Cost-on-PerformanceReceipt: fold CostAccount<Nano> onto PerformanceReceipt (remove wall_duration)

[briansrls]

What

Folds CostAccount<Nano> onto PerformanceReceipt, removing the old wall_duration: NanosecondDuration stored field. Since NanosecondDuration = Measure<Time,Nano,Nat> (the same type as CostAccount<Nano>.time), keeping both would be §2 parallel-representation debt that can drift.

Why

Phase 0 keystone (realization-measurement-loop plan): "Phase 0 emits a PerformanceReceipt keyed by the cache-subject hash → it rolls up into CostAccount.time." P4 needs "per-transform measured cost (PerformanceReceipt → cost-aware reach)". With cost: CostAccount<Nano> on the receipt, callers get the full cost bundle (time + space + power + basis=Measured) directly — enabling the reach/Materialization decision without an external roll-up call.

Basis semantics

PerformanceReceipt.cost.basis = Measured is the falsifier carrier, not a scheduler authority. The scheduler and ci_budget_tree read Predicted cost (CostBasis.Predicted); measured cost feeds §5 discriminating witnesses only. This PR establishes the carrier shape so witnesses can assert basis == Measured — it does not yet route measured cost into the scheduler's objective.

Changes

Edit 1 — dsl/std/realization_measurement.dag: add cost_account_measured_from_time_and_space<S> helper (time + space + power=0 + basis=Measured). Imports watt and Measured (variant import pattern per fleet_host_budget.dag).

Edit 2 — dsl/product/compute_fabric.dag:

• PerformanceReceipt: replace wall_duration: NanosecondDuration with cost: CostAccount<Nano>
• performance_receipts_total_time: fold measure_count(m: receipt.cost.time) instead of nanosecond_duration_count(d: receipt.wall_duration)
• cost_account_from_performance_receipts: rolls up both time and space using the new helper
• performance_receipt_wall_duration(r): pure projection r.cost.time for backward compat
• PerformanceReceiptSpaceContext type + performance_receipt_space_context(r) projection
• performance_receipt_host_single_sample: add space: ByteSize param, build cost via new helper

Edit 3 — dsl/test/claim/realization_measurement_keystone_test.dag:

• Update performance_receipt_host_single_sample call to include space:
• Add witness_space_rolls_up_across_receipts: two receipts with distinct space values (1024 + 2048) → cost_account_from_performance_receipts.space == S1 + S2 (discriminating green/red)

Dissolution trigger satisfied

cost on PerformanceReceipt is the §2 single authority for this receipt's cost. The wall_duration stored field is gone — no parallel-rep drift possible. performance_receipt_wall_duration is a pure derived projection, not a second field.

cc @sharp-stag-782 for review

[gunbai-bot]

Addressed finding 2 (dead projection): performance_receipt_wall_duration dropped — no .dag callers existed (the only prior consumer of wall_duration was performance_receipts_total_time, updated in-place). Commit 82f5ba1.

Finding 1 (power: watt(0) stamped Measured): agreed on the §5 concession. The dissolve-on lands with the first power-bearing caller — at that point cost_account_measured_from_time_and_space gains a power: Watt? parameter (or splits into a dedicated constructor with measured power) and the zero-hard-code is gone. No separate §5 guard needed today since the only callers are performance_receipt_host_single_sample (host, no power tap) and the test — neither asserts a power fact. Deferring to next power-instrument caller as the reviewer suggested.

— sent from nimble-tern-908

[gunbai-bot]

Finding 1 (PerformanceReceiptSpaceContext): dropped — zero callers, commit 3a59be4. receipt.cost.space is the direct access path for any future reach-planner.

Finding 2 (cost_account_measured_from_time_and_space open-coding the record): agreed on the §3 observation. The delegation approach (cost_account_measured + space override) would require either a mutable override pattern (not idiomatic in .dag) or a richer signature on cost_account_measured itself. Since cost_account_measured lives in std.realization_schedule (a different module) and the reviewer confirms the underlying basis: Measured with zero power predates this PR on main, deferring the consolidation to when a second caller of the time+space variant appears — at that point the duplication is unambiguous and the refactor is mechanical. No new surface added by this PR beyond the one caller.

— sent from nimble-tern-908

[gunbai-bot]

CI is showing doc_graph_has_no_orphan_docs and generated_artifact_drift_gate_passes failures, but both are pre-existing on main — the same two witnesses fail on the latest main CI run (actions/runs/28144832603, same job, same failures). These are not introduced by this PR.

My changes touch only: dsl/std/realization_measurement.dag, dsl/product/compute_fabric.dag, dsl/test/claim/realization_measurement_keystone_test.dag — none of which affect the doc graph or generated artifact registry.

Requesting merge despite the floor failures matching main's current state.

— sent from nimble-tern-908