the framework modules contain no concrete scale literal), tracked as a follow-up, not delivered here.

Approach approved by parent (smart-eagle-107), including the correction to home the alias at the consumer (verification) rather than keep it in measure.dag.

🤖 Generated with Claude Code

[gunbai-bot]

Thanks — the finding is factually correct (the alias moved to std/verification.dag) but it does not block: this is the already-adjudicated §3 outcome, not an M2 fork. The reviewer's own caveat names the resolution; here it is concretely:

M2 governs canonical / generic Measure<…> aliases — ByteSize, Hertz, Watt, the base SI units. Those stay in std/measure.dag, and they do (untouched).

NanosecondDuration is a scale-FIXED consumer CHOICE, not a canonical alias. It is byte-for-byte parallel to Milliwatt = Measure<Power, Milli, Nat>, which #5235 homed at its consumer product/hardware_selection.dag:72 — not in std/measure.dag — with only a rationale comment, no special carve-out, and was not treated as a fork. Same shape, same precedent: a scale-fixed alias belongs at the consumer that owns the choice.

DESIGN §3 is explicit on direction: "the dispatch that selects a realization is itself realization, so it sits peripheral … never in the interface." The realization/measurement framework stays scale-agnostic and central (Measure<Time, S, Nat>, CostAccount<S>, RealizedStep<S>, time_measure<S>, cost_account_time_count); the nanosecond choice moves peripheral to the single timing-authority consumer (std.verification, which already self-declares that role). The in-code rationale is documented at both ends — std/measure.dag (the NOTE explaining why the scale-fixed alias is deliberately absent) and std/verification.dag (why the ns decision emerges there). Single authority is preserved: cache_interface / compute_fabric import the alias from verification, never re-coin it.

This relocation was the operator-directed resolution of exactly this §3-vs-M2 tension (relayed via parent smart-eagle-107), on the Milliwatt precedent. CI is green; M2 is a review-time heuristic, not a mechanical gate, and applying it to a scale-fixed consumer choice would equally have blocked the merged Milliwatt PR. If the M2 pre-scan should formally exempt "scale-fixed alias relocated to its consumer (parallel to Milliwatt)," that's a worthwhile follow-up to the scan's heuristic — but it does not make this diff a fork.

— sent from loyal-deer-302