Conversation
…rung), derived from the measured inventory Operator 2026-09-21: "lets ensure this release has MORE than enough tokens for our needs (overdo it), and our verbs are dogfooded, i.e chat with and without thinking, ditto run, ditto code, etc,". #3712 row B1, the judge side; stacked on row A. - scripts/lib/model_ladder_cells.py (called from check_model_ladder.sh): the owed set is short rungs up to each model's context_length plus long rungs for qwen35 and one representative per other arch, x cells.verbs x the thinking modes its TEMPLATE evidence derives (#3723). fit := weights + kv_bytes_per_token*tokens + workspace <= the host's TOTAL GPU memory (operator "a", #3710). A fitting cell must pass (cuda, no fallback, rc 0, prompt >= rung, an answer, thinking closed); a non-fitting one must be an honest refusal; every owed cell passes on >= 1 required host; every arch has a held representative. Rungs come from evidence/release/context-rungs.json (#3715), one declaration shared with pv; missing or unsized is a named FAIL. - contract: ladder.cells {verbs, long_rungs_for}; MCL-INV-008; FALSIFY-MCL-016. The cells block is an anti-shrink floor vs origin/main. - case table 26 -> 45; 8 cells-module mutants, each killed by its case. Refs #3712 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Contributor
Author
|
quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false) {
"ticket": "PMAT-3712",
"head": "7d7a338f301eb938632f567c3a94f5cbe9266f76",
"width": 3,
"executor": "agy",
"agreed": false,
"auto_merge": {
"checked": true,
"was_armed": false,
"disarmed": false,
"note": "auto-merge not armed"
},
"lanes": [
{
"lane": 1,
"verdict": "FAIL",
"findings": 3
},
{
"lane": 2,
"verdict": "FAIL",
"findings": 1
},
{
"lane": 3,
"verdict": "PASS",
"findings": 5
}
]
} |
…loor is the context-rungs file Quorum round 1 on #3738 (1 PASS / 2 FAIL) found two real defects: - a non-fitting cell that reported `pass` was accepted, contradicting MCL-INV-008 (not fits => refused). The arithmetic decides which cells are owed, so a pass beyond it proves the arithmetic wrong: now a named FAIL. - the rung-id anti-shrink read the cells block, which has no rungs since they moved to evidence/release/context-rungs.json: vacuous. The floor is now that file at origin/main (rung ids and consumers; absent at main = bootstrap). The module also keeps only REQUIRED hosts' receipts itself instead of relying on the caller. The third finding (qwen3-8b required: false) is answered by the issue author's done_when 1 amendment on #3712, quoted verbatim in the fragment. Case table 45 -> 47; cells-module mutants 8 -> 10, each killed by its case. Refs #3712 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Contributor
Author
|
quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false) {
"ticket": "PMAT-3712",
"head": "bc37412ebda3b253e5dae7d95d2bd3fe55fd4231",
"width": 3,
"executor": "agy",
"agreed": false,
"auto_merge": {
"checked": true,
"was_armed": false,
"disarmed": false,
"note": "auto-merge not armed"
},
"lanes": [
{
"lane": 1,
"verdict": "NO-VERDICT",
"findings": 0
},
{
"lane": 2,
"verdict": "FAIL",
"findings": 2
},
{
"lane": 3,
"verdict": "PASS",
"findings": 0
}
]
} |
…ent and done_when 5; title matches Quorum round 2 on #3738 judged the cells matrix unrequested scope (the operator's bar raise was a comment, not a done_when) and read the fragment title's "no Q4_K rung optional" against the ruled required: false. The issue author amended #3712 (comments 5764277522 and 5764495191); the fragment quotes both verbatim, and the title now states the amended criterion. Refs #3712 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Contributor
Author
|
quorum-review (AD-04): NOT agreed (auto_merge: checked=true was_armed=false disarmed=false) {
"ticket": "PMAT-3712",
"head": "234d7a96f55acc616413d61707fdffa1b52465df",
"width": 3,
"executor": "agy",
"agreed": false,
"auto_merge": {
"checked": true,
"was_armed": false,
"disarmed": false,
"note": "auto-merge not armed"
},
"lanes": [
{
"lane": 1,
"verdict": "FAIL",
"findings": 1
},
{
"lane": 2,
"verdict": "PASS",
"findings": 0
},
{
"lane": 3,
"verdict": "PASS",
"findings": 0
}
]
} |
…he arch it matches is measured Quorum round 3 on #3738 (2 PASS / 1 FAIL): the FAIL claimed qwen35-family GGUFs report general.architecture = qwen2, so families: [qwen35] never matches, and that the tests hold no qwen35 model. Both are refuted by measurement: Qwen3.5-0.8B/4B report qwen35 (apr inspect and a direct header read), and the fixture's q.gguf is arch qwen35 and owes every long rung. The refutation is now executable: cells mutant family-long (the families check deleted) is killed by red-cells-missing-cell. The fragment records the measurement. Refs #3712 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Contributor
Author
|
quorum-review (AD-04): three PASS — agreed (auto_merge: checked=true was_armed=false disarmed=false) {
"ticket": "PMAT-3712",
"head": "fd636b3d734cdc305c93a8aed1e773eb4c1bc5b9",
"width": 3,
"executor": "agy",
"agreed": true,
"auto_merge": {
"checked": true,
"was_armed": false,
"disarmed": false,
"note": "auto-merge not armed"
},
"lanes": [
{
"lane": 1,
"verdict": "PASS",
"findings": 1
},
{
"lane": 2,
"verdict": "PASS",
"findings": 1
},
{
"lane": 3,
"verdict": "PASS",
"findings": 0
}
]
} |
noahgift
added a commit
that referenced
this pull request
Sep 21, 2026
…— the vocabulary is S1's surface (#3745 S3) Operator 2026-09-21: "the way to prevent testing leakage is tests are dervived from surface and SHACL enforced...WE ARE NOT ENFORCING our interface AT ALL..it is hand coded not derived". #3752 (#3745 part S3), stacked on S1 (#3749, surface::emit()). crates/apr-cli/src/hand_list_guard.rs, a #[cfg(test)] module of apr-cli's lib: - vocabulary = crate::surface::emit(): verbs are command path tokens and aliases; flags of a verb are its --long/-short/aliases plus global_args. Nothing here names a verb. - an enumeration (S3.1 as amended): a literal sequence whose verbs, or flags of one verb, are >= 2 AND a strict majority of its items. Read from YAML, shell arrays (+= merged), for-in lists, verb case arms, comma/assigned strings, and bracketed string-literal lists in Python heredocs and Rust (merged across `+`). An argv, a script's own option parser and clap's help/version are not enumerations. - scan set: the Cargo.toml dogfood gates, scripts/release/**, the dogfood scripts, the ladder and release-readiness contracts, the release-evidence extractor. No allow-list. - runs in workspace-test (required) on every PR and in the pre-publish dogfood through `make coverage-check` (llvm-cov --workspace --lib; a failure is a NO-GO). Measured: 4/4 green on main + S1; RED on #3738's ladder cells.verbs when dropped in. S3.4 (the ladder/B1 migration to S2's derived cells) lands on S2. Refs #3752 #3745 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This was referenced Sep 21, 2026
Contributor
Author
|
closed by triage (aprender-dd; cop aprender-cf approved): FOLDED, not abandoned. Per the release-process owner aprender-6c [8b6b78]'s ruling on #3712, this row was folded into the 0.69.1 squash batch #3772; its fold order lists it, and its cases are on main. The squash is why its head is not an ancestor of main. Branch kept. |
This was referenced Sep 23, 2026
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
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.
#3712 row B1: the model gate judges cells (verb × thinking × context rung), derived from each model's measured inventory
Stacked on #3721 (row A). Scope: done_when 5 on #3712 (issue author, comment 5764495191) adds the cells matrix; done_when 1 is amended in comment 5764277522 (the qwen3-8b flip rides the 0.69.1 fold). The fragment quotes both verbatim. The base is
PMAT-3712-ladder-universe-inventory, so this diff is B1 only.Operator 2026-09-21, verbatim: "lets ensure this release has MORE than enough tokens for our needs (overdo it), and our verbs are dogfooded, i.e chat with and without thinking, ditto run, ditto code, etc,". A cell is (model, host, verb, thinking, context rung). This PR is the judge side. The producer side (B2, measuring the cells) depends on apr work in flight and is not here.
What the judge now requires (
scripts/lib/model_ladder_cells.py, called fromcheck_model_ladder.sh)context_length, plus long rungs forqwen35and ONE representative per other arch (named in the contract), ×cells.verbs× thinking modesenable_thinking→ both; an always-open<think>→ on; otherwise off. The producer's own claim can't shrink itevidence/release/context-rungs.json(apr-release-context-rungs/v1, #3715), one declaration shared with pv. A missing file, or a consumer without a number, is a named FAILcellsblock (block, verbs, long-rung families and representatives) andcontext-rungs.json(rung ids, consumers), since that file is where the rungs liveMeasured on this head
check_model_ladder.sh --self-test: 47 cases, 0 bad (21 new cells cases, each red for its own reason). All mutants are killed: 3 judge, 4 producer-lock, and 11 cells-module mutants (missing-cell,cotenant,passes-nowhere,think-closed,fell-back,prompt-short,modes-evidence,no-representative,pass-beyond-fit,rungs-floor,family-long), each by its named case.general.architecture = qwen35(not qwen2), sofamilies: [qwen35]matches by arch. The fixture's qwen35 model owes every long rung, and thefamily-longmutant proves the rule is tested.context-rungs.json. The third finding (required: false) is covered by the issue author's done_when 1 amendment on P0 release-process: the train can publish with "most models working" — the model gate is a hand-picked ladder with OPTIONAL rungs, a skip reads as pass, and the dogfood can be inherited. Every Q4_K model × {lambda, gx10} × CUDA must be green or NOTHING ships #3712, which the fragment quotes verbatim.guard_tree --no-cargo76 checks, 0 failed.pv validateok ·pv lint contracts/PASS · census andcontracts.ntunchanged · bashrs 0 errors · shell-lint ratchet PASS.Where the gate runs
check_model_ladder.shis declared only inCargo.toml[package.metadata.dogfood]gates (the T-2 pre-publish dogfood). It is also listed inscripts/unwired_guards_baseline.txt: no PR workflow runs it. Its real run is RED until the producer writes cells and #3715 lands the rung file. That is the operator's release bar, not a PR red.Not here
apr inspectdropsqwen35.*) and Time-to-first-token on Qwen3.5-4B-Q4_K_M is 14-16s vs llama.cpp's 1.8-2.9s on the same box (7.7x); apr's decode rate is unmeasurable with shipped tooling #3596 (capacity::plan, the pre-load refusal).Refs #3712
keep-open: row B1 of #3712 only. Row B2 (the producer's cell measurement) and done_when 3 and 4 remain; the rung file lands with the pv shape; the qwen35 declared rung waits on the inspect fix.
🤖 Generated with Claude Code