Triage 0.70 scope: 341 → 121 in 0.70.0; 0.69.x leftovers, unmilestoned and unlabelled dispositions (proposal, list-first) #4159

Activity

1. noahgift

   added this to the 0.70.0 milestone on Sep 24, 2026
2. noahgift commented on Sep 24, 2026

   

   noahgift

   on Sep 24, 2026

   MemberAuthor

   More actions

   Part 1 executed (aprender-dd, 2026-09-24). Read back from GitHub:

   • 4/4 closes and 65/65 moves, each with its reason comment;
   • the 4 kind:* labels created;
   • 139/139 labels applied: code 111, measurement 17, triage 8, docs 3.

   The 0.70 → 0.71–0.74 scope moves are HELD for the operator.

   The 72 unlabelled issues the derived rule cannot decide, left for a later pass (no guessing).
   Each needs a kind: of code / triage / docs / measurement from its owner or the cop:

   #	milestone	title
   #3343	0.70.0	ci_resolve_dirty binds merge.roadmap.driver to a path no .gitattributes selects — the id-keyed roadm
   #3345	0.70.0	A clean-room runner lost its pinned 1.93.0 toolchain mid-build (ENOENT on rustc) and a live build lo
   #3531	0.70.0	PACK-001 first main run: three shards of one push chose three tiers (reflog-derived base) and partit
   #3742	0.74.0	tokenizer parity: the APR/HF encode paths are sibling copies #3726 did not touch — aprender-core Bpe
   #3747	0.69.1	#3568 PR 1/4: the TokenConstraint hook and the llguidance engine behind it — no generation loop wire
   #3777	0.70.0	#3745 S2: cli-surface extractor + derived release cells (pairwise) — release-readiness-v1 targets th
   #3876	0.71.0	ladder receipt: bytes describes the symlink, sha256 describes the model — one row, two objects
   #3878	0.71.0	ladder receipt: a dangling symlink records "sha256": "" — an identity claim of nothing
   #3901	0.71.0	a ladder row red only on serve prints reason 'unknown' — #3886 moved the verdict and left the explan
   #3904	0.70.0	#3872 removed the four slices someone FOUND; nobody enumerated the surface — here it is, and the 5th
   #3916	0.70.0	#3904 enumerated the surface I could SEE: its scan is Python-shaped, and Rust truncates with .take(N
   #3918	0.74.0	detect_format_from_name's ordered substring table has produced three defects — one a correct fix tha
   #3919	0.71.0	The thinking-ON leg exists twice and has now drifted twice: the hybrid copy takes neither #3907's bu
   #3920	0.71.0	#3911's embedded tokenizer splits every ChatML control token into six pieces — 32 ids where the refe
   #3921	—	P0: chat/code judged by rc and serve by http — a verb emitting 'zombie zombie zombie' is recorded as
   #3930	—	Qwen3NoThink prefills \n\n where the model's own template uses \n\n\n\
   #3936	0.71.0	the ladder's serve /health window is a fixed 90s, so the 27B is red for load time while passing capa
   #3943	—	ladder_serve_teardown reads 'not answering /health' as 'server gone', so a still-loading server is r
   #3945	0.74.0	Enumerated conformance table: derive (qtype × op × backend) proof obligations from the dispatch regi
   #3947	0.71.0	aprender-core has no IQ dequantizer, so tensor_contract cannot inspect three release-blocking models
   #3950	—	IQ2_XXS (ggml 16) has no CUDA GEMV — 0.69.1 release blocker
   #3951	—	Qwen3.5-0.8B-IQ4_XS never closes its think block while Q4_K_M does: CUDA IQ4 GEMV or real quant dama
   #3952	0.70.0	CRUX: add vLLM as a pre-release comparator engine (plugin driver; lambda x86/4090 + gx10 aarch64/GB1
   #3953	—	IQ2_S (ggml 22) has no CUDA GEMV — 0.69.1 release blocker
   #3955	0.72.0	False provenance: chat banner claims 'fused Q4K, F2-validated' on BF16 with F2 SKIPPED; chat envelop
   #3959	0.74.0	Dedupe IQ decoders: aprender-serve onto trueno_quant::iq (one decoder, one oracle)
   #3960	—	Q2_K (ggml 10) has no CUDA GEMV — 0.69.1 release blocker
   #3961	—	golden thinking-ON leg discards its backend (_used_gpu) and passes an empty as think
   #3962	—	0.69.1 BLOCKER: CRUX producer — per-verb prompt sets (incl. hard prompts) + code verb + hf/vllm quor
   #3963	—	IQ3_XXS (ggml 18) has no CUDA GEMV — 0.69.1 release blocker
   #3968	0.71.0	GPU whitelist admits a quant TYPE on one model's shapes — admission must key on per-shape device con
   #3977	0.71.0	0.70.0: qwen35moe (Qwen3.5-35B-A3B hybrid SSM MoE) CUDA forward
   #3980	0.72.0	Batch serving F2 GPU validation has never run (empty probe) — needs a representative probe at batch-
   #3987	—	qwen3moe CUDA reaches run+qa only: chat rc 8, serve 501/500, code rc 1 — serve loads no mapped model
   #3990	0.73.0	Qwen2.5 chat template drops the default system prompt (26 vs llama.cpp's 47 tokens) — apr 1/18 vs ll
   #4008	—	arch constraints: qwen35moe (Qwen3.5-35B-A3B) has is_moe=false; replace #3992's substring MoE predic
   #4015	0.70.0	ladder serve wait: #3943's stall rule counts time blocked in flock as no progress, a false 'stalled'
   #4017	0.73.0	apr --gpu holds ~2.3× the GGUF size in host RAM (27B Q4_K_M: 36 GiB RSS for a 15.6 GiB file) — llama
   #4021	0.70.0	CRUX judge: thinking-ON cells must not be judged on closure alone — use answer-when-both-cl
   #4027	—	Operator standing rules → global ~/.claude/CLAUDE.md (cop verbatim relays; quorum Opus fallback unti
   #4028	0.70.0	aprender-serve lib tests accumulate memory across tests in one process: 30 GB peak single-threaded,
   #4030	0.71.0	0.71: qwen35-0.8b-q4km golden thinking-ON never closes on apr (greedy near-tie path, no localized de
   #4031	0.70.0	Load-time CRUX certification status for every model (certified / not certified / known issue)
   #4032	0.69.1	0.69.1 hotfix: load-time warning for Qwen3.5-0.8B (known issue #4030)
   #4034	0.70.0	ladder: run a cell's CPU and CUDA lanes concurrently (the CPU lane doesn't hold the GPU lock)
   #4037	0.70.0	certification: carry a cell's receipt forward when the diff since its apr_sha doesn't touch that cel
   #4038	0.70.0	ladder: CPU==GPU think-block parity on a fixed prefix (e.g. 256 tokens); full 2048-token budget on G
   #4039	0.70.0	sweep: order cells so CRUX-certified models run first (CRUX can start early), then largest first to
   #4040	0.70.0	nightly full ladder + CRUX on main; the release sweep runs only the delta since the last green night
   #4043	0.70.0	test_forward_batched_graphed_replay passes when graph capture FAILS (falls back silently); parallel
   #4049	—	certification: per-architecture carry-forward — a diff invalidates only the cells whose inference pa
   #4052	0.70.0	ladder: short-circuit a cell's verbs once F10 RED-UNSUPPORTED is proven by name (not-run, never pass
   #4053	0.70.0	ladder: WORK dir off /tmp onto NVMe + free-space preflight (gx10 ENOSPC at 27B) (#4033)
   #4054	0.70.0	Wire aprender-serve --test tokenizer_nonascii_4005 into CI (dark test since 9684a7a, #4005)
   #4059	0.70.0	binary-debt #2: the KEPT binaries' CLI tests are dark in CI — pv (17 files), aprender-profile (13),
   #4063	0.72.0	binary-debt #6: aprender-train-* → apr (distill / finetune --plan / inspect / bench); delete train-s
   #4065	0.74.0	binary-debt #8: sunset the 13 pre-monorepo crates.io names — a final release pointing at aprender, a
   #4087	0.70.0	qa tensor_contract: cache keyed by model sha + checker-source hash (#4033 lever; 19–46% of qa)
   #4090	0.70.0	ladder: serve health wait reads a GPU-LOCK WAIT as a stalled server (no log, no CPU) — a foreign loc
   #4111	—	provable-contracts facades stuck at 0.4.0 (-> aprender-contracts ^0.64): tier-14 publishes nothing
   #4124	0.70.0	PVL EV-5a orphan group ProvableContracts.Theorems: 55 module(s) outside the root import cone
   #4125	0.70.0	PVL EV-5a orphan group ProvableContracts.Defs: 4 module(s) outside the root import cone
   #4126	0.70.0	ladder: serve probe writes "http":000 (invalid JSON) for a timed-out route, so the whole serve recor
   #4127	0.70.0	Published aprender pulls two unmaintained proc-macro crates (tabled 0.16 → proc-macro-error; validat
   #4130	0.70.0	Published tarballs of aprender-core, aprender-orchestrate, aprender-cbtop, aprender-test-showcase sh
   #4131	0.70.0	ladder: every run sha256s the whole 109 GB inventory (~473 s on lambda, ~60% of a --only 9B run); ca
   #4140	0.70.0	#4129 on the 0.69.1 lineage: apr-cli and aprender-orchestrate tests include #3990 fixtures / thinkin
   #4153	0.70.0	clippy debt ratchet 1/5 (0.70): apr-cli + aprender-core + aprender-serve residual (~522)
   #4154	0.71.0	clippy debt ratchet 2/5 (0.71): qa-runner, first half (~560)
   #4155	0.72.0	clippy debt ratchet 3/5 (0.72): qa-runner, second half (~559)
   #4156	0.73.0	clippy debt ratchet 4/5 (0.73): test-cli + half the small crates (~510)
   #4157	0.74.0	clippy debt ratchet 5/5 (0.74): aprender-graph + remaining small crates (~376) → 0
3. noahgift commented on Sep 24, 2026

   

   noahgift

   on Sep 24, 2026

   MemberAuthor

   More actions

   The 72 undecidables are labelled with the cop's defaults (aprender-dd, 2026-09-24): 71 applied and read back; #4032 skipped because it was closed in part 1.

   • kind: from the title's own verb (fix → code, measure/bench → measurement, docs/spec → docs), otherwise kind:triage.
   • milestone: every one already had a milestone, so none moved. The backlog amendment changes only the fallback.

   Coin-flips: 68 of 71. These titles are STATEMENTS of a defect ("X does Y wrong"), not imperatives, so the verb rule almost never matched and they fell to kind:triage. Most read as code defects, so the by-exception pass will have many rows to flip.

   #	milestone	kind applied	coin-flip	title
   #3343	0.70.0	triage	yes	ci_resolve_dirty binds merge.roadmap.driver to a path no .gitattributes selects — the id-keyed
   #3345	0.70.0	triage	yes	A clean-room runner lost its pinned 1.93.0 toolchain mid-build (ENOENT on rustc) and a live bui
   #3531	0.70.0	triage	yes	PACK-001 first main run: three shards of one push chose three tiers (reflog-derived base) and p
   #3742	0.74.0	triage	yes	tokenizer parity: the APR/HF encode paths are sibling copies #3726 did not touch — aprender-cor
   #3747	0.72.0	triage	yes	#3568 PR 1/4: the TokenConstraint hook and the llguidance engine behind it — no generation loop
   #3777	0.70.0	triage	yes	#3745 S2: cli-surface extractor + derived release cells (pairwise) — release-readiness-v1 targe
   #3876	0.71.0	triage	yes	ladder receipt: bytes describes the symlink, sha256 describes the model — one row, two objects
   #3878	0.71.0	triage	yes	ladder receipt: a dangling symlink records "sha256": "" — an identity claim of nothing
   #3901	0.71.0	triage	yes	a ladder row red only on serve prints reason 'unknown' — #3886 moved the verdict and left the e
   #3904	0.70.0	triage	yes	#3872 removed the four slices someone FOUND; nobody enumerated the surface — here it is, and th
   #3916	0.70.0	triage	yes	#3904 enumerated the surface I could SEE: its scan is Python-shaped, and Rust truncates with .t
   #3918	0.74.0	triage	yes	detect_format_from_name's ordered substring table has produced three defects — one a correct fi
   #3919	0.71.0	triage	yes	The thinking-ON leg exists twice and has now drifted twice: the hybrid copy takes neither #3907
   #3920	0.71.0	triage	yes	#3911's embedded tokenizer splits every ChatML control token into six pieces — 32 ids where the
   #3921	0.72.0	triage	yes	P0: chat/code judged by rc and serve by http — a verb emitting 'zombie zombie zombie' is record
   #3930	0.73.0	triage	yes	Qwen3NoThink prefills \n\n where the model's own template uses \n\n</thin
   #3936	0.71.0	triage	yes	the ladder's serve /health window is a fixed 90s, so the 27B is red for load time while passing
   #3943	0.74.0	triage	yes	ladder_serve_teardown reads 'not answering /health' as 'server gone', so a still-loading server
   #3945	0.74.0	triage	yes	Enumerated conformance table: derive (qtype × op × backend) proof obligations from the dispatch
   #3947	0.71.0	triage	yes	aprender-core has no IQ dequantizer, so tensor_contract cannot inspect three release-blocking m
   #3950	0.73.0	triage	yes	IQ2_XXS (ggml 16) has no CUDA GEMV — 0.69.1 release blocker
   #3951	0.73.0	triage	yes	Qwen3.5-0.8B-IQ4_XS never closes its think block while Q4_K_M does: CUDA IQ4 GEMV or real quant
   #3952	0.70.0	code		CRUX: add vLLM as a pre-release comparator engine (plugin driver; lambda x86/4090 + gx10 aarch6
   #3953	0.73.0	triage	yes	IQ2_S (ggml 22) has no CUDA GEMV — 0.69.1 release blocker
   #3955	0.72.0	triage	yes	False provenance: chat banner claims 'fused Q4K, F2-validated' on BF16 with F2 SKIPPED; chat en
   #3959	0.74.0	triage	yes	Dedupe IQ decoders: aprender-serve onto trueno_quant::iq (one decoder, one oracle)
   #3960	0.73.0	triage	yes	Q2_K (ggml 10) has no CUDA GEMV — 0.69.1 release blocker
   #3961	0.71.0	triage	yes	golden thinking-ON leg discards its backend (_used_gpu) and passes an empty as
   #3962	0.72.0	triage	yes	0.69.1 BLOCKER: CRUX producer — per-verb prompt sets (incl. hard prompts) + code verb + hf/vllm
   #3963	0.73.0	triage	yes	IQ3_XXS (ggml 18) has no CUDA GEMV — 0.69.1 release blocker
   #3968	0.71.0	triage	yes	GPU whitelist admits a quant TYPE on one model's shapes — admission must key on per-shape devic
   #3977	0.71.0	triage	yes	0.70.0: qwen35moe (Qwen3.5-35B-A3B hybrid SSM MoE) CUDA forward
   #3980	0.72.0	triage	yes	Batch serving F2 GPU validation has never run (empty probe) — needs a representative probe at b
   #3987	0.74.0	triage	yes	qwen3moe CUDA reaches run+qa only: chat rc 8, serve 501/500, code rc 1 — serve loads no mapped
   #3990	0.73.0	triage	yes	Qwen2.5 chat template drops the default system prompt (26 vs llama.cpp's 47 tokens) — apr 1/18
   #4008	0.74.0	triage	yes	arch constraints: qwen35moe (Qwen3.5-35B-A3B) has is_moe=false; replace #3992's substring MoE p
   #4015	0.70.0	triage	yes	ladder serve wait: #3943's stall rule counts time blocked in flock as no progress, a false 'sta
   #4017	0.73.0	triage	yes	apr --gpu holds ~2.3× the GGUF size in host RAM (27B Q4_K_M: 36 GiB RSS for a 15.6 GiB file) —
   #4021	0.70.0	triage	yes	CRUX judge: thinking-ON cells must not be judged on closure alone — use answer-when-bo
   #4027	0.74.0	triage	yes	Operator standing rules → global ~/.claude/CLAUDE.md (cop verbatim relays; quorum Opus fallback
   #4028	0.70.0	triage	yes	aprender-serve lib tests accumulate memory across tests in one process: 30 GB peak single-threa
   #4030	0.71.0	triage	yes	0.71: qwen35-0.8b-q4km golden thinking-ON never closes on apr (greedy near-tie path, no localiz
   #4031	0.70.0	triage	yes	Load-time CRUX certification status for every model (certified / not certified / known issue)
   #4034	0.70.0	code		ladder: run a cell's CPU and CUDA lanes concurrently (the CPU lane doesn't hold the GPU lock)
   #4037	0.70.0	triage	yes	certification: carry a cell's receipt forward when the diff since its apr_sha doesn't touch tha
   #4038	0.70.0	triage	yes	ladder: CPU==GPU think-block parity on a fixed prefix (e.g. 256 tokens); full 2048-token budget
   #4039	0.70.0	triage	yes	sweep: order cells so CRUX-certified models run first (CRUX can start early), then largest firs
   #4040	0.70.0	triage	yes	nightly full ladder + CRUX on main; the release sweep runs only the delta since the last green
   #4043	0.70.0	triage	yes	test_forward_batched_graphed_replay passes when graph capture FAILS (falls back silently); para
   #4049	0.74.0	triage	yes	certification: per-architecture carry-forward — a diff invalidates only the cells whose inferen
   #4052	0.70.0	triage	yes	ladder: short-circuit a cell's verbs once F10 RED-UNSUPPORTED is proven by name (not-run, never
   #4053	0.70.0	triage	yes	ladder: WORK dir off /tmp onto NVMe + free-space preflight (gx10 ENOSPC at 27B) (#4033)
   #4054	0.70.0	code		Wire aprender-serve --test tokenizer_nonascii_4005 into CI (dark test since 9684a7a, #4005)
   #4059	0.70.0	triage	yes	binary-debt #2: the KEPT binaries' CLI tests are dark in CI — pv (17 files), aprender-profile (
   #4063	0.72.0	triage	yes	binary-debt #6: aprender-train-* → apr (distill / finetune --plan / inspect / bench); delete tr
   #4065	0.74.0	triage	yes	binary-debt #8: sunset the 13 pre-monorepo crates.io names — a final release pointing at aprend
   #4087	0.70.0	triage	yes	qa tensor_contract: cache keyed by model sha + checker-source hash (#4033 lever; 19–46% of qa)
   #4090	0.70.0	triage	yes	ladder: serve health wait reads a GPU-LOCK WAIT as a stalled server (no log, no CPU) — a foreig
   #4111	0.73.0	triage	yes	provable-contracts facades stuck at 0.4.0 (-> aprender-contracts ^0.64): tier-14 publishes noth
   #4124	0.70.0	triage	yes	PVL EV-5a orphan group ProvableContracts.Theorems: 55 module(s) outside the root import cone
   #4125	0.70.0	triage	yes	PVL EV-5a orphan group ProvableContracts.Defs: 4 module(s) outside the root import cone
   #4126	0.70.0	triage	yes	ladder: serve probe writes "http":000 (invalid JSON) for a timed-out route, so the whole serve
   #4127	0.70.0	triage	yes	Published aprender pulls two unmaintained proc-macro crates (tabled 0.16 → proc-macro-error; va
   #4130	0.70.0	triage	yes	Published tarballs of aprender-core, aprender-orchestrate, aprender-cbtop, aprender-test-showca
   #4131	0.70.0	triage	yes	ladder: every run sha256s the whole 109 GB inventory (~473 s on lambda, ~60% of a --only 9B run
   #4140	0.70.0	triage	yes	#4129 on the 0.69.1 lineage: apr-cli and aprender-orchestrate tests include #3990 fixtures / th
   #4153	0.70.0	triage	yes	clippy debt ratchet 1/5 (0.70): apr-cli + aprender-core + aprender-serve residual (~522)
   #4154	0.71.0	triage	yes	clippy debt ratchet 2/5 (0.71): qa-runner, first half (~560)
   #4155	0.72.0	triage	yes	clippy debt ratchet 3/5 (0.72): qa-runner, second half (~559)
   #4156	0.73.0	triage	yes	clippy debt ratchet 4/5 (0.73): test-cli + half the small crates (~510)
   #4157	0.74.0	triage	yes	clippy debt ratchet 5/5 (0.74): aprender-graph + remaining small crates (~376) → 0
4. noahgift commented on Sep 24, 2026

   

   noahgift

   on Sep 24, 2026

   MemberAuthor

   More actions

   Cop rulings applied and read back (aprender-dd, 2026-09-24):

   • 26/26 no-theme tooling items moved 0.71–0.74 → backlog, each with a comment citing the operator's backlog rule.
   • The 68 coin-flip kinds were re-ruled. Before: triage 68. After: code 22, measurement 10, docs 1, triage 35 (33 changed). The earlier 69 counted 0.69.1 hotfix: load-time warning for Qwen3.5-0.8B (known issue #4030) #4032, closed in part 1.
   • The re-planned 0.70 scope cut stays HELD for the operator's GO.
5. noahgift commented on Sep 24, 2026

   

   noahgift

   on Sep 24, 2026

   MemberAuthor

   More actions

   Epic membership, measured 2026-09-24 (aprender-dd). Operator rule: a milestone is an epic only if an epic issue lists its members.

   milestone	epic	open members	listed before	UNLISTED before	action
   0.70.0	#3998	359	5+0 via sub-epic	353	NONE: held for the scope-cut GO
   0.71.0	#3994	137	4+0 via sub-epic	132	+132 appended to #3994's checklist, read back
   0.72.0	#4000	71	5+3 via sub-epic	62	+62 appended to #4000's checklist, read back
   0.73.0	#3999	101	0+0 via sub-epic	100	+100 appended to #3999's checklist, read back
   0.74.0	#4001	33	5+0 via sub-epic	27	+27 appended to #4001's checklist, read back
   0.75.0	#4002	19	1+0 via sub-epic	17	+17 appended to #4002's checklist, read back
   0.76.0	none	49	0+0 via sub-epic	49	epic CREATED #4164, all 49 listed

   Listed = the epic names it, or names a sub-epic that names it. The checklists are marked members-derived:v1 so a later re-derivation replaces them instead of stacking.

   0.70.0's 353 unlisted members, not written; the scope cut produces the final list:

   #2307 #2373 #2463 #2470 #2473 #2482 #2492 #2494 #2500 #2503 #2526 #2530 #2556 #2585 #2590 #2614 #2615 #2616 #2646 #2648 #2653 #2655 #2699 #2701 #2704 #2750 #2757 #2766 #2775 #2784 #2797 #2798 #2808 #2814 #2856 #2877 #2878 #2881 #2889 #2890 #2891 #2892 #2894 #2896 #2897 #2898 #2899 #2900 #2901 #2902 #2911 #2912 #2913 #2914 #2915 #2916 #2917 #2918 #2931 #2978 #2979 #2983 #2984 #2989 #2992 #2993 #2995 #2996 #2997 #2999 #3014 #3015 #3018 #3029 #3043 #3054 #3055 #3058 #3062 #3081 #3132 #3133 #3140 #3141 #3142 #3143 #3144 #3170 #3171 #3172 #3173 #3176 #3177 #3182 #3183 #3190 #3203 #3204 #3206 #3217 #3218 #3251 #3254 #3258 #3261 #3269 #3279 #3282 #3284 #3286 #3291 #3293 #3298 #3304 #3306 #3324 #3343 #3345 #3353 #3397 #3398 #3399 #3416 #3446 #3454 #3457 #3458 #3459 #3460 #3462 #3464 #3466 #3478 #3493 paiml/infra#1144 #3522 #3531 #3543 #3544 #3546 #3554 #3559 #3560 #3561 #3562 #3563 #3567 #3569 #3577 #3579 #3584 #3585 #3586 #3587 #3591 #3592 #3593 #3594 #3597 #3603 #3607 #3610 #3611 paiml/infra#1145 #3615 #3624 #3625 #3628 #3635 #3636 #3639 #3643 #3646 #3649 #3650 #3652 #3667 #3668 #3676 #3679 #3680 #3683 #3684 #3686 #3687 #3690 #3694 #3697 #3699 #3701 #3703 #3705 #3708 #3709 #3712 #3713 #3714 #3715 #3717 #3722 #3731 #3732 #3739 #3740 #3741 #3745 #3748 #3749 #3761 #3765 #3768 #3770 #3771 #3774 #3777 #3778 #3788 #3790 #3792 #3795 #3797 #3805 #3806 #3809 #3818 #3821 #3829 #3831 #3833 #3834 #3836 #3837 #3839 #3841 #3844 #3845 #3847 #3854 #3855 #3856 #3857 #3863 #3864 #3865 #3874 #3879 #3887 #3902 #3904 #3909 #3912 #3916 #3917 #3952 #3956 #3964 #3966 #3972 #3985 #3989 #3996 #4015 #4020 #4021 #4025 #4026 #4028 #4031 #4033 #4034 #4036 #4037 #4038 #4039 #4040 #4041 #4043 #4044 #4045 #4047 #4048 #4050 #4051 #4052 #4053 #4054 #4055 #4056 #4057 #4058 #4059 #4060 #4061 #4067 #4068 #4069 #4070 #4071 #4072 #4073 #4074 #4075 #4076 #4077 #4078 #4079 #4080 #4081 #4082 #4083 #4084 #4085 #4086 #4087 #4088 #4089 #4090 #4096 #4099 #4101 #4102 #4105 #4108 #4112 #4113 #4114 #4117 #4118 #4120 #4122 #4124 #4125 #4126 #4127 #4128 #4129 #4130 #4131 #4132 #4133 #4134 #4140 #4143 #4145 #4148 #4149 #4151 #4152 #4153 #4158 #4159 #4160 #4162

   Empty open milestones, for the cop:

   • 0.66.0: CLOSED (0 open; released long ago).
   • CRUX Phase 3: 0 open / 9 closed, open. Candidate to close.
   • APR-CLUSTER-001: 0 open / 0 closed, open. Candidate to close.
   • 0.69.0: 0 open / 48 closed, open. Candidate to close.
   • 0.69.1: 0 open / 38 closed, open. Candidate to close.
   • Inference dispatch & architecture consolidation: 1 open / 0 closed, open. Candidate to close.
   • Note: the dispatch milestone's epic EPIC: Inference dispatch & architecture consolidation (dedicated cycle) #3423 is itself milestoned in 0.74.0 and is listed there. The milestone has no open members of its own.
6. noahgift commented on Sep 24, 2026

   

   noahgift

   on Sep 24, 2026

   MemberAuthor

   More actions

   Correction to the comment above (aprender-dd): the dispatch milestone's "1 open" is PR #3424 (spec PP-ARCH-001), not an issue. The milestone API counts PRs, and my member count was issues only. So that milestone has 0 open issues and one open spec PR. It is not empty, so it is not a close candidate while #3424 is open.
7. noahgift commented on Sep 24, 2026

   

   noahgift

   on Sep 24, 2026

   MemberAuthor

   More actions

   MUST-row sweep (read-only, then act), aprender-dd 2026-09-24. For each 0.70 MUST row: merged PRs (gh pr list --state merged --search) and origin/main commits referencing it; a fix-claiming candidate had its named guard / self-test run on a clean origin/main worktree.

   Result: 76 rows. 1 already CLOSED (#3561, fixed e6f77c9). 0 new CLOSE. 2 PARTIAL (commented). 73 still-open: no fix-claiming commit; 8 of them have commits that only reference them.

   issue	candidate fix	ancestor of main?	guard result	verdict
   #3561	e6f77c9 / PR #3614	yes (and v0.69.1)	check_t2_preflight_verdict PASS; self-test PASSED	CLOSED earlier today
   #2701	—	—	—	still-open
   #3140	—	—	—	still-open
   #3141	—	—	—	still-open
   #3142	—	—	—	still-open
   #3258	—	—	—	still-open
   #3269	refs only: 5f9bd93 +1	—	—	still-open
   #3298	—	—	—	still-open
   #3459	6a657fc / PR #3617 (part 1)	yes	check_tag_step_gated.sh self-test PASSED	partial: part 2 open (f5)
   #3543	refs only: 8c7822f	—	—	still-open
   #3544	—	—	—	still-open
   #3559	refs only: 82c64c6	—	—	still-open
   #3560	refs only: 82c64c6	—	—	still-open
   #3567	82c64c6 / PR #3633	yes	guard PASS (lambda, pin 0.69.1); self-test 19/19	partial: ask 1 (yoga has no pv; gx10 pin 0.68.2) left
   #3569	refs only: 5f9bd93 +3	—	—	still-open
   #3577	refs only: 5f9bd93 +2	—	—	still-open
   #3594	—	—	—	still-open
   #3610	refs only: 5f9bd93 +1	—	—	still-open
   #3611	—	—	—	still-open
   #3624	—	—	—	still-open
   #3679	refs only: 5f9bd93 +1	—	—	still-open
   #3687	—	—	—	still-open
   #3809	—	—	—	still-open
   #3821	—	—	—	still-open
   #3844	—	—	—	still-open
   #3847	—	—	—	still-open
   #3856	—	—	—	still-open
   #3857	—	—	—	still-open
   #3863	—	—	—	still-open
   #3972	—	—	—	still-open
   paiml/infra#1146	—	—	—	still-open
   #3997	—	—	—	still-open
   #4031	—	—	—	still-open
   #4033	—	—	—	still-open
   #4034	—	—	—	still-open
   #4036	—	—	—	still-open
   #4037	—	—	—	still-open
   #4038	—	—	—	still-open
   #4039	—	—	—	still-open
   #4040	—	—	—	still-open
   #4045	—	—	—	still-open
   #4047	—	—	—	still-open
   #4051	—	—	—	still-open
   #4052	—	—	—	still-open
   #4053	—	—	—	still-open
   #4057	—	—	—	still-open
   #4058	—	—	—	still-open
   #4069	—	—	—	still-open
   #4070	—	—	—	still-open
   #4071	—	—	—	still-open
   #4072	—	—	—	still-open
   #4073	—	—	—	still-open
   #4074	—	—	—	still-open
   #4075	—	—	—	still-open
   #4076	—	—	—	still-open
   #4077	—	—	—	still-open
   #4078	—	—	—	still-open
   #4079	—	—	—	still-open
   #4080	—	—	—	still-open
   #4081	—	—	—	still-open
   #4082	—	—	—	still-open
   #4083	—	—	—	still-open
   #4084	—	—	—	still-open
   #4087	—	—	—	still-open
   #4088	—	—	—	still-open
   #4102	—	—	—	still-open
   #4112	—	—	—	still-open
   #4117	—	—	—	still-open
   #4118	—	—	—	still-open
   #4122	—	—	—	still-open
   #4124	—	—	—	still-open
   #4125	—	—	—	still-open
   #4131	—	—	—	still-open
   #4151	—	—	—	still-open
   #4152	—	—	—	still-open
   #4153	—	—	—	still-open
8. noahgift

   added

   kind:triageEpic, tracker, decision or hygiene work (derived rule, #4159)Epic, tracker, decision or hygiene work (derived rule, #4159)

   on Sep 24, 2026
9. noahgift commented on Sep 24, 2026

   

   noahgift

   on Sep 24, 2026

   MemberAuthor

   More actions

   Closed by the 0.70 triage (aprender-dd, operator "go .70" relayed 2026-09-24): the triage this tracked has landed; per-item decisions are in the 0.70.0 / 0.71.0 / backlog milestones.
10. noahgift

    closed this as completedon Sep 24, 2026