Skip to content

Port gHashTag/trinity:fpga/openxc7-synth/d_clk_direct.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_c - #5595

Merged
3 commits merged into
masterfrom
queen-5591
Oct 4, 2026
Merged

Port gHashTag/trinity:fpga/openxc7-synth/d_clk_direct.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_c#5595
3 commits merged into
masterfrom
queen-5591

Conversation

@gHashTag

@gHashTag gHashTag commented Oct 2, 2026

Copy link
Copy Markdown
Owner

Closes #5591

Written by a bee on queen-5591 and published by tools/queen/publish.py. The branch itself is the bee's; the second commit is the coordination entry every pull request must add, which a bee has no way to know about.

1 file changed, 34 insertions(+)

🤖 Generated with Claude Code

Trinity Bee and others added 2 commits October 2, 2026 14:12
Trinity Bee
salvage(queen-5591): commit what the turn left uncommitted
b5a1839 
The turn ended with these files edited and never committed. Uncommitted
work is invisible to the review - it reads the branch - so the attempt
would have been released as empty and the next bee would have started
beside this work rather than from it.

This commit is not a claim that the work is correct. It is the bee's
work, committed on its behalf, and it is judged exactly like any other:
the adversarial reviewer reads it, the compiler runs on it, and the
issue's own criteria are measured against it.

Issue: #5591
Turn: 05d17d19-6f1d-4ab6-ade9-57db4f609f3c
Ending: finished (the turn closed)
Committed: 1 path(s)
Left uncommitted: 0 path(s) outside the declared boundary
@claude
docs: the coordination entry this branch needs to land
280ab43 
A pull request must add exactly one docs/now entry and a bee has no way
to know that: its brief names a boundary file and acceptance criteria,
and docs/now/ is neither. The publisher adds it rather than failing the
gate.

Closes #5591

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@github-actions

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

This notebook contains session context, decisions, and artifacts for this work.

@github-actions

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-02 14:23:50 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 46
PRs with All Checks Green 4
READY 3
FAILING 46
PENDING 0
NO CHECKS YET 0

These columns do not partition: 3 + 46 + 0 + 0 = 49, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=b7d5cc5c4cf1 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

This was referenced Oct 2, 2026
Merged
Merged
Merged
Merged
Merged
@gHashTag

gHashTag commented Oct 2, 2026

Copy link
Copy Markdown
Owner Author

Reviewer bee W -- changes requested at head 280ab435a9. Not merging.

t27c typecheck passes, but the port is not ready:

  1. No seal. t27c seal specs/port/trinity/fpga/openxc7-synth/d_clk_direct.t27 --verify gives No saved seal found at .trinity/seals/openxc7-synth_trinity_top.json. Run --save with a t27c built from current master and commit the json. The module name trinity_top is shared by about 10 sibling ports, so check that the seal path does not collide with an existing one.
  2. The check (now-freshness) job is red. The NOW header must match # NOW -- <title> (YYYY-MM-DD). The parenthesised "(Verilog, 1 module)" inside the title breaks that pattern. The job also reports that the entry says nothing checkable, so name the module, the source file and the verification you ran.
  3. Check L1 TRACEABILITY is red. No commit carries Closes #5591 (commit-level count was 0). Put it in the commit message, not only in the PR body.
  4. The invariant is assert(true) under generated filler comments. That is assertionless in substance. Assert something real about the clock-direct wiring (for example the port widths or the pass-through relation), or drop the invariant.

Not my push to make: whoever fixes this should leave the merge to a different reviewer bee.

This was referenced Oct 2, 2026
Merged
Open
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
@gHashTag

gHashTag commented Oct 2, 2026

Copy link
Copy Markdown
Owner Author

Seal for specs/port/trinity/fpga/openxc7-synth/d_clk_direct.t27: not added by this bee, on purpose.

  • Seals are keyed by directory + module name. This spec declares module trinity_top, like all 11 sibling ports already on master in specs/port/trinity/fpga/openxc7-synth/, none of which is sealed. t27c seal --save writes it to .trinity/seals/openxc7-synth_trinity_top.json. That key would belong to whichever of the 12 identically named modules was sealed last.
  • No gate on this PR asks for the seal. coverage (check_seal_coverage.py) passes. spec-guards fails in check_ring_spec_drift.py, which does not involve this spec. The red checks that do involve this PR are Check L1 TRACEABILITY, where the two commits carry no Refs #5591, and check, where the docs/now entry "does not say anything checkable".

If the seal is still wanted, rename the module first (for example module d_clk_direct). Then run it with a t27c built from the same master revision the PR merges onto:

git fetch origin master && git merge origin/master
cargo build --release -p t27c
./target/release/t27c seal --save specs/port/trinity/fpga/openxc7-synth/d_clk_direct.t27
./target/release/t27c seal --verify specs/port/trinity/fpga/openxc7-synth/d_clk_direct.t27
git add .trinity/seals/ && git commit -m "seal: d_clk_direct port (Refs #5591)"

For reference, a t27c from 2026-10-02 (not master's revision, so do not commit these values) gives spec_hash=sha256:43d098e9….

@github-actions github-actions Bot mentioned this pull request Oct 2, 2026
Closed
This was referenced Oct 3, 2026
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
gHashTag added a commit that referenced this pull request Oct 3, 2026
@gHashTag @claude
bees/reviewer: run the agent on z.ai's free GLM flash, on the keys we…
0be7b95 
… hold (Refs #5776)

The reviewer needed a `claude setup-token` in the Keychain because the CLI's
own OAuth login cannot refresh under launchd. It does not need Anthropic at
all: `claude -p` is only the harness (its sandbox flags keep the agent
read-only), and z.ai serves the same API.

Measured 2026-10-04 on the five keys in ~/.claude/.env: glm-4.7-flash and
glm-4.5-flash answer, free; every paid model answers [1113][Insufficient
balance or no resource package]. So:

- `--provider zai` (default; BEE_REVIEWER_PROVIDER): ANTHROPIC_BASE_URL is
  z.ai's Anthropic endpoint, model glm-4.7-flash, glm-4.5-flash as the CLI's
  fallback when the first is overloaded (1305).
- Keys: ZAI_API_KEY, ZAI_API_KEY_2.. from the environment, then ZAI_KEY_N
  from ~/.claude/.env (BEE_ZAI_ENV_FILE names another file). Round-robin;
  a key z.ai refuses (1113, 401, 1302/1303) hands the same review to the
  next key; all refused is AgentUnavailable, which charges no head.
- The agent sees one key as ANTHROPIC_AUTH_TOKEN, never the pool, and none
  of the desktop app's login or model settings. No key is ever logged.
- `probe` sends one tiny turn per key: 5 of 5 answer under `env -i`.
- `--provider claude` keeps the Keychain setup-token path.

Dry run on two live pull requests: #5595 REQUEST_CHANGES in 5 turns / 50 s;
#4498 contradicted itself (APPROVE plus a blocking-check line) and the
runner's own gate called it incomplete, posting nothing.

self-test: 85 checks, 0 failures.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Oct 3, 2026
Merged
Merged
Merged
Merged
gHashTag added a commit that referenced this pull request Oct 3, 2026
@gHashTag @claude
bees/reviewer: repair an approval that leaves a line out; tick and it…
cec8b61 
…s trend (Refs #5776)

An APPROVE whose block only leaves something out -- no summary, no criterion
line, a red check with no discounted-check line -- now gets the same one repair
turn an answer with no block gets, told exactly what is missing and which
checks are red. It may discount a check only where the review already said
why; otherwise it must answer REQUEST_CHANGES. The corrected block replaces the
old one (two blocks would be two verdicts). A block that contradicts itself
(an unmet criterion, a blocking-check) gets no repair turn: that is a
judgement. Measured on the live log of 2026-10-03: #5595 was thrown away whole
as "red check(s) not discounted"; 4 of 7 live reviews ended without a usable
verdict.

`reviewer.py tick` (and `tri review tick`): one look -- health, queue by
reason, reviews and merges since the last look, the merger's merges told from a
person's by mergedBy.is_bot -- appended to ~/.local/state/t27-bees/ticks.jsonl,
then what only a run of looks shows: work queued on three looks with no review,
the same failure on two looks in a row (a repair that did not hold), approved
and labelled pull requests not merged after two hours. It repairs nothing.
`doctor_findings` is split out of `cmd_doctor` so tick reuses it unprinted.

reviewer.py self-test: 130 checks, 0 failures.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag added a commit that referenced this pull request Oct 3, 2026
@gHashTag @claude
bees/reviewer: read again when GitHub fails a read for a moment (Refs #…
a2bba25 
…5776)

Live, 2026-10-03T20:10Z: the head re-read before posting #5595's verdict got
GraphQL's "HTTP 401: Requires authentication" on a valid keyring login; the
next read two minutes later worked. The verdict -- six minutes of model time --
was dropped, nothing was recorded, and the head goes back in the queue.

Every call through `Gh` is a read (writes go through the bot's token), so a
transient failure is asked again after 5 s and 20 s. A REST 401 (a dead
login) and a 404 fail at once. Four self-test checks; with the retry turned
off, the two that need it fail.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Oct 3, 2026
Merged
Merged
Merged
Merged
Merged
Merged
Closed
Merged
Closed
Merged
gHashTag added a commit that referenced this pull request Oct 4, 2026
@gHashTag @claude
bees/reviewer: a fallback is z.ai's load, not the head's -- it spends…
8492e0c 
… no attempt (B18) (Refs #5776)

When z.ai is overloaded (1305), the CLI falls back to the second flash
model partway through a review. That review cannot be seconded, so it ends
incomplete, and until now it also used up one of the head's two attempts.
During the 23:32-00:21Z overload, #5781, #5783, #5820 and #5822 each took
a fallback and ran out of attempts. They then sat waiting for a new push
even though nothing in them had failed.

head_history no longer counts an incomplete whose first review fell back.
An agent-failed row still counts. The predicate is now one function,
fell_back, and the fallback rate in stats uses it too. Replayed over the
live rows, those four heads become reviewable again; #5595 and #5689 stay
out because both of their failed tries were their own.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Oct 4, 2026
Merged
Merged
@dmitrii-f-t27 dmitrii-f-t27 closed this pull request by merging all changes into master in 4ddd9ee Oct 4, 2026
@gHashTag gHashTag mentioned this pull request Oct 4, 2026
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Port gHashTag/trinity:fpga/openxc7-synth/d_clk_direct.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_clk_direct.t27

3 participants

@gHashTag @dmitrii-f-t27 @claude