perf: publish U5 FPGA loopback throughput evidence

[ichigo-6301]

Summary

• publish one bounded XC7Z100 synchronous PL-loopback board observation
• preserve the exact operator-transcribed debugger fields and recompute all rates with Python Decimal
• publish the SDK source only as a reproduction reference, not as retained source-to-binary build provenance
• register a partial, public, non-resume-eligible FPGA Claim under the base-owned policy merged through PR policy: tighten FPGA evidence publication boundaries #16
• leave RTL, profiles, ASIC Evidence, Async64 simulation Evidence, and frozen release content unchanged

Observation

• profile: 13 RX / 13 TX, TX0 to RX0, synchronous 512-bit PL-local loopback
• memory boundary: existing 64-bit HP0 at 100 MHz
• workload: 1024 x 4096-byte frames
• payload: 4,194,304 bytes
• timer: 8,969,535 XTime ticks at COUNTS_PER_SECOND=333333343
• rounded equivalent PL cycles: 2,690,860
• post-start-write-return completion-window rate: 1.558722 MB/s/MHz
• at 100 MHz: 155.872225 MB/s / 1.246978 Gb/s
• relative to the conservative 4 B/cycle shared-HP0 model: 38.968056%

Evidence Boundary

Classification: FPGA_DEBUGGER_TRANSCRIBED_SINGLE_RUN.

The raw debugger fields were transcribed after the timing and correctness gates. No independent screenshot, debugger session export, or memory dump was retained. The public helloworld.c is a fixed source-only reproduction reference; the retained private ELF and bitstream are identified by size and SHA-256, but no retained build manifest cryptographically links those binaries to the public source.

The timer starts after the descriptor-start register write returns. The metric is therefore a post-start-write-return completion-window payload rate, not launch-to-completion hardware end-to-end throughput.

Review Fixes

• add and hash debugger_capture_transcript.txt
• downgrade capture wording from debugger-captured to operator-transcribed
• remove the unretained source-to-ELF/bitstream build-chain assertion
• compute model efficiency from the unrounded tick rational
• authorize only the fixed source-only fpga/u5/benchmark exception
• rename and bind the measured timing window honestly

Validation

• local/WSL make showcase-check: 337 tests passed, 1 expected skip
• FPGA_EMULATION_EVIDENCE_PASS
• browser render: 6 assets, 256 text nodes, 37 regions, 4 homepage cases
• public hygiene: 346 checksums, 14 README claims, 8 showcase assets
• ASIC Evidence and Async64 publication gates: pass
• frozen v0.1.0-rc1: unchanged and pass
• checksum and git diff --check: pass
• final GitHub tree equals locally validated tree 7d20a9e996d246dcff43a0397308991f02431a86

Nonclaims

• automated UART capture: no
• repeated-run statistics: no
• steady-state first-frame-excluded result: no
• Async64 CDC board test: no
• Aurora performance: no
• DDR peak or FPGA Fmax: no
• 64 B/cycle Writer result: no
• ASIC result: no
• resume eligible: no

Scope

• production RTL changed: no
• profile/config/filelist/constraint changed: no
• existing ASIC/Async64 Claim numeric changed: no
• release tag changed: no
• PR merged: no

[ichigo-6301]

@codex review

Please review the bounded U5 FPGA evidence publication at 8dbaf7672e8453a664ae5aa35e5dfeac62f37faa. Focus on counter derivation, artifact/source binding, claim boundaries, sensitive-data hygiene, and base-owned scope enforcement.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8dbaf7672e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[ichigo-6301]

@codex review

Please review corrected head 60c9af361717bf6f95e253e6451c98e21ddb5f92 against merged base policy 8a0ac6be0482ddce84fd4e56a0bb51fbc7917b19. Focus on the operator-transcribed single-run classification, absence of retained source-to-binary build provenance, direct unrounded Decimal efficiency derivation, exact source-only public-scope exception, and the post-start-write-return timing boundary.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🚀

Reviewed commit: 60c9af3617

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".