fix(ENG-RELEASE-BINARIES): the roadmap REL row carries no release lifecycle, and its 'no published binary exists' clause is false against tag v0.0.2 - #3340
Open
phantomic12 wants to merge 3 commits into
Conversation
phantomic12
force-pushed
the
row/RELEASE-LIFECYCLE-ROADMAP
branch
from
September 28, 2026 19:43
f407119 to
22c9249
Compare
Collaborator
|
Review status: the roadmap statements are statically consistent with tags and release records, but PASS is not established without focused release-checker runs and lifecycle/publication mutations. Keeping this open for executable verification. |
…ed, and reconcile the tree to its gate Merge 4099082 resolved scripts/check-agent-record.py to its pre-restructure parent, silently discarding the restructured checker that the merge's second parent carried: fence-aware link extraction, strip_code_spans, the record-anchor ratchet, the mudler#3099 claim-state consistency gate, the roadmap issue-row refusal, and derived matrix counts. The tests re-pointed at the restructured surface (1de097c) then failed against a checker that no longer exposed it: 124 tests, 93 failures and 31 errors at the stale local main, still 72F/83P at origin/main. RESTORED: scripts/check-agent-record.py now carries the restructured implementation again (merge parent 8b5435b), whose counts are DERIVED from the rows rather than pinned. The model-inventory constants are regenerated to the current tree (rows 324, memberships 373, architectures 356, targets 309, modules 245) with an annotation naming the SystemOne decision-row closure that explains the delta, and the citation-preserving padding is kept intact so the tracked line anchors still resolve. RECONCILED, each because the lax interim checker had accepted it: seven claim files now annotate their row's matrix lifecycle state; roadmap_v1 no longer stores fifty GitHub issue table rows and points at .agentss/issues/ authority instead; three issue files move to row-named directories whose Row field names a canonical claimable row (MODEL-DSV4-EXL3, the DeepSeek-V4 text row, MODEL-GLINER25); the new _owed issue is listed under the owning spec's ## Owed; and 44 stale or broken record anchors are re-pointed to the lines that now hold what the prose says — including the four server-CLI anchors, whose file shrank to a thin ABI client and whose flags now parse in src/vllm/entrypoints/openai/server_main.cpp. TESTS: the mutation tests that pinned the retired per-matrix count constants are deleted with a comment naming the retired capability — the restored checker derives counts, so there is no constant to ratchet. The check_issue_records test shim is dropped (the restructured checker owns the real entry point), the derived-count membership tests are re-composed against the path-only MATRICES registry, and test_record_merge_shape reads the ClaimRecord shape the claim-state gate needs. GATES: python3 scripts/check-agent-record.py OK (ENGINE=179 MODEL=384 QUANT=87 KERNEL=60 BACKEND=90 ANCHOR-ROT=0); pytest tests/scripts/test_agent_record.py 123 passed, 60 subtests passed; test_record_merge_shape and test_issue_records green. Fixes ISSUE-LOCAL-01M3JMSD9P7REKNQVEG9HCEWTD FOLLOWING_AGENTS_PROTOCOL Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:zai/glm-5.3-flash [maki]
…oncile the drifted records The 4099082 merge dropped the restructured check-agent-record checker, so main carried the old count-constant monolith against a test suite written for the restructured API: tests/scripts/test_agent_record.py failed with 29 errors and 6 failures on a tree the old checker called green. Restore the restructured checker (cherry-pick of bdcf37fa9's scripts/check-agent-record.py and scripts/issue_records.py, whose counts are derived from the rows, so the stale ENGINE/MODEL/QUANT/BACKEND count constants and the model-inventory dict drift disappear as a class), then reconcile the tree the restored gate actually flagged: - Claims/specs: muse-glimmer-parity.md and nano-nemotron-vl-radio.md listed three row-owned issue IDs (01M3S0ZTRRVQWCZ35W33AJJNN8, 01M3RY6G385D41W5SNF1C85RRS, 01M3S07X6Y4ADHHQ02FYMR4RXF) under ## Owed; a spec's Owed section may only own _owed IDs, so each reference moved to a new ## Tracked section with the same traceability. The MiMoV2 issue carried the invalid State PARTIAL; the row is ACTIVE (W1 landed, CLAIM-MIMOV2-PORT), so the issue is OPEN. - Record anchors: 10 citations in engine-matrix.md named lines the prose no longer holds (handle_metrics 1322->1354, handle_tokenize 1333->1365, set_tokenizer_info_enabled 396->408, kAcceptedInertArgs 339->409, handle_abort_requests 1582->1614, set_abort_requests 403->415, LoadedEngine::FromModelDir 2548->2556, process_inputs 249->266, plus the two duplicate row-cell spellings); anchor rot returns to 0 against the unchanged baseline. - Checker-line citations: six tracked specs cite check-agent-record.py line numbers the restore moved; re-pointed each to the line that now holds the cited statement (link_bases region 1078-1107->1094-1184, check_spec 1169->1857, the active-claim Row-ID refusal 1712-1714 -> 1975 with the ID_RE prefix list 327-338->903-906, ISSUE_ROW 1960->2052, owed_issues 2000->2017, check_issue_index 2018->2044). - Tests: test_malformed_row_is_still_seen_downstream asserted the retired count-constant error format; the restructured checker already reports both the shape and the duplicate finding in one run, so the obsolete third require is re-pointed at that contract instead. The checker-line citation scan now skips a tracked symlink dereferenced to a directory (.agents/evidence/.../latest), which carries no line citation. test_agent_record.py: 106/106. Files: ISSUE-LOCAL-01M3SQTG14MVDB5MWCQPB418QM FOLLOWING_AGENTS_PROTOCOL Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:zai/glm-5.3-flash [maki]
phantomic12
force-pushed
the
row/RELEASE-LIFECYCLE-ROADMAP
branch
from
October 1, 2026 18:31
22c9249 to
dcd4196
Compare
…ycle, and drop its false "no published binary exists" clause check-release-binary-contract and check-windows-release-state were both red, and neither for a reason a reader could act on: the text they require has never been in .agents/roadmap_v1.md. git log -S "v0.0.2 published eight primary archive" over that path returns nothing, and at 1de097c the file contains no v0.0.2, no v0.0.3-pre.1 and no ENG-RELEASE-WINDOWS anchor. The v0.0.2 facts live in the ENG-RELEASE-BINARIES row of .agents/engine-matrix.md, so the two gates disagree about which file owns the lifecycle -- the same shape as check-benchmark-index, where a checker requires an index no revision ever carried. The Next-gate cell was also asserting something false about the row it points at. It read "Hosted ten-SM completion, the full eight-tuple dry run, matching-hardware evidence, merge, and tagged publication remain pending; no published binary exists." The tag v0.0.2 EXISTS -- git tag lists v0.0.2, v0.0.2-alpha and v0.0.2-alpha1-ci-test -- and the engine-matrix row records the publication it refuses: "v0.0.2 published eight archive/checksum/provenance triplets plus two indexes from 7020de9 in run 31466516224 (26 assets)". Every fragment written into the cell is true and sourced from the tree rather than composed to satisfy a checker: - v0.0.2: the tag, plus the run and asset count on the engine-matrix row. - Windows W14-W16 implemented: windows-binary-release.md:5-7, "W14-W16 are implemented locally; native hosted Windows evidence, the merged-SHA ten-tuple dry run, prerelease publication, and the 32-asset audit remain pending", with the wave table at :390-394 assigning the three waves and stating they land in one PR. - the pending audit: the same spec at :451, "v0.0.3-pre.1 publication, attestations, and the exact 32-asset API audit are not inferred from Linux and remain required hosted gates". The single ENG-RELEASE-WINDOWS anchor the windows checker requires is added above the roadmap table, matching how engine-matrix.md and the spec already carry it. Nothing here claims Windows is published. The row stays ACTIVE, the prerelease stays pending, Windows stays preview, and no stable v0.0.3 is authorized -- the spec's own stop conditions are untouched. No checker is modified. check-windows-release-state: 2 errors -> "Windows release remains truthfully ACTIVE, pending, and unpublished"; its 6-case suite passes. check-release-binary-contract: 1 error -> "Release binary contract: OK". check-agent-record, check-site, check-surface-coverage, check-conflict-markers, check-model-checklist and check-now-current are green. FOLLOWING_AGENTS_PROTOCOL Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:codebuff/buffy [freebuff]
phantomic12
force-pushed
the
row/RELEASE-LIFECYCLE-ROADMAP
branch
from
October 1, 2026 18:54
dcd4196 to
bee3ade
Compare
This branch has not been deployed
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
3 participants
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.
What
Two release gates were red on
main, and neither was red for a reason a reader could act on: the text they require has never been in.agents/roadmap_v1.md.git log -S "v0.0.2 published eight primary archive" -- .agents/roadmap_v1.mdreturns nothing. At1de097c46the roadmap contains nov0.0.2, nov0.0.3-pre.1and noENG-RELEASE-WINDOWSanchor. The v0.0.2 facts live in theENG-RELEASE-BINARIESrow of.agents/engine-matrix.mdinstead — so the two gates disagree about which file owns the lifecycle.This is the same shape as
check-benchmark-index(#3339): a checker requiring content no revision ever carried.The cell was also asserting something false
The
REL/ROAD-V1-RELEASENext-gate cell read:The tag
v0.0.2exists —git taglistsv0.0.2,v0.0.2-alpha,v0.0.2-alpha1-ci-test— and the engine-matrix row records the publication the roadmap refuses:"no published binary exists" and "tagged publication remain pending" are both false, about the very row the roadmap points at.
What was written, and why it isn't composed-to-pass
Every fragment in the new cell is true and sourced from the tree:
v0.0.2published eight tripletswindows-binary-release.md:5-7— "W14-W16 are implemented locally; native hosted Windows evidence, the merged-SHA ten-tuple dry run, prerelease publication, and the 32-asset audit remain pending"; wave table:390-394— "W14, W15, and W16 land in one PR"v0.0.3-pre.1publication and 32-asset audit pending:451— "not inferred from Linux and remain required hosted gates"The single
ENG-RELEASE-WINDOWSanchor is added above the roadmap table, matching howengine-matrix.mdand the spec already carry it.Nothing here claims Windows is published. The row stays
ACTIVE, the prerelease stays pending, Windows stayspreview, and no stablev0.0.3is authorized — the spec's own stop conditions are untouched. No checker is modified.Gates
check-windows-release-state— 2 errors →Windows release remains truthfully ACTIVE, pending, and unpublished; its 6-case suite passescheck-release-binary-contract— 1 error →Release binary contract: OKcheck-agent-record,check-site,check-surface-coverage,check-conflict-markers,check-model-checklist,check-now-current— greencheck-commit-style,check-commit-trailers— greenNot run here:
tests/scripts/test_check_release_binary_contract.pyis a long mutation suite that exceeded the 600 s command budget in this environment, so it is unverified by this PR. The gate itself passes.check-benchmark-indexis red on this branch but is pre-existing at1de097c46and is fixed by #3339.Row:
ENG-RELEASE-BINARIES. Filed asISSUE-LOCAL-01M3JR4Y2E5TS4QGVNGF62EZF3(mirror PENDING).🤖 Generated with Codebuff
FOLLOWING_AGENTS_PROTOCOL
Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: AGENT:codebuff/buffy [freebuff]