feat(bench): RFC 0031 — latency_p50 channel (RFC0031.7 becomes measurable)

[jensholdgaard]

What

The latency_p50 channel RFC 0031 §3.6 specifies but the harness never implemented. After this, scenario RFC0031.7 (ourios.latency_p50 ≤ F_L6 × loki.latency_p50) is measurable as written — and measured.

Measured result (pre-merge validation, run #18 on this branch — first latency numbers in the program)

pair	Ourios p50	Loki p50	ratio (>1 = Ourios faster)
severity (1 row)	82.0 ms	875.0 ms	10.7×
L3 trace (9 rows)	74.6 ms	24.10 s	323×
L1 template (2 rows)	75.7 ms	23.32 s	308×
window k=100	40.2 ms	13.8 ms	0.34 — RFC0031.7 floor (3) PASSES
window k=2000	85.9 ms	294.8 ms	3.4× — Ourios faster; floor passes
selective-resource k=100	38.8 ms	51.2 ms	1.3× — Ourios faster; floor passes

Three headline findings:

1. The "young engine" latency risk the RFC hedged against did not materialize. Ourios answers every pair in 39–86 ms — a remarkably flat profile — while Loki spans 14 ms to 24 s. On the needle classes the human-facing difference is interactive-vs-coffee-break: 75 ms vs 23–24 seconds.
2. The L6 floor, evaluated as the RFC actually wrote it (latency), PASSES on all three window pairs (0.34 / 3.43 / 1.32, all ≥ 1/3) — and Ourios is outright faster on two of the three windows it loses on bytes. The §9.13 draft's "published loss" is a bytes-channel artifact; on the RFC's own gate the window tier holds the floor.
3. The asymmetry caveat (in-process vs localhost HTTP) is immaterial at these magnitudes.

All byte figures and equivalence unchanged from run #17 (exit 0).

Checks run

cargo fmt --all --check, cargo clippy --all-targets --all-features -- -D warnings (workspace-wide), cargo nextest run -p ourios-bench (140 passed). Pre-merge bench: run #18 (29210202343), exit 0, six pairs, equivalence held.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WQY9wfrfRggqSpMLH8Xj3Y

[coderabbitai]

📝 Walkthrough

Walkthrough

The RFC0031 comparative harness now measures warm p50 latency for Loki and Ourios queries, propagates latency through measurement handling, reports latency ratios and Floor gate results, and tests median and gate calculations.

Changes

Warm latency comparison

Layer / File(s)	Summary
Latency measurement and gates crates/ourios-bench/tests/rfc0031_comparative.rs	Adds repeated Loki and Ourios query timing, median Duration calculation, and nanosecond-based latency Floor gating.
Measurement collection and propagation crates/ourios-bench/tests/rfc0031_comparative.rs	Carries optional latency p50 values through Ourios and Loki measurement vectors and split_measurements.
Latency reporting and tests crates/ourios-bench/tests/rfc0031_comparative.rs	Reports warm latency values, Loki/Ourios ratios, and Floor outcomes, with unit tests for median and gate behavior.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
  participant ComparativeRunner
  participant Ourios
  participant Loki
  participant IndicativeReport
  ComparativeRunner->>Ourios: Execute repeated in-process queries
  Ourios-->>ComparativeRunner: Return latency samples
  ComparativeRunner->>Loki: Execute repeated query_range requests
  Loki-->>ComparativeRunner: Return latency samples
  ComparativeRunner->>ComparativeRunner: Store and split measurements
  ComparativeRunner->>IndicativeReport: Render p50 values, ratio, and Floor result

Loading

Possibly related PRs

• jensholdgaard/ourios#467: Introduces the comparative core types and equivalence functions extended by this latency work.
• jensholdgaard/ourios#474: Establishes the comparative workflow and bytes-read reporting extended here with warm latency reporting.
• jensholdgaard/ourios#481: Adds Floor gate behavior for bytes, which this change parallels for latency.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	It covers the change and validation, but it doesn't follow the template: Summary, Related, and Checklist sections are missing or incomplete.	Add the required Summary and Related sections, and include the Checklist with checkbox items for fmt, clippy, tests, docs/CHANGELOG, and RFC link if applicable.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	It matches the main change: adding the RFC 0031 latency_p50 channel and making RFC0031.7 measurable.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch rfc0031-latency-channel

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Adds the missing RFC 0031 §3.6 latency_p50 “corroborating channel” to the comparative benchmark harness so scenario RFC0031.7’s latency floor can be measured and reported alongside the existing bytes gates.

Changes:

• Introduces warm-p50 latency measurement for both Ourios (in-process) and Loki (localhost HTTP), using 7 post-correctness repetitions and reporting the median.
• Implements an RFC0031.7 latency floor check by reusing ourios_bench::bytes_within_floor over nanosecond integers.
• Extends the per-pair measurement plumbing/report output and adds unit tests for median + floor behavior.

---

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

[Copilot]

Pull request overview

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

[Copilot]

Pull request overview

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

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.