Skip to content

spec(queen): keyed_batch.t27 -- the bee admission admits a batch of ready issues per queend choose (Closes #10033) - #10051

Merged
gHashTag merged 1 commit into
masterfrom
claude/festive-golick-124a9d
Oct 17, 2026
Merged

spec(queen): keyed_batch.t27 -- the bee admission admits a batch of ready issues per queend choose (Closes #10033)#10051
gHashTag merged 1 commit into
masterfrom
claude/festive-golick-124a9d

Conversation

@gHashTag

@gHashTag gHashTag commented Oct 17, 2026

Copy link
Copy Markdown
Owner

Closes #10033. Epic #7851.

Why

On 2026-10-17 the keyed-actors trial (trios 42042a5, TRIOS_QUEEN_DISPATCH=actors) started 2 runners in 20 minutes. The round loop starts 18.5 in the same time, so the trial was reverted (evidence).

The cause is that admission is serial:

  • The round readies about 826 candidates.
  • The one bee-admission actor then awaits one queend choose per issue (queen-bee-actors.ts:298).
  • The loop asks queend once per round.

What

This PR adds specs/queen/keyed_batch.t27, a new card. It changes how many queend calls the admission makes, not what it answers: a batch holding one ask answers exactly as keyed_guard.t27 admit_answer does.

Part Functions Rule
Who joins joins_batch, answer_now Only the asks admit_answer would send to queend join a batch. The rest are answered at once.
Group commit batch_goes_out, batch_take, chooses_for A batch goes out when no choose is in flight, with no timer. It holds up to BATCH_ASKS_MAX = 1024 asks, which is at least the measured 826.
Quota new_lane_quota Uses control.t27 batch_quota, the loop's own cap.
Answer word_of, batch_answer, batch_grants Each issue gets one word: PICK, LATER or NO. A kept ask takes no new lane. Picks past the free lanes wait, and a batch grants exactly its quota.
Wake batch_wake_count A freed lane wakes every waiting issue into one batch (up to MAX), not one issue per lane.
Model round_chooses, chooses_of, grants_of For 826 candidates on 47 free lanes with one pick in ten: 461 serial chooses for 47 grants, against 1 batched choose for the same 47.
Bounds ask_wait_ms, ask_bound_ok, word_after_choose An ask's call must outlast two chooses: ASK_BOUND_MS 30000 > 2 x CHOOSE_BOUND_MS 12000. The choose bound is chosen, not measured. A choose that ends with no answer gives LATER, so it moves no lane and refuses nobody.

The card has 13 tests and 6 invariants.

Lab evidence (Railway lab only, nothing run on the Mac)

The lab ran on a tree of master 9a7d54a with t27c-bootstrap 0.5.2 and zig 0.16.0. The spec's sha256 is 80709724eab9fa180050984b0586ef224c87f6a8b70db245d3d04bb5af177006.

  • t27c test-report: 13 of 13 tests pass, with 0 vacuous. The 6 invariants are proved at comptime.
  • Backends: gen, gen-c, gen-rust, gen-verilog, gen-js and gen-ts all exit 0.
  • tri mutate spec --jobs 8: 89 mutants, 85 killed (68 by a test and 17 by an invariant at compile time). The other 4:
    • 2 survivors, both equivalent: < to <= at line 93 (batch_take) and line 168 (batch_wake_count). At open == MAX both branches return MAX.
    • 1 hang: line 154 i = i in batch_grants, which is the mutant's own endless loop.
    • 1 unviable: line 234 ask_bound_ok -> return false. The invariant the_bounds_this_card_sets_fit rejects it at comptime.
    • An earlier run had 2 more survivors: candidates - i to + i, and j < i+chunk to <=. Both showed a test gap. The added (1500, 400, 5) assert now kills them.
  • Hand constant plants (tri mutate plant): all 10 are caught.
    • Killed by a test: 1024 to 2048.
    • Blocked by an invariant: 1024 to 825, 826 to 1025, 185 to 200, 20 to 18, PICK 0 to 3, LATER 1 to 0, NO 2 to 1, 30000 to 24000, and 12000 to 15000.
  • Wasm card:
    • Build: t27c gen-c followed by zig cc --target=wasm32-wasi -Oz -DNDEBUG -mexec-model=reactor. It exports the 16 card functions plus lane_free, admit_answer, wake_count and batch_quota.
    • Result: two builds give one sha256, 1ed6bcb13de9ff4c462ff6d0f8a1ba6e3c590d0235756f3aa4ce4bd8fcacfd8a (1187 bytes).
  • Seal: .trinity/seals/queen_QueenKeyedBatch.json, built by t27c-bootstrap@0.5.2+cc80a305d.
    • Its closure is control, keyed and keyed_guard.
    • 13 tests passed. --verify reports all hashes MATCH.

Runtime

The thin TypeScript wiring is a gHashTag/trios issue (linked in a comment below). The before/after numbers against the loop go on #7851.

Follow-up noticed

gen-c does not escape C keywords. A local named short broke the wasm build, so this card names it rest.

🤖 Generated with Claude Code

@gHashTag @claude
spec(queen): keyed_batch.t27 -- the bee admission admits a batch of r…
ea34ba0 
…eady issues per queend choose (Closes #10033)

The actors trial on trios 42042a5 started 2 bees in 20 minutes, against
the loop's 18.5, because the one bee-admission actor asked queend once
per issue. This card says how a batch of asks is answered with one
choose: which asks join a batch, group commit with no timer, the loop's
batch_quota, one policy word per issue, a batched wake, and the bound
an ask's call must keep over a choose.

Lab evidence (Railway lab): 13/13 tests, 0 vacuous, 6 invariants;
mutate 85/89 killed (2 equivalent, 1 hang, 1 unviable); 10/10 constant
plants caught; reproducible wasm; seal verified.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag gHashTag mentioned this pull request Oct 17, 2026
Open
@gHashTag

gHashTag commented Oct 17, 2026

Copy link
Copy Markdown
Owner Author

Runtime wiring: gHashTag/trios#2043

@gHashTag gHashTag mentioned this pull request Oct 17, 2026
Open
5 tasks
@gHashTag
gHashTag enabled auto-merge (squash) October 17, 2026 17:33
@github-actions

github-actions Bot commented Oct 17, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-17 17:34:33 UTC

Summary

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

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e769ef021713 != 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).

@gHashTag
gHashTag merged commit 38b4f74 into master Oct 17, 2026
25 of 29 checks passed
This was referenced Oct 17, 2026
Closed
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.

spec(queen): keyed_batch.t27 -- the bee admission admits a batch of ready issues per queend choose

1 participant

@gHashTag