Conversation
secup
added a commit
that referenced
this pull request
May 3, 2026
Tonight's autonomous code-quality pass on the new TNC + HARQ work, informed by Codex review (full notes /tmp/codex_quality_findings.md, 22 ranked findings). Tackled the safe ones with high impact: Safety / correctness (#3, #7-#8, #14): - ultra_tnc: skip config loading on --help / --list-audio-devices so a malformed config doesn't break operator recovery - tnc_session: cmdStats() now null-guards cmd_emit_, matching the rest of the emit path (was the one place that could throw std::bad_function_call on a null callback) - tnc_bridge: postPTT() callbacks no longer fire under ptt_mutex_. Computes the PTT transition under the lock, releases, then invokes sink + hardware callback. Closes a deadlock window when a slow hardware-PTT serial write blocks under the mutex. - (#7 PTT setLine() return-check deferred — needs decision on whether to fail-startup or fall-back-to-VOX. Punted to follow-up.) Tests (#12, #19-#20): - 1 new test_tnc_session: corrupt-deflate RX silently dropped (regression guard for the marker-strip path) - 6 new test_soft_combine cases: * setEnabled(false) clears retained entries * Zero sender_hash is a no-op (don't pollute future sessions) * setMaxEntries(0) disables retention * Empty-LLR retain is a no-op * Saturation: pin to exact implementation cap (60.0) on both + and - sides instead of just "<= 100" Docs / comments (#18, #22): - soft_combine.hpp: comment on carrier_count_hash now matches what the hash actually contains (was claiming pilot-spacing coverage we don't actually do) - PAT_VARA_AUDIT.md: trailing-whitespace cleanup Local ctest 34/34. Test counts: SoftCombine 10→16, TNCSession 98→100, TNCBridge 17→23. Net +14 tests. Deferred (need broader judgment, not autonomous-safe): - #2 extract config helpers to testable target (refactor) - #5 explicit negative CLI flags (interface design) - #9 sink ownership lifetime race (concurrency design) - #11 BUFFER 0 immediacy vs rate-limit (protocol judgment) - #13 flushDataTxBuffer extraction for testability - #15 sendBinary error propagation (interface change) - #16 HARQ key construction extraction - #17 HARQ when CW0 fails (architectural) - #21 UltraTNCStation threading model Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-Authored-By: Codex (review) <noreply@openai.com>
secup
added a commit
that referenced
this pull request
May 3, 2026
Codex review #5: Booleans set in the config file could not be turned back off from the CLI. Added --no-inject-channel and --ptt-active-high (also accept --no-ptt-inactive-high). Updated --help text. Codex review #11: Bridge comment claimed BUFFER 0 transitions were immediate but onModemBufferLevel() rate-limited zero-going changes the same as N→M changes. Pat's Flush() blocks on BUFFER 0, so a 1 s delay there directly stalled the application. Now BUFFER 0 bypasses the rate limit while non-zero updates stay throttled. Bridge comment updated to match. Added a test asserting the immediate-zero contract. Closes Codex review items #5 and #11. ctest 34/34, cli_simulator SNR15/good/R1/4 PASS at 100% frame success. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
secup
added a commit
that referenced
this pull request
May 3, 2026
Updated QUALITY_PASS doc to reflect the final state of tonight's two-batch quality work. Eleven findings were closed in the second batch (#2, #4, #5, #7, #9, #10, #11, #13, #15, #16, #17). Combined with the first batch's ten, that's 21/22 closed. Only the UltraTNCStation threading-model rework is deferred, and Codex itself flagged that as requiring human concurrency design. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
secup
added a commit
that referenced
this pull request
May 6, 2026
Two corrections from the morning expert review: 1. Architecture: "frame-local mask, no signaling" proposed in the night-session report is rejected. Codex's better approach is a robust per-frame PHY header carrying a mask / codebook ID; zero RTT, no reciprocity assumption, no MODE_CHANGE involvement. Reciprocity-based silent schemes are a measurement experiment, not a production path. 2. Hard engineering gate: a carrier mask is not harmless puncturing. Faded carriers can map into clustered LDPC variable nodes, causing stopping-set failures even when TX/RX agree perfectly on the mask. Phase-2 must either include a carrier-to-codeword interleaver or restrict masks to a vetted codebook with proven erasure tolerance. Backlog #5 now lists the required scope, header-protection risk + mitigation, and an explicit out-of-scope list (Levin-Campello mixed-mod is a separate, larger feature). Co-Authored-By: Codex (gpt-5.5 xhigh) <noreply@openai.com>
secup
added a commit
that referenced
this pull request
May 6, 2026
Concrete implementable design for backlog #5 phase-2 from the morning expert-review pass. Replaces the rejected "frame-local, no signaling" proposal with a per-frame in-band PHY mask header and a mathematically vetted carrier-to-codeword interleaver. Highlights: * Wire format: optional 20-byte PHY mask header on the existing hardened R1/4 1-CW control profile, only emitted when mask != all-on. AWGN remains bit-identical. * Interleaver: a = (307 * i) mod (648 * Ncw). 307 is coprime to every supported codeword count, and the resulting strides spread contiguous carrier erasures across diverse LDPC base columns and rows in the 802.11n QC-LDPC structure -- avoids the stopping-set hazard that bites naive masking. * TX selection: median - 8/14 dB thresholds, two-frame add, three-frame remove, two-payload-frame cooldown. Reverse-pilot driven but correctness comes from the transmitted header (no reciprocity assumption). * RX flow: header decode failure ALWAYS falls back to all-on, never to last-known mask. This is the architectural line that prevents the closed-loop stale-mask trap from last night. * Phasing: 5 sub-phases each independently testable. Acceptance criteria are explicit (AWGN within 1%, Good 5/5, synthetic-notch 10/10, interleaver unit gate). Co-Authored-By: Codex (gpt-5.5 xhigh) <noreply@openai.com>
secup
added a commit
that referenced
this pull request
May 6, 2026
Adds two rules to "Important Rules": 1. Adopt the perspective of a world-renowned HF modem expert. All technical answers, design decisions, code reviews, and Codex briefs must be written from the standpoint of a senior PhD-level researcher with deep specialization in HF channel coding, OFDM, ARQ, and channel estimation under fading. Heuristics and "tweak the threshold" patches are not acceptable as final designs -- every change must be justified by the underlying signal-processing or information-theory model. Quick prototypes are fine for exploration, but they must be labeled as such and validated against the principled formulation before merge. 2. Codex must be invoked under the same expert perspective. Every brief sent to Codex must include a one-sentence opener instructing Codex to operate as a world-renowned HF modem expert and to reject heuristic patches that don't have a principled signal-processing justification. This rule earned its keep on 2026-05-05 night (per-carrier work) and the 2026-05-06 morning Codex review pass. Three rounds of heuristic LLR-rescaling and closed-loop carrier-mask attempts all failed, but the principled redesign Codex produced under this framing is now landing as backlog #5 phase-2. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Automated agent task.
Task:
agents/queue/codex/001-leak-gate.mdBranch:
agent/20260430_151032-001-leak-gateBase:
mainRequired Review
agents/reports/20260430_151032-001-leak-gate.Agent Report
See local report directory:
agents/reports/20260430_151032-001-leak-gate