-
Contributor (2026-11-07): runtime Cargo builds now resolve relative
CARGO_TARGET_DIRfrom the caller, matching staticlib discovery. Four CLI regressions pass (two failed before); real Cargo smoke passes all four path modes and fresh reuse. Pinned macOS validation on4cda40a: workspace 2945/2945, corpus 135/135, differential 45/45 + 90/90 and guards pass. Hosted CI/review pending; see PLAN.md. -
Contributor (2026-11-07):
hyb buildpreserves unmarked/unreadable Go sidecars before compilation; marked outputs remain reusable, while legacy unmarked output requires explicit migration. Six CLI regressions, pinned macOS workspace 2947/2947, corpus 135/135 and differential 45/45 + 90/90 pass. Hosted CI/review pending; see PLAN.md. -
Contributor (2026-11-06): triage board updates validate the configured issue or one trusted discovery result; 12 offline regressions pass. Baseline overwrote a marker-quoting issue. No live writes or compiler tests; CI/review pending. See PLAN.md.
-
Contributor (2026-11-05): default Actions watch selects the heavy workflow before pagination; three focused regressions pass (two failed before). The full Linux job suite passes 95/95 with no skips; earlier Windows environment limitations remain recorded in PLAN.md. Hosted CI/review pending.
-
Contributor (2026-11-01): PR #2400 integrated with main
7e770ea; pinned Linux workspace 2097/2097, corpus 130/130, differential 42/42 + 88/88 and all 12 local stages pass. Tested code/lockfile unchanged by final documentation. Hosted CI for the refreshed head, review and any award remain pending. See PLAN.md. -
Contributor (2026-11-06): claims-link checker now resolves angle-delimited Markdown evidence paths (including spaces) and checks
.hyverdicts; 11 focused tests and repository guards pass. Hosted CI, review, and any bounty award remain pending. See PLAN.md. -
Contributor #1165: range bounds spelled as different names of one Go integer type no longer raise a false HYB-E2009 (shared canonicaliser in
hy-check+ regression pin); workspace/corpus/differential/CI remain delivery gates. -
Contributor #2199: a store whose target is a type name (
n string = "str",q.n int = 5) is rejected as HYB-E1002 (Row 13 BUILDABLE) — the #778 check now also covers assignment targets; declarations and plain stores unchanged; 3 new tests intype_name_in_value_position.rs(16/16 green). -
Contributor #2320:
collect_callee_const_return_rootsno longer fails open — aconstmap returned through five wrappers (m := h5()) or through a fn value (cb := leaf; m := cb()) is rejected as HYB-E3004 (spawn) + HYB-E3006 (element write) instead ofOK (0 diagnostics)with ago run -racedata race in 10/10 runs; past the walk bound and for a callee whose body the file cannot see, the walk reports every may-share file-levelconst, gated on the result being able to share (a chain returning a plain value keeps its old verdict) (SPEC-LANGUAGE O5.6, O4.9). 11 new tests incallee_const_return_roots_2320.rs, 5 of them fail pre-fix; corpus 127/127 (39/2/86) and census/markers/registration guards green; no corpus verdict changed. -
Contributor #1598 follow-up (2026-10-24): specialized smoke gate also requires rejection exit 1 and an exact diagnostic token. Eight new process regressions; four fail before, sixteen combined pass after. New-head compiler CI/review pending.
-
Contributor (2026-11-06): installer stages and rolls back runtime/CLI payloads on copy or promotion failure; offline installer suite passes 9/9. Hosted CI, review, and any bounty award remain pending. See PLAN.md.
-
Contributor candidate (2026-10-24): author-stage repro evidence distinguishes checker failure from rejection and requires an exact diagnostic token. Eight subprocess/Git regressions, baseline 4 failures. No compiler soundness claim; see PLAN.md. Upstream review/CI pending.
-
Contributor #1347: composite receive operands (
<-chs[0],<-chs[i:],<-append(chs, ch)[0],<-outer[0][0]) now resolve to their channel, so a receive from a send-only channel is rejected (HYB-E2008); regression pin inrecv_composite_channel_operands.rs; workspace/corpus/differential/CI remain delivery gates. -
Contributor #1347-sibling (2026-10-31): locals bound from a composite (
sub := append(chs, ch)) now bind the projectedty, so<-sub[0]in a type position rejects HYB-E2008 on send-only channels instead of passing with 0 diagnostics (the "known remaining sibling" NOTE in recv_composite_channel_operands.rs is closed); HYB-E2008 tier0_rule_id BUILDABLE; corpus 129; pin in recv_local_composite_binding.rs. -
AUTOMATION (session-35): the daily human-chore is gone — triage-board.yml now runs
triage_bot.py approve --max 60every 04:30 UTC after the board post (board FIRST: run 39897148251 proved the approve POSTs drain the shared per-repo GITHUB_TOKEN hour pool and RateGuard kills the scan otherwise; approve is best-effort with continue-on-error) · session-side sweep cleared 172 pending action_required runs (all 3 per-PR workflows, 0 left at 03:56Z) ·triage_bot.Gh.reqnow retries rate-limit 403s with 5/20/45s backoff (GitHub secondary limits surface as 403, not 429; shepherd.py inherits via the shared Gh class) · BOT AUDIT: board #940 fresh, stale+rebase green on schedule, autoland gates proven honest (758/827/929/905 landed via session-side autoland.py per the session-34 GH006 finding — the workflow push transport remains the missing infra: bundle→branch→PR→checks→auto-merge) -
LANDED (session-35, Pareto round 6): #1095 afe0c03 (ryder777777, package_release.sh dot-segment/dash/extra-arg rejection, bug live ×4 cases) · #1097 ff446e2 (ryder777777, install.sh HYB_VERSION path-component validation, bug live via regression suite failing on main) · #1098 37364a8 (ryder777777, check_test_registration start-dir escape; live-bug proof = direct probe:
-s ..+ package-chained root bypasses the guard exit-0 on main, fixed tree refuses; PR's own test passes pre-fix too — quoted-pattern weakness noted to the author) — each: merge gate clean, battery 1770/1770, corpus 126 tripwire-not-fired, differential FULL 40+86, CI compiler+quick green on heads (preflight reds were the shared-token 403 saturation, not code), credit comments posted -
Contributor #656 follow-up: apply the existing receiver guard to task callees and bound method values; 19 focused regressions cover residual defects and valid controls. Full workspace, corpus, differential, and CI are delivery gates.
-
Phase: MVP DONE + v0.1.0 released + community OPEN — session-23: triage bot shipped (PR #941) + Pareto landed 12 PRs verified (#892 #741 #869 #901 #915 #899 #913 #917 #671 #938 #885 #870) + #804/#793 evidence-closed (Go 1.26.9 accepts directional makes) — main tip b84fff6, battery cargo green, corpus 113/113, differential FULL 37+76, census 7/7
-
M1 VERDICTS: 21/21 drafts + late-arrival addendum READY (
/home/z/my-project/VERDICTS-M1-DRAFT.md): post to discussion #365 on 2026-11-01 00:00 UTC per the file's checklist; spike evidence now includes #709 (W-006, CI pending at last check), #704/#635/#637 (all CI green, December-window posture), #379 (W-015, green) -
Session-22 family resolutions: float-const (#703+#689+#685→9a41d35/89b6370), untyped-nil (#673+#691→7aaf193/aeac72a), float ops/comparisons (#587+#578→6fec37b/ee34307, #576 closed), arrays (#648, #650/#649 closed), LX6 (#677, #676 closed), guard quad (#602+#600+#540+#705→89bc13e), N5 (#662, #663 closed w/ defect analysis), set/option (#653, #675/#692 closed), sonson ×10, tools ×12, kaggle/auth ×6, corpus fixtures ×6 (corpus 79→109)
-
NEW SOUNDNESS ISSUE FILED: #722 — generic struct emission broken (
struct box[t]→type box [t]struct {, accepted-but-unbuildable); found by 22-d2 during N5 verification -
Remaining wave tail (NEXT lists in worklog): #604 #584 #591 #563 #580 #613 #501 (loan/closure) · #690 #671 #439 #619 #608 #611 (22-n tail) · #391 #696 #700 #617 #389 (22-o tail) · #402 — plus NEW arrivals #711–#724 unreconned
-
Wildcard holds (do NOT merge before window): #635 #637 #704 #709 #616 #446 #577 #572 #569 #566 #556 #545 #516 #505 #488 #480 #474 #464 #452 #405 #384 #381 #370 #379 #507 #412 #336; drafts #531/#465 stay with authors
-
CI quota healthy (4980/5000 after ~190 approvals this session); race tempo ~15 min/landing at peak — worktree-merge-push protocol held (two ort auto-merge traps caught: silent hunk deletion ×2, unpaired conflict marker ×2)
-
Next queue: wave tail + #711–#724 recon, then M1 posting (11-01), then $500 Wildcard Track ops
-
AUTOMATION (session-23):
tools/triage_bot.pyLIVE on main (merged via #941) + daily Triage Board (#940) + label-only stale guard; round-2 scan: land-candidates DRAINED (0 left), queue = ~147 stale-base (600+ commits behind, need extraction/close) + #620 (conflicts, +848 C9/C10 arity) + 27 wildcard holds; session start =scan+approveper AUTOMATION.md -
CLEANUP (session-24): behavioral subsumption matrix (every cleanup PR's .hy fixtures vs main's hyb) + git-cherry patch-id sweep → #620 CLOSED-credit (content already landed verbatim via extraction 8518b93+0d0f68d — checkpoint note was stale), #834 (premise dead, #804/#793 ruling), #911 (dup of landed #899), #891 (landed via pr-678); #837 policy pointer; VERIFIED-GAP queue posted to board #940: #784+#856 (typed-const overflow, main accepts, two takes) → #910 (const index OOB) → #942 (&m[k] non-binding) → #794+#796 (#778) → #926+#960 (#925); round-3 scan: 189 open = 117 stale-base / 23 land-cand / 27 wildcard / 13 conflicts / 5 drafts / 3 dup / 1 recon
-
LANDED (session-24): #784 isHeSatoshi (fixes #725, merge 7fcd25d) — typed const expression folded-result overflow → HYB-E6002 BUILDABLE (folder Expr::Call arm + result fold-range-check + shifts + negation; ACCEPT control pins SPEC X1/X2 run-time wrapping). Dup policy: #856 (Dimany4, Morphexe report) verified as sibling take, credited in landing msg (tier0_rule unset fails corpus contract). Battery: merge gate clean · workspace green · corpus 115/115 (+2 fixtures) · census 7/7 (36/2/77) · differential FULL 38/38 build+run + 77/77. Queue next: #910 → #942 → #794+#796 → #926+#960 → new Pareto (#617 first).
-
AUTOMATION (session-25): Shepherd W0-W7 LIVE on main (#965, merged 63f5298, push-and-assert LAND-CONFIRMED) — W1 event-driven board (minutes not daily; fixed latent triage-board.yml flag bug that had kept it red since #941), W2 CAS rebase 3/hour, W3 preflight, W4 dup closer (Phase-A dispatched = bulk-cleanup proposals; Phase-B needs verified+7d+zero-unique), W5 verify-and-land w/ push-and-assert, W6 stale-premise probe (#804 class), W7 JSONL log on automation/session-log; SHEPHERD_MODE=live via repo variable (kill switch); manual: SHEPHERD.md; landing queue unchanged: #910 → #942 → #794+#796 → #926+#960 → #617
-
LANDED (session-26): verified-gap queue DRAINED — #910 (Sam, array const-index OOB → HYB-E1002, b80fc0e+fix1571342) · #942 (Sam, &m[k] non-binding → HYB-E3104, f5c7f12) · #794 (isHeSatoshi, type-name value position, 42c75f3; #796 hardik4love-lang credited) · #926 (rickymulya, return-call arity+void, 614f18b; #960 airdropia credited) · #617 (bazingagdy-star, relative install dir, 57a010a) — corpus 115→120 (census 7/7 = 38/2/80, differential FULL 40/40+80/80), issues #725/#778/#925 closed, CI green on tip
-
GUARDRAILS (session-26): branch protection ON main (required checks quick+compiler — the only ones reporting on fork PRs; strict; force-push/deletion banned; admins exempt; linear-history OFF so two-parent landing merges keep flowing). NEW PROTOCOL RULE from the #910 landing: index/worktree split shipped counts=115 while battery verified 116 (compiler CI red on b80fc0e, fixed forward 1571342) — after ANY conflict-resolution edit run
git add -Abefore commit and the pre-push gate isgit diff HEADEMPTY; also close PRs only after GitHub push-event merge detection flips (early close froze #794 badge merged=False) -
LANDED (round 4, session-27): 8 verified from the fresh Pareto — #993 (Dimany4, generic-receiver method-arity fold, fixes #880, 349ff9e) · #964 (string-index mutation + map key types, 5aa2b28) · #995 (slice-conversion arity/operand, 3c51088) · #969 (conversion operand types + void args, c203eca) · #989 (array const slice bounds, 158352a) · #909 (slice/map literal item types, c9edb09) · #919 (address-of non-addressable, 0271750) · #928 (copy/clear/panic/recover, 08e1863) — every landing: merge gate CLEAN, main repro + go-build cross-check, merged-tree repro-fixed + over-reach controls, battery green (workspace · corpus 120 · census 7/7 = 38/2/80 · differential FULL 40+80 · git diff HEAD empty), push-and-assert LAND-CONFIRMED; issues #880 #963 #994 #968 #988 #908 #918 #927 auto-closed
-
WILDCARD (session-28, wildcard-focus run): hold integrity verified (27 holds, 0 merged, 0 bot:dup-of — Phase-B cannot eat one) · CI freshness swept 23/27 green, 4 red-quick heads (#405 #569 #572 #709) root-caused as stale-merge-ref artifact (pre-fix ci.yml
"sha":"none"self-test; compiler GREEN on all 4; self-heals on author push) · W-023 enrolled as slot 22 (#565/#577 Khanhtnq, Go-toolchain oracle INVARIANT 4): ledger row ce53c67, intake-ack+review on #577, same-seed re-run on main d7ab63a shows residual 38.03%→2.78% (rejects 38→85), corpus audit 40/40+80/80; review defect named (check_one EXEC_ERROR regression); oracle's last 2 shapes filed as #998/#999 = next landing-queue head · W-018 RESULTS condition DISCHARGED (+154 lines on #572) · W-022 strengthening delivered (a7ba19f) · M1 verdict file now 22 slots + freshness addendum, ready for 2026-11-01 -
GHOST-FORK (session-31): review's five directives executed → M11 PASS (Rust import-level coverage closed 0→89-100/case; precision root-caused on BOTH sides — 0/184 FP graph artifacts, labels measure edit-set, direct-only P 0.158; counterfactuals C1+C2 caught on real repo) · C1 repro routed to the bounty track: independent reproduction at main tip ec3fe0b posted on #982 (2026-10-21, issuecomment-6315014401) · metrics charter institutionalized (P1 verification coverage + P2 counterfactual outcomes primary, pooled strict precision per-layer only) · queue study round 1 SCORED: #998/#999 pre-landing predictions P1 3/3, P2 2/2 (repro flips 0→1 under the experiment engine, gates ok) — reports/12-queue-study-score.md
-
LANDED (session-31, autoland era): #998 + #999 in one atomic landing 36a6273 (both W-023 oracle residuals) — #998: assign-side E6002 fold now resolves named consts via the file const map (
x = krejects like the declaration twin); #999: place_read mirrors the index base into expression form so the #910 bound fires on assignment places (a[300] = 1on var arrays) · 6 new tests · corpus 122 (38/2/82, +2 fixtures, all count surfaces updated) · battery: hy-check 1340/1340, workspace 1747/1747, corpus 122/122 tripwire-not-fired, differential FULL 40/40+82/82 · issues auto-closed on push -
AUTOLAND (session-31): the loop is closed —
tools/autoland.py+autolandworkflow verify-and-land PREPARED fix bundles autonomously (baseline ACCEPT → patched REJECT w/ expected code → full battery → commit Fixes #N → push-and-assert → bundle consumed → shepherd JSONL log; any gate failure = BLOCKED with evidence, nothing lands) · kill switch unchanged: SHEPHERD_MODE=live unlocks, dry-run = baseline-only · format + trust boundaries: automation/land-queue/README.md — authoring stays with sessions; verification/landing/logging is the machine's -
AUTHOR-STAGE (session-32): the author layer is decided + scaffolded — taxonomy 2026-10-21: 63.4% of 112 open issues = const-eval/overflow (39) + bounds/index (32), the two classes #998/#999 just templated → author round-1 scope = exactly those (PRs/claims refused forever) · Ornith-1.0-9B verified REAL (HF ornith-ai/Ornith-1.0-9B[-GGUF], MIT, qwen3_5, Q4_K_M 5.63GB, Kaggle mirrors + working token) but 69.4% SWE-bench Verified is provider-reported & unranked → bake-off candidate, NOT default brain (Ornith-1.5-9B joins; brain host = Kaggle GPU reusing kaggle.yml/tools/job.py, $0 compute) · vex-harness due diligence: real, v0.2.0, 2 releases, 1 author, litellm pinned → observe-mode candidate without credentials; mini-swe-agent = auditable fallback · default harness = tools/author_stage.py v0 (stdlib-only, bundle-only, scope-guarded compiler/+tests/, accept-controls read from HEAD, PRs refused, trace.log everything, selftest 8/8) + author-stage.yml (AUTHOR_ENABLED arm → AUTHOR_MODE dry-run/live → SHEPHERD_MODE unchanged) — the author writes bundles, autoland stays the ONLY path to main · bake-off protocol + beat-humans scoreboard pre-declared in AUTHOR-STAGE.md (gates frozen before any run)
-
LANDED (session-33, author-stage round 1): the autonomous loop ran end-to-end twice — the specialized arm authored, autoland landed — #758 (named/derived const zero divisors, HYB-E6001, landing 0107658) + #827 (range-bound type forms, HYB-E2009, landing d0a2db5), both READY in round 1, $0, 0 over-reach; corpus 122->124 (fixtures registered session-side, sibling-take credit to sonson0910's stale-base companions #759/#828); census drift at HEAD repaired first (ci.yml/Makefile said 120 vs corpus 122); AUTHOR-STAGE.md §6b scoreboard
-
LANDED (session-34, author-stage round 2): T-EBIND binding-shape family — #929 (98fe6c2) + #905 (3f671b3) via autoland bundles, #903 closed subsumed-with-credit — #929: T23 source-count rules cover non-call sources (receive/map-index <=2, slice/array/string index =1, select-recv 3-name, 1-result call to 2+ names); #905: parallel decl/assign count guards + typed-decl initializer compat + positional parallel-assign kinds (O8 loan fail-open preserved; Try excluded as provably-single — corpus 10-a5 pinned the two-value
f()?shape) — both READY round 1, $0, 0 over-reach, P1 seeds committed BEFORE authoring (round-1 miss repaired); corpus 124->126 (38/2/86), census 8/8; credits magomedta (#930/#906) + rickymulya (#904) -
AUTOLAND-TRANSPORT (session-34 finding): the workflow's live push has NEVER worked — GH006 (GITHUB_TOKEN non-admin + required checks strict); round-1's 'autoland landed' was session-side autoland.py with the same gates (commit-evidence: committer Z User) — this session landed both bundles the same way; the baseline gate also caught #905's never-clean verbatim fence honestly (AUTOLAND-STALE-PREMISE -> sanitized session-side). NEXT-session infra: bundle -> branch -> PR -> required checks -> auto-merge (keeps protection meaningful, makes the loop unattended)
-
OLD-NEXT (round 7): author-stage round 2 — the template/LLM boundary: try the closure/loan families (16 issues) with a BLOCK-expectation, seed Ghost-Fork P1 predictions BEFORE authoring (round-1 miss, recorded); Ornith-1.0/1.5-9B Kaggle arms on the residual (AUTHOR-STAGE.md §5); landing queue #801 #930 #985 #984 (bounty (T,error)-tail) #935 #904 #906 #815 #774 #991 #800 #923 #751 #931 #937 #972 #780 #818 #766 #824 #828; wildcard verdicts post 2026-11-01 to #365 (22 slots ready); W4 Phase-B ~2026-10-27 (do NOT hand-flip labels)
-
Contributor #1518 (2026-10-30): rebased to main
5435169; all 11 fresh Linux checks pass on Rust 1.99.0/Go 1.26.9: runtime 4/4, real Go bridge, workspace 1891/1891, corpus 127/127, differential 41/41+86/86, format/clippy. CI/review/merge/award/payment pending; details in PLAN.md. -
Contributor (2026-11-03, fixes #3412): composite literal elements/keys/values whose composite INNER type differs (
[]chan int32{ bad_chan }withchan int64,[1][3]int{ bad_arr }with[4]int,map[string]chan int32{ "k": bad_chan }) now reject HYB-E1002 instead of passing with 0 diagnostics and failinggo build— buildable.rs deep-compares resolved composite spellings (deep_composite_conflict) wherever the coarse kind model collapses container inners; channel direction mismatches stay with the E3012 coercion layer (no double report), aliases/type parameters/non-numeric array lengths fail open. Corpus +2 (135 post-merge on current main, counts reconciled in-merge), regression pin in composite_lit_inner_types.rs.