docs(benchmarks): refresh post-#699 baseline and update the Claude git rules

[tylerkron]

What was wrong

• The benchmark README still presented the 2a59fd1 run as the current baseline, and ended with the protobuf first-sample outlier "filed as perf(sdcard): the protobuf log parser spends 1.6ms and 6.7MB before its first sample, while CSV and JSON take 2us #697 rather than fixed here". perf(sdcard): stop the protobuf log parser decoding a whole 64 KB buffer before its first sample #699 fixed perf(sdcard): the protobuf log parser spends 1.6ms and 6.7MB before its first sample, while CSV and JSON take 2us #697 weeks ago.
• StreamDecodeBenchmarks' remarks said a regression would show as allocation "climbing off zero", but the baseline in the same README is already ~1.38 KB per frame.
• .claude/rules/git-workflow.md (loaded into every Claude Code session in this repo) still said main/master, used generic git add . recipes, and said nothing about how main is actually gated or how commits are titled.

How it was fixed

• README: the 2a59fd1 table is labelled before perf(sdcard): stop the protobuf log parser decoding a whole 64 KB buffer before its first sample #699, and the "filed, not fixed" section is replaced by perf(sdcard): stop the protobuf log parser decoding a whole 64 KB buffer before its first sample #699's before/after figures (protobuf time-to-first-sample 1.6 ms / 6.7 MB → 9.4 µs / 263 KB). These numbers are copied from perf(sdcard): stop the protobuf log parser decoding a whole 64 KB buffer before its first sample #699's PR description, not re-measured here, and the text now says so.
• Benchmark remarks: the ~1.38 KB baseline is attributed to each channel's DataSample plus the SampleReceivedEventArgs raised for the subscriber.
• Git rules: updated, not deleted. They keep the never-push-main guardrail and now state what the repo actually does: merge-queue-gated, squash-only main; branch from a fresh origin/main; conventional-commit type(scope): summary titles; PR descriptions that lead with the problem; no git stash (the stash stack is shared across worktrees). The master mentions and generic recipes are gone.

Docs and comments only; no production code.

[qodo-code-review]

PR Summary by Qodo

Refresh post-#699 benchmarks and centralize git workflow guidance

📝 Documentation ⚙️ Configuration changes 🕐 Less than 10 minutes

AI Description

• Records #699’s protobuf latency and allocation improvements beside preserved pre-fix baselines.
• Corrects stream decode guidance to acknowledge unavoidable subscriber event allocations.
• Replaces stale Claude git instructions with canonical contribution guidance.

Diagram

graph TD
  A["Claude agents"] --> B["Git rule stub"] --> C["CONTRIBUTING.md"]
  D["PR #699 results"] --> E["Benchmark README"]
  F["Stream benchmark"] -->|"documents allocation floor"| E

Loading

High-Level Assessment

The approach is appropriate: preserving the original baseline while adding same-machine #699 results maintains historical comparability, and replacing duplicated Claude instructions with a short CONTRIBUTING.md pointer prevents workflow drift. Rerunning benchmarks elsewhere would produce less comparable timing data, while deleting the Claude rule entirely would reduce discoverability.

Files changed (3) +22 / -80

Documentation (2) +19 / -12

README.mdDocument protobuf benchmark improvements from #699 +17/-11

Document protobuf benchmark improvements from #699

• Relabels the existing '2a59fd1' results as the pre-#699 baseline and adds same-machine before-and-after measurements from #699. It explains the first-sample latency, full-drain, and allocation improvements without referring to #697 as still open.

src/Daqifi.Core.Benchmarks/README.md

StreamDecodeBenchmarks.csClarify the stream decode allocation baseline +2/-1

Clarify the stream decode allocation baseline

• Updates benchmark remarks to explain that regressions appear as allocation growth from the existing approximately 1.38 KB baseline, rather than growth from zero. The baseline comes from 'SampleReceived' event arguments paid by real subscribers.

src/Daqifi.Core.Benchmarks/StreamDecodeBenchmarks.cs

Other (1) +3 / -68

git-workflow.mdDelegate Claude git workflow rules to CONTRIBUTING.md +3/-68

Delegate Claude git workflow rules to CONTRIBUTING.md

• Replaces duplicated branch, commit, and pull-request recipes with a concise pointer to the repository’s canonical contribution policy. It explicitly prohibits pushing 'main' and inventing a separate commit-message convention.

.claude/rules/git-workflow.md

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📜 Skill insights (0)

1. Allocation source is misstated ✓ Resolved 🐞 Bug ⚙ Maintainability

Description

The StreamDecodeBenchmarks remarks attribute the full ~1.38 KB per-frame allocation to
SampleReceivedEventArgs, although decoding also creates one DataSample object per enabled
channel. This misdirects future allocation investigations because the 16 sample objects and 16
event-argument objects both contribute to the reported baseline.

Code

src/Daqifi.Core.Benchmarks/StreamDecodeBenchmarks.cs[R27-28]

+/// as allocation climbing, not as a departure from zero — the table is already ~1.38 KB from the
+/// SampleReceived args a real subscriber pays for.

Relevance

●●● Strong

Recent benchmark review accepted correcting misleading allocation attribution in documentation.

PR-#706
PR-#698

ⓘ Recommendations generated based on similar findings in past PRs

Evidence

The benchmark subscribes to SampleReceived for every enabled channel, while the decoder explicitly
constructs a DataSample for each decoded channel and SetActiveSample separately constructs
SampleReceivedEventArgs when invoking the subscriber. Therefore, the measured per-frame allocation
cannot be attributed solely to the event arguments.

src/Daqifi.Core.Benchmarks/StreamDecodeBenchmarks.cs[103-110]
src/Daqifi.Core/Device/Internal/StreamFrameDecoder.cs[595-596]
src/Daqifi.Core/Channel/AnalogChannel.cs[342-344]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The benchmark remarks incorrectly say the full ~1.38 KB baseline comes from `SampleReceivedEventArgs`, while each decoded channel also allocates a `DataSample`.

## Fix Focus Areas
- src/Daqifi.Core.Benchmarks/StreamDecodeBenchmarks.cs[27-28]

## Recommended Fix
Revise the remarks to say the baseline includes both each channel's decoded `DataSample` and the `SampleReceivedEventArgs` created for the benchmark subscriber, without attributing the entire measured amount to either object type.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

Context sources

Review mode: ⏭️ Skipped: The push only changes documentation and benchmark comments, with no runtime, configuration, test, or build behavior changes.

Tip of the day

💡 Did you know, you can add REVIEW.md to your repo root and Qodo follows it on every PR

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

[tylerkron]

/agentic_review

[qodo-code-review]

Code review by qodo was updated up to the latest commit 9a61dd1

[tylerkron]

Reviewed (Claude): partial - the benchmark refresh is right and its numbers are #699's real measurements (now labelled as copied, not re-run); the Claude workflow rules were mostly correct live guardrails, so they're updated to the actual merge-queue/squash/conventional-commit process instead of being cut to a two-line stub. Qodo-clean on 9a61dd1, CI green — ready for review.