Validate latest PR40 assembly discovery batch

[cpunion]

Fork-only Draft staging for xgo-dev/plan9asm#40, targeting its existing contribution branch. The upstream-connected head remains unchanged. This is not completed ecosystem coverage or a runtime certificate.

Current repair batch

Head 46fa4bca contains 20 batched commits after 0c61d468. The previous run 38096927749 finished naturally before publication. Fresh current-head CI and exact-version replay are required; local component passes do not promote external versions.

Repair class	Independent verification
Scalar integer lane insertion and GP/X transfers	Complete typed families, exact-width conditional native reads and result-byte effects; current/old Go, five-target objects and original-Go/LLVM live/dead numerical/fault controls
PSHUFB/VPSHUFB	Complete grammar, aliases, E4NF reads, caller/result dependencies, raw WIG and feature inference; restore legacy vector dispatch to the same lowerer, with handled rejection before IR
Mixed TEXT/DATA frame observation	Complete original Go objects/listings and immutable readonly-literal consumer; retain mutable/address/escape and source-lifetime boundaries
Scalar MOV and extension boundaries	Five native B/W/L/Q/D load spellings, volatile dead reads and result-byte effects; preserve independent compare reads, Go high-byte/REX identity and reject bare extension symbols
Packed integer add/subtract	All 16 wrapping/saturating vector names share typed lowering/read/alias and result effects; 50 related roots pass in 323s, with 144 bounded kernels and independent native/masked fault controls
Packed unsigned widening	All 12 PMOVZX/VPMOVZX names and six ratios retain exact shrinking tuples and E5 suppression; 88 related roots pass in 377s, including 78 bounded live/dead/masked kernels
Legacy downloaded-module origin	Distinct SumDB-authenticated metadata for ZIPs without go.mod; exact cache-only dependency queries preserve original language/module/source/export identity and mutation revocation
Frozen header queries	Retain the actual SDK patch in the owned query module without changing original library bytes or weakening immutable guards; 23 related planning roots pass in 313s
Ordinary frame CPP	Share complete actual-Go macro-registration AST checks between frame capture and profile evidence; 12 OS/profile configurations check real Go object values plus 72 LLVM object compilations
ARM64 min/max test ownership	Four disjoint native roots preserve all 14,600 original states and 175,200 LLVM comparisons; each family passed, and a separate guard keeps them in distinct ordinary hash shards

These repair already recognized instruction families and their formats/effects; no new-mnemonic total is claimed. Native entry/frame/backing, generated-header and unregistered effect permissions remain separate. LLVM 22 only; development Go 1.27.1, compatibility checks Go 1.20.14. No local Module Index scan or external-library compilation batch was launched.

Local gate	Outcome and limitation
Both nested CLI suites after CPP repair	Passed; plan9asmll has 258 passing roots, zero failures/skips, 1,191s
Integrated CPP/dispatcher selection	24 root tests and three shared-registration roots passed, 84s; related vet and official gate passed
Feature/dispatch follow-up	25 current-Go and three Go 1.20 roots passed; five earlier dispatcher/segment roots also passed on both SDKs
Shared actual macro-registration package	Race checks passed; 86.7% statement coverage, not overall/patch coverage approval
Frozen complete main-module run at 4856910c	Still running its remaining partitions; failures in compare negative, legacy shuffle dispatch, segment assertion and opcode-only feature fixture have separate fixes above. Original report remains failed
Current-head full CI, exact external replays and verified ledger	Not established

Component evidence cannot be credited to a different frozen full run. Rosetta/portable execution does not certify native Linux or AVX512 execution. Required cross-runtime owners remain required; earlier failures/timeouts are retained.

Completed previous CI

Policy preflight passed. All six priority shards failed; 36/42 hit the unchanged 330-minute deadline and uploaded partial reports.

Diagnostic shard	Reported versions	Passed	Failed	Source N/A	Explicit skips
5	96	38	42	16	0
29	106	53	40	12	1
36 (partial)	72	27	32	11	2
38	94	35	41	18	0
39	97	41	45	11	0
42 (partial)	63	24	30	9	0
Reported subset	528	218	230	77	3

The reports retain 2,559 ordinary translations; 5,539 versions remain unreported. The three skips are two invalid-source and one superseded-version outcome. These are old-source diagnostic outcomes, not new-head ledger credit or necessarily distinct module counts.

The strict aggregate audited reports in 1,116s, compared with 4,778s in the preceding run; the partial subsets differ, so this is an observed timing, not a controlled speedup. It correctly rejected six reports instead of 64 complete shards. Downstream module/cross completion failed for missing owner artifacts after priority failure prevented their jobs from running, not newly observed unit failures. No completeness guard or deadline was relaxed.

Remaining categories include native ABI/entry/TLS, SP/backing/alignment, raw continuation/CFG joins, FP padding and result availability. The fixes above must be replayed against their original exact versions; moving the first error is not candidate closure. Native-layout exclusions remain provisional and require review.

Discovery funnel

Committed stage	Count
Contiguous official Module Index records	18,758,534
Unique module paths recorded	965,347
Successfully inspected retained exact versions	952,900
Assembly-bearing exact versions queued	6,067
Original assembly paths inventoried	78,961
Inspection failures retained for retry	16,390

The 215 contiguous ranges cover 2025-10-01T00:00:00Z through 2026-09-22T22:43:08Z, not the complete historical index or today's head. No reliable global unique-module denominator is available. Keep the newest version per (module path, Go-semver major); v0/v1 lines and separate major paths may coexist. Deduplication never rewinds already covered ranges.

Independent, not-imported inventory: 21,257,856 index records, 1,051,454 module paths and 6,700 assembly versions. These are discoveries, not additional tested passes. cgo inventory remains outside this repository.

Assembly-evidence cohort	Passed	Failed	Pending
Stored historical v3 ledger	5	1	6,061
Current repair head	Not established	Not established	Complete matching reports required

The ledger remains complete=false, verified=false, with zero complete shards. Scan records prove inspection; assembly evidence proves its exact producer's translation/object tests. No manual promotion, mixed-revision report, N/A or skip replaces a pass. Only the automatic writer may publish fully matched evidence.

Official classification and remaining gates

Architecture	Observed Go forms	Supported	Requires context
386	57	48	9
amd64	6,736	6,688	48
arm	499	455	44
arm64	2,048	1,979	69
wasm	120	0	120
Total	9,460	9,170	290

The unchanged official gate has zero unsupported forms/parse errors. This is not complete encoder-combination, ABI or runtime coverage; context-dependent wasm forms do not mean an unsupported architecture.

The strict SDK benchmark at 9f9c1225 remains failed: 163/184 files passed and 21 failed, zero N/A/unattempted. Per-target passes are 386 29/33, amd64 48/55, arm 27/30, arm64 40/47 and wasm 19/19. Its 59 target-seconds plus three build-seconds met unchanged time budgets, but functionality did not pass. Native entry/ABIInternal/TLS, dynamic frames, native result tails and backing/alignment remain blockers.

CI retains policy preflight, all six priority shards, then other entry jobs only after priority success. All 64 shards and strict aggregates remain required, without an explicit matrix concurrency cap. Let the new run finish before another batch push.

Keep fork PR4 and upstream PR40 Draft until current-head CI, review, coverage and automatically verified ledger pass. Only then integrate into PR40's existing fork head. No upstream push, merge, approval or closure is authorized. Fork PRs 1–3 are already closed; the separate llgo integration PR259 remains necessary.

[codecov]

Codecov Report

❌ Patch coverage is 90.60606% with 93 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (codex/expand-ecosystem-corpus-20260913@b93666f). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
amd64_frame_vector.go	75.18%	33 Missing ⚠️
amd64_raw_enqueue.go	90.41%	7 Missing ⚠️
arm64_raw_pool_mask_relation.go	88.13%	7 Missing ⚠️
arm64_raw_pool_stride.go	92.10%	6 Missing ⚠️
go_translate.go	89.09%	6 Missing ⚠️
amd64_lower_enqueue.go	88.88%	5 Missing ⚠️
arm64_raw_pool_relational_loop.go	87.09%	4 Missing ⚠️
amd64_blocks.go	50.00%	3 Missing ⚠️
amd64_lower_branch.go	75.00%	3 Missing ⚠️
arm64_raw_pool_affine.go	86.95%	3 Missing ⚠️
... and 11 more

Additional details and impacted files

@@                            Coverage Diff                            @@
##             codex/expand-ecosystem-corpus-20260913       #4   +/-   ##
=========================================================================
  Coverage                                          ?   89.93%           
=========================================================================
  Files                                             ?      661           
  Lines                                             ?    75174           
  Branches                                          ?        0           
=========================================================================
  Hits                                              ?    67611           
  Misses                                            ?     7558           
  Partials                                          ?        5           

Flag	Coverage Δ
unittests	89.93% <90.60%> (?)
windows	89.93% <90.60%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.