fix(latency): a declared range budget is read, under its own property set, and a breach fails the run

[avrabe]

#455 item (2), partially — REQ-ANALYSIS-LATENCY-BUDGET-001 goes to implemented, not verified, because two of its clauses didn't ship. Said first so this isn't read as closing the item.

The comparison already existed — four defects kept it from firing

	defect	this PR
(a)	get_timing_property is scalar-only; the standard types Latency as Time_Range, so #455's own 300 ns .. 250 us read as no budget	fixed
(d)	wrong property set — Latency is Communication_Properties; the lookup searched Timing_Properties then "", so the correctly qualified spelling was the one never found	fixed (found while fixing (a))
(c)	a Warning — even the one firing combination exited 0	fixed — now an Error
(b)	reads the owner component, never the flow's own { Latency => … }	not here — needs a HIR change

End to end, through the real CLI

Published v0.45.0 against this branch; two devices summing 80 ms against a 1 ms budget:

model	v0.45.0	this PR
range, bare Latency	0	1 [ERROR] … exceeds bound
range, Communication_Properties::Latency	0	1
scalar — previously only a Warning	0	1
range budget that is met	0	0
no budget	0	0

⚠️ Behaviour change

A breach is now an Error, so models that are over budget today — and were told so only at Warning — will start failing. That is the intent: a budget nobody can gate on is the defect REQ-CLI-ANALYZE-EXITCODE-001 fixed one layer down. Needs a release-notes line.

Deliberately NOT shipped: the below-minimum check

best_case_ps is not a best case. It sums get_execution_time, whose doc says "worst-case from range" — which is why two 20 ms .. 40 ms devices report [80.000 ms .. 80.000 ms] rather than 40 .. 80.

A minimum check over that figure compares an inflated lower bound, so it stays silent on real violations whenever the inflated sum clears the minimum — unsound in exactly the direction #455's kill-criterion needs. Filed as REQ-ANALYSIS-LATENCY-BEST-CASE-001 (the provenance class again: a printed figure labelled as something it isn't). A test pins that no below-minimum diagnostic ships until it's fixed.

Relatedly, a scalar budget is read as an upper bound only, never min == max — otherwise the moment a minimum check exists, every within-budget flow breaches it.

Two existing tests went vacuous — demonstrated, not assumed

latency_exactly_at_bound_no_warning and latency_within_bound_no_warning filtered on Severity::Warning and asserted empty — so once the breach became an Error, a spurious Error passed them. Proof: > → >= survived the at-bound test before the fix, and is caught by it after. Negative tests now match at any severity; positive ones require Error.

Mutation evidence

Each mutant confirmed surviving the suite before its case was written:

mutant	killed by
range string path returns None (the original bug)	only a_range_budget_is_read_and_a_breach_is_an_error
Communication_Properties dropped	only a_budget_under_its_standard_property_set_is_found
range max reads the min	the met/breached range pair
severity back to Warning	the three positive tests
> → >=	only the at-bound test, after its fix
scalar sets min == max	survived the whole analysis suite (nothing consumes min_ps yet) — pinned at the accessor, caught by exactly one

cargo mutants --in-diff locally: 10 mutants, 7 caught, 3 unviable, 0 missed, end_time populated. The block this replaces carried no survivors in main's last full run (38500597645), so the full-crate ratchet is expected to hold at 292 — a prediction backed by that artifact, not by reasoning alone.

Verified

cargo test --workspace exit 0 — 3238 passed across 102 result lines, zero failures · spar-analysis 941 · clippy -D warnings clean · fmt clean · workspace builds all targets · rivet validate PASS.

Refs #455

🤖 Generated with Claude Code

https://claude.ai/code/session_01BnBWxgVEUPdsBTLNKtrA6Z

[github-actions]

Rivet verification gate

✅ 20/20 passed

	count
Passed	20
Failed	0
Skipped (no steps)	0

Filter: (and (= type "feature") (or (has-tag "v093") (has-tag "v0100")))

Failed artifacts

(none)

_{Updated automatically by tools/post_verification_comment.py. Source of truth: artifacts/verification.yaml.}

[codecov]

Codecov Report

❌ Patch coverage is 87.30159% with 16 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/spar-analysis/src/property_accessors.rs	72.72%	15 Missing ⚠️
crates/spar-analysis/src/latency.rs	98.59%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!