fix(benchmarks): bound lifecycle peak envelope without derivative false positives

[DecisionNerd]

Verified independent blocker

PR #1271's complete CI run 34869384852 passed Rust, platform bindings, durability and concurrency tests. Its real tiny/ownership-growth producer completed all lifecycle phases but rejected aggregate temporary-storage growth: positive_slopes requires adjacent derivatives of the lifecycle maximum to differ by no more than factor two.

The producer defines lifecycle peak as the maximum across independently validated operation peaks. A maximum of linear operation envelopes can change its dominant operation and violate that derivative test without superlinear total growth. Candidate SCALE6/7/8 peaks are 1,093,632 / 1,343,488 / 2,449,408 bytes for work 1,088 / 2,176 / 4,352. Adjacent slopes differ by 2.213, but peak/work decreases 1005.18 → 617.41 → 562.82. Baseline successful CI 34864684237 peaks were 1,101,824 / 1,515,520 / 2,842,624: the candidate reduces every peak. This does not identify the exact winning operation; do not invent phase attribution.

Scope

Correct the aggregate-maximum growth invariant in the existing tiny lifecycle gate. Keep exact receipt/raw-owner reconciliation, all owner EOF derivative checks, retained-allocation checks, positive aggregate growth, fixed absent/lock owners and current factor-two ceilings. Do not raise a numeric tolerance, skip a check, alter allocation accounting, pad output to fit slopes, add retries, change the workload, or introduce a new certification workflow.

For this fixed 1x/2x/4x workload, require adjacent aggregate peak/work to be nonincreasing: P_next * W_previous <= P_previous * W_next, alongside strict positive peak growth. This is a tighter upper-growth bound than the previous factor-two normalized ceiling and admits the maximum of affine operation envelopes with nonnegative fixed overhead. A changing derivative of that maximum is not itself a storage regression.

Acceptance criteria

• Add a regression demonstrating the old derivative check rejects a legitimate maximum of two affine operation envelopes and the corrected gate admits it.
• Admit the preserved real perf(storage): retire consumed roots before shaping merge peak #1271 observations while retaining exact owner, ratio, row, schema and positive-growth checks.
• Add a case rejected by the new adjacent normalized bound despite fitting the old factor-two ceiling. Keep flat, inflated, missing and malformed evidence rejection tests.
• Preserve full failed CI evidence and baseline comparison; do not claim the failed run passed.
• Pass targeted harness checks and required exact-head CI Gate; squash merge and verify closure. Then refresh perf(storage): retire consumed roots before shaping merge peak #1271 against current main and rerun its complete gate once.

Focused native blocker of #1268 and canonical #1194. No implementation scope expansion or capacity certification is added.

Verified completion: #1273 squash-merged as 880c9a7acbe9f648ade9ce60c08a0057c93e9cfa after exact-head CI34872510638 and CI Gate passed. All400 local harness tests passed on stable head078469fb; independent review found no blocker. #1271 was refreshed onto current main with its Rust diff unchanged.

[DecisionNerd]

Evidence relocated from the docs tree (#1625). The 1 file(s) below were committed under docs/development/ as measurement evidence for this issue. They are removed from the tree by #1628 (the #1625 pull request); the repository keeps method and content digests, results live here. Every file remains available at its permalink on commit 29a7b34ebe44, and its SHA-256 is recorded next to it.

Files (1 total, 17 lines)

• docs/development/evidence/lifecycle-peak-envelope-1272.md sha256:a8d1e44b594ed6163c4031ef8e548a1666258fb75b39dac22f6d0f35d6118240