fix(replication): strengthen audit cadence and penalties

[mickvandijke]

Summary:

• Run replication storage audits every 10-20 minutes instead of 30-60 minutes.
• Raise confirmed audit failure trust penalty weight from 2.0 to 5.0.
• Document the audit penalty weight in the replication reference profile.

Semver: patch

Tests:

• cargo test replication::config --lib

[Copilot]

Pull request overview

This PR tightens replication storage audit behavior by increasing audit frequency and strengthening the trust penalty applied to confirmed audit failures, while updating the replication design reference profile accordingly.

Changes:

• Reduce audit scheduler jitter window from 30–60 minutes to 10–20 minutes in the replication reference config.
• Increase AUDIT_FAILURE_TRUST_WEIGHT from 2.0 to 5.0 and add/adjust unit tests to reflect the new reference profile.
• Document the audit failure trust penalty weight in REPLICATION_DESIGN.md.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/replication/config.rs	Updates audit tick interval bounds and the confirmed audit failure trust penalty weight; adjusts/adds tests for the new reference values.
docs/REPLICATION_DESIGN.md	Updates the reference profile table to reflect the new audit cadence and documents AUDIT_FAILURE_TRUST_WEIGHT.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

[dirvine]

Review: fix(replication): strengthen audit cadence and penalties

Overall Verdict

Changes are sensible and well-scoped — tighter audit cadence and stronger penalties should improve data integrity. A few notes below.

Analysis

1. Audit frequency 30-60min → 10-20min (3-6x increase)
Safe given the audit workload scales as floor(sqrt(local_key_count)). For a node with 10K keys, that is ~100 keys per round. At worst cadence (10min) that is 1 key checked every 6 seconds — negligible overhead.

2. AUDIT_FAILURE_TRUST_WEIGHT 2.0 → 5.0 (2.5x increase)
This is the most impactful change. The weight amplifies the trust penalty per confirmed failure. Worth verifying that the eviction/penalty threshold is proportionally tuned — otherwise a small number of transient network failures could trigger premature distrust. Recommend documenting the threshold relationship.

3. New test: audit_failure_weight_matches_max_consumer_trust_event_weight
The f64 bit-pattern comparison via to_bits() is fragile — it passes only for the exact binary representation of 5.0. While 5.0 is exactly representable in binary64, this pattern is brittle to future constant changes. Consider using a semantic comparison or the float-cmp/approx crate pattern if similar tests exist in the codebase. Not a blocker.

4. Test scenario_31_audit_cadence_within_jitter_bounds
Adequately validates jitter produces varying intervals. With 100 samples across a 10-minute range there is effectively zero chance of all being identical. Good.

5. macOS CI: Test (macos-latest) FAILURE
Root cause is Install Foundry step failure (infrastructure/network issue on the macOS runner) — not a code failure. Unit tests were skipped as a result. Not a PR concern.

Summary

Aspect	Status
Code correctness	✅
Test coverage	✅ (minor nit on f64 comparison pattern)
Docs updated	✅
Build (all platforms)	✅
Format/Lint	✅
Security audit	✅
CI (macOS)	⚠️ Infrastructure failure only (not code)

The PR is clean and well-targeted. Nice work @mickvandijke.

[dirvine]

Reviewed via Hermes. Code is clean, well-scoped, all builds/lint/security pass. Test coverage adequate. Changes are sound.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.