Finish ci_runner.dag as the one-binary CI floor + collapse ci.yml shim (LOAD-BEARING; design-sign-first; AWAIT full mandate message from zesty-swift-79 before acting) by briansrls · Pull Request #4960 · gunb-ai/gunbc

briansrls · 2026-06-15T15:29:36Z

Summary

Part (1) Logging: ci_runner.dag emits modeled GHA LogAnnotations (AnnotationGroup/EndGroup, AnnotationNotice on pass, AnnotationError on fail) via extdeps/github/log_annotations.dag — no raw echo [PASS]/[FAIL].
Part (2) Floor model (v2): Four blocking gates — dsl-compile-clean (full-tree v2 parse minus enumerated allowlist), fmt, affected-tests (AffectedTestsGate fail-safe superset battery), ci-floor-parity (DagParityGate). Demote spine dissolved from ci.dag; former demoted jobs run inside gate-3.
Part (3) Shim collapse: .github/workflows/ci.yml → infra_isolation + single ci job running gunbc run --source-root dsl --function run_ci_pipeline.

Gate-1 allowlist (by-execution, post file-level folds)

231 dsl files; 2 v2-parser-debt entries remain after folding:

todo_service.dag: let → data (file-level)
schedule_model.dag: if/then/else → match (file-level)

Grandfathered (dsl/gunbc/parse_allowlist.dag):

dsl/gunbc/ci_emission.dag — record-literal Workflow body
dsl/std/unicode.dag — fn expression body (= match)

Transport: scripts/v2-dsl-full-tree-parse-gate.sh (shrink-only ratchet).

Notes

Gate-3 runs fail-safe superset (scripts/v4-affected-tests-gate.sh) until diff-scoped dispatch lands in the one-binary runner.
Branch protection must move to job id ci (+ infra_isolation).

Test plan

cargo test -p ci_workflow_ratchet (17/18; workflow_no_path_regex_policy_ci_yml needs git repo cwd)
bash scripts/v2-dsl-full-tree-parse-gate.sh → PASS (2 allowlisted files)
Local gunbc run … run_ci_pipeline emits ::group:: / ::notice:: / ::error:: wire lines
CI on self-hosted fleet after branch protection update

…loor CI dissolution (#4960, ci-runner-one-binary-floor) collapses ci.yml to one job running ci_runner.dag; the v4_lens_ci job this step lived in is going away. Per sunny-crab-863's coordination, drop the raw workflow step here and re-home the time-affine keystone into the gate-3 AffectedTestsGate battery (scripts/v4-affected-tests-gate.sh, after the probe-selector keystone) once #4960 lands. This PR now touches scripts/ + dsl/ only -- zero ci.yml collision regardless of merge order. The gate script + witnesses are kept; the one wiring line follows on rebase atop #4960. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The M1.c witness is kernel-side and fixtures-only; the v4_lens_ci_perturb 20->30m timeout bump + budget-table edits do not belong here. ci.yml is being collapsed by #4960 (which deletes v4_lens_ci_perturb), so a timeout patch there collides with the dissolution. The underlying ~169s resolve cost of affected_set_ci_runner.dag is routed to sunny-crab-863 for the one-binary floor gate-3 budget. This PR now touches only the witness .dag. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

gunbai-bot · 2026-06-15T20:31:51Z

ci: reopen to re-trigger workflow (stays draft)

…l shim (

…line) Replace ~17 parallel jobs with a thin workflow: runner-isolation guard, checkout/toolchain/cache, and a single gunbc run of run_ci_pipeline. Gate membership and GHA log annotations remain in dsl/gunbc/{ci,ci_runner}.dag. Co-authored-by: Cursor <cursoragent@cursor.com>

Apply zesty design feedback: gate-1 full-tree parse minus enumerated allowlist (2 v2-parser-debt files after folding todo_service + schedule_model), gate-3 AffectedTestsGate superset battery, gate-4 blocking DagParityGate; dissolve demote spine from ci.dag. Co-authored-by: Cursor <cursoragent@cursor.com>

gunbc run without --entry indexes all 232 dsl modules and fails on allowlisted v2-parser-debt files before any gate runs. CI shim now passes --entry dsl/gunbc/tools/ci_runner.dag; gate-1 still enforces full-tree parse via v2-dsl-full-tree-parse-gate.sh. Co-authored-by: Cursor <cursoragent@cursor.com>

…l shim (

Operator bankruptcy (msg_684c8383): floor is exactly dsl-compile-clean → fmt → affected-tests. No emit-determinism 4th gate, no reach-proof, no ci_floor freshness/emit preservation. Co-authored-by: Cursor <cursoragent@cursor.com>

Add scripts/ci-floor-witnesses.sh (gate-1/3/overall RED demos) as a post-pipeline step in the one-binary ci job so design-sign evidence runs by-execution in live CI, not local-only. Co-authored-by: Cursor <cursoragent@cursor.com>

Add scripts/ci-floor-witnesses.sh for gate-1/gate-3/overall +/- witnesses. Ratchet tests pin 3-gate topology (DslCompileClean, Fmt, AffectedTests). Co-authored-by: Cursor <cursoragent@cursor.com>

Witness script now supports the ci.yml RED-demo invocation; fixes CI failure from unknown mode after witness wiring. Co-authored-by: Cursor <cursoragent@cursor.com>

Include gate-3 perturb-check in the hermetic witness step so collapsed CI exercises gate-1, gate-3, and overall floor RED assertions. Co-authored-by: Cursor <cursoragent@cursor.com>

Login shells (sh -lc) reset PATH on self-hosted runners, causing all floor gate scripts to fail instantly. Also improve witness RED log lines (PLANTED/OBSERVED/DETECTED) for design-sign log review. Co-authored-by: Cursor <cursoragent@cursor.com>

gunbc shell.Exec buffers all subprocess stdout/stderr in memory. Gate-3 runs ~30+ minutes with heavy output; on arm64 CI this likely OOM-killed the step after ~23m with no useful log. Redirect to RUNNER_TEMP log and tail last 120 lines; print tail on gate failure. Co-authored-by: Cursor <cursoragent@cursor.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

infra_isolation de-priv guard removed per operator. sccache server bring-up will be modeled fail-closed inside the floor binary (next commit); the shim keeps only CARGO_HOME/RUSTUP_HOME/CARGO_BUILD_JOBS isolation, which must precede the GHA Setup-Rust/Cache-Cargo steps. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…nto vendor-agnostic CpuFacts std/cpu owned a one-vendor SKU enum (AmpereCpuModel) and dispatched cpu_catalog_row on it — std (a standard library) enumerating one vendor's parts is a purchase order, not a concept. Path is the discriminator: vendor hardware is an external dependency. - dsl/std/cpu/ampere.dag -> dsl/extdeps/cpu/ampere.dag (Altra rows as CpuModelCatalogRow values carrying the new vendor field; imports the catalog SHAPE from std.cpu). - std/cpu/types.dag: dissolve AmpereCpuModel + CpuIdentity; CpuModelCatalogRow gains vendor: CpuVendor; CpuFacts now { catalog, deployment } (identity-by-part). - std/cpu.dag: drop cpu_catalog_row SKU dispatch; cpu_vendor / cpu_facts_* project straight off cpu.catalog.*. std/cpu is the general CPU concept only. - integration.rs parser smoke + placement_supply doc-comment repointed. scratch.txt items 3/4/5. gate-1 full-tree parse PASS. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…stances -> ctrl) The concrete srv1/srv2 DIMM populations are the operator's deployment, not a gunbc standard-library fact — re-homed to ctrl plans/models/fabric/operator_fleet.dag alongside the srv1/srv2/Hetzner ComputeOffers (item 6). No .dag consumed the file; only a Rust parser-smoke include_str + two doc-comments referenced it. std/memory keeps the DRAM taxonomy (types) + the SK Hynix catalog DISPATCH — note that SK Hynix vendor DATA in std/memory/sk_hynix.dag is the same misplacement smell as the now-moved cpu/ampere (scratch item 10), tracked separately. scratch.txt items 2 + 6 (gunbc leg). gate-1 full-tree parse PASS. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…frozen) Revert ci.yml (queue:max, v4_lens_gate 45m), compute_fabric.dag, and check_ci_timeout_budgets.py to main. CI-infra belongs to CI-floor lane (#4960/#4997); not pushed until #4997 lands + clean window per zesty. Keeps: walkthrough, program.dag M1a header, program_runs corpus enrollment. Co-authored-by: Cursor <cursoragent@cursor.com>

…ot a std primitive) std/ is a standard library of universal primitives; the compute fabric is a supply/demand/placement MARKET — a domain/product model. It imports std/ only and sits as a product peer alongside the existing non-core trees (gunbc/, demos/, ctrl/), so this does NOT alter the INVARIANTS §P2 core DAG (std->extdeps->compiler->workflow). - dsl/std/compute_fabric.dag -> dsl/product/compute_fabric.dag; module std.compute_fabric -> product.compute_fabric. - Functional blast radius was one real import (probe_selector repro witness) + one Rust include_str parser-smoke; both repointed. All other references were comments, updated for coherence (skipped crisp-bear-37's in-flight corpus file + dated docs/planning notes). scratch.txt item 7. gate-1 full-tree parse PASS. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ose CpuSocket string Two modeling smells from the audit (scratch items 10, 11): item 10 (mirror of the cpu/ampere dissolution): SK Hynix vendor catalog was baked into std/, with std.memory dispatching on an owned-SKU enum. - dsl/std/memory/sk_hynix.dag -> dsl/extdeps/memory/sk_hynix.dag (rows as DramModuleCatalogRow values; vendor data is an external dependency). - Dissolve DramModuleIdentity + SkHynixDdr4Module SKU enums and the std.memory.dram_module_catalog_row dispatch. MemoryPopulationRow now carries directly; std/memory imports no vendor catalog. item 11 (stringly atom): socket: "LGA4926" decomposed into a grounded type. - New std.cpu.types CpuSocket { package: SocketPackage, contact_count: Int }; SocketPackage = LandGridArray | PinGridArray | BallGridArray. LGA = Land Grid Array, 4926 = land/contact count — the trailing number is now a grounded Int, not an opaque token. Anchor cited (Wikipedia CPU socket). - extdeps/cpu/ampere rows construct CpuSocket { package: LandGridArray, contact_count: 4926 }. integration.rs parser-smoke repointed (sk_hynix). gate-1 full-tree parse PASS. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The M1.c witness is kernel-side and fixtures-only; the v4_lens_ci_perturb 20->30m timeout bump + budget-table edits do not belong here. ci.yml is being collapsed by #4960 (which deletes v4_lens_ci_perturb), so a timeout patch there collides with the dissolution. The underlying ~169s resolve cost of affected_set_ci_runner.dag is routed to sunny-crab-863 for the one-binary floor gate-3 budget. This PR now touches only the witness .dag. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…floor # Conflicts: # .github/workflows/ci.yml # dsl/product/compute_fabric.dag

…endor closed-enum-in-std) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…gn pass Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…floor

…verer gap) + item 14 (CI gates are bash scripts; same root cause) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The M1.c witness is kernel-side and fixtures-only; the v4_lens_ci_perturb 20->30m timeout bump + budget-table edits do not belong here. ci.yml is being collapsed by #4960 (which deletes v4_lens_ci_perturb), so a timeout patch there collides with the dissolution. The underlying ~169s resolve cost of affected_set_ci_runner.dag is routed to sunny-crab-863 for the one-binary floor gate-3 budget. This PR now touches only the witness .dag. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Stack modeled ci_base_build_graph pre-warm onto merged one-binary floor (#4960): runs after sccache_upsert_preflight, before gate fan-out. Re-homes gate-3 consumers (doc_refs, timeout_budgets, bootstrap ratchets), threads SourceRoot/allowlist authority, and fixes log annotation rendering. Co-authored-by: Cursor <cursoragent@cursor.com>

Drop carried CI-infra commits (sccache EAGAIN, perturb timeout, concurrency comment) that collided with #4960 ci.yml collapse; OCI enrollment only. Co-authored-by: Cursor <cursoragent@cursor.com>

…eipts (#4978) Apply budget-script ceilings only (no ci.yml touch post-#4960 collapse). Co-authored-by: Cursor <cursoragent@cursor.com>

#4960 ci.yml has only the `ci` job; remove infra_isolation from the timeout budget table so gate-3 check passes. Apply cargo fmt to ratchet test. Co-authored-by: Cursor <cursoragent@cursor.com>

…abric.dag, dsl/std/cpu/types.dag) These files do not exist in origin/main (compute_fabric moved to dsl/product/ in #4960; cpu/types.dag later dissolved). They were swept into this branch by a previous auto-commit that captured unmerged index state. Removing them keeps the PR diff scoped to wiring-liveness changes only. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

gunbai-bot Bot marked this pull request as ready for review June 15, 2026 16:09

gunbai-bot Bot marked this pull request as draft June 15, 2026 16:10

briansrls mentioned this pull request Jun 15, 2026

std.time_affine: affine-instant time carrier (#4883 TemporalAggregator unit fix candidate) #4928

Closed

gunbai-bot Bot mentioned this pull request Jun 15, 2026

gunbc M1.c: add ci_runner_strict_subset_discriminates_holds claim-run witness to affected_set_ci_runner.dag (fixtures-only, signed-off shape) #4981

Closed

briansrls marked this pull request as ready for review June 15, 2026 18:58

briansrls mentioned this pull request Jun 15, 2026

fix(ci): fleet throttle + docs-only lens skip + sccache stop-server removal #4991

Merged

gunbai-bot Bot closed this Jun 15, 2026

gunbai-bot Bot reopened this Jun 15, 2026

gunbai-bot Bot marked this pull request as draft June 15, 2026 20:32

gunbai-bot Bot mentioned this pull request Jun 15, 2026

CI Rework #4996

Closed

6 tasks

briansrls and others added 17 commits June 15, 2026 21:07

WIP: Finish ci_runner.dag as the one-binary CI floor + collapse ci.ym…

56f6b73

…l shim (

WIP: Finish ci_runner.dag as the one-binary CI floor + collapse ci.ym…

9c07208

…l shim (

WIP: Finish ci_runner.dag as the one-binary CI floor + collapse ci.ym…

6c90e2e

…l shim (

WIP: Finish ci_runner.dag as the one-binary CI floor + collapse ci.ym…

997b406

…l shim (

ci: wire hermetic floor +/- witnesses in collapsed CI

be327bc

Add scripts/ci-floor-witnesses.sh (gate-1/3/overall RED demos) as a post-pipeline step in the one-binary ci job so design-sign evidence runs by-execution in live CI, not local-only. Co-authored-by: Cursor <cursoragent@cursor.com>

ci: wire floor witnesses + finalize 3-gate bankruptcy floor

a5592c1

Add scripts/ci-floor-witnesses.sh for gate-1/gate-3/overall +/- witnesses. Ratchet tests pin 3-gate topology (DslCompileClean, Fmt, AffectedTests). Co-authored-by: Cursor <cursoragent@cursor.com>

fix(ci): add --red-demo witness mode for ci.yml teeth step

7df91a0

Witness script now supports the ci.yml RED-demo invocation; fixes CI failure from unknown mode after witness wiring. Co-authored-by: Cursor <cursoragent@cursor.com>

ci: run all 3 RED teeth witnesses in --red-demo CI step

01f937b

Include gate-3 perturb-check in the hermetic witness step so collapsed CI exercises gate-1, gate-3, and overall floor RED assertions. Co-authored-by: Cursor <cursoragent@cursor.com>

ci: re-trigger workflow for gate-3 OOM fix (1be7320)

9c7f422

Co-authored-by: Cursor <cursoragent@cursor.com>

ci: trigger workflow after canceling stuck run

eb61d2d

Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls and others added 3 commits June 15, 2026 21:57

WIP: CI Rework

c13bb28

briansrls and others added 2 commits June 15, 2026 22:17

WIP: CI Rework

36f3647

briansrls and others added 3 commits June 15, 2026 22:41

Merge remote-tracking branch 'origin/main' into ci-runner-one-binary-…

908adb6

…floor # Conflicts: # .github/workflows/ci.yml # dsl/product/compute_fabric.dag

scratch: log audit items 12 (arch enum triplication) + 13 (hardware-v…

3a80ae8

…endor closed-enum-in-std) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

scratch: park items 12/13 (std/extdeps boundary) for a dedicated desi…

7adf887

…gn pass Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

gunbai-bot Bot marked this pull request as ready for review June 15, 2026 22:52

briansrls and others added 2 commits June 15, 2026 23:04

Merge remote-tracking branch 'origin/main' into ci-runner-one-binary-…

59b10d5

…floor

scratch: item 8 root-cause (bash text-scrape enrollment + 20/54 disco…

b5f8bd2

…verer gap) + item 14 (CI gates are bash scripts; same root cause) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

briansrls merged commit 87ae7c4 into main Jun 15, 2026
1 check failed

briansrls deleted the ci-runner-one-binary-floor branch June 15, 2026 23:18

gunbai-bot Bot mentioned this pull request Jun 15, 2026

§4 demo: TS-first walkthrough + enroll program_runs_holds in CI corpus #4980

Closed

briansrls mentioned this pull request Jun 15, 2026

§10 Slice-0 OCI: CI-enroll ctrl-session image+runtime witness into claim-witness corpus gate #4971

Closed

briansrls added a commit that referenced this pull request Jun 15, 2026

fix(ci): raise v4_lens_gate/perturb timeout ceilings with on-wave rec…

29a1b4b

…eipts (#4978) Apply budget-script ceilings only (no ci.yml touch post-#4960 collapse). Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls mentioned this pull request Jun 15, 2026

CI floor: pre-warm sccache with gunbc base build graph per pin BEFORE lens/gate fan-out (kills cold-start stampede) #4992

Closed

2 of 3 tasks

gunbai-bot Bot mentioned this pull request Jun 16, 2026

SP-M3: mixed-signal analog+digital on shared map/fold substrate (verilog.dag digital side) #5007

Closed

6 tasks

briansrls mentioned this pull request Jun 16, 2026

docs(roadmap): 2026-06-16 wave reconciliation — landed vs unmerged work #5020

Closed

Navigation Menu

Repository navigation

Finish ci_runner.dag as the one-binary CI floor + collapse ci.yml shim (LOAD-BEARING; design-sign-first; AWAIT full mandate message from zesty-swift-79 before acting) - #4960

Finish ci_runner.dag as the one-binary CI floor + collapse ci.yml shim (LOAD-BEARING; design-sign-first; AWAIT full mandate message from zesty-swift-79 before acting)#4960
briansrls merged 43 commits into
mainfrom
ci-runner-one-binary-floor

briansrls commented Jun 15, 2026 •

edited by gunbai-bot Bot

Loading

Uh oh!

gunbai-bot Bot commented Jun 15, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

briansrls commented Jun 15, 2026 • edited by gunbai-bot Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Gate-1 allowlist (by-execution, post file-level folds)

Notes

Test plan

Uh oh!

gunbai-bot Bot commented Jun 15, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

briansrls commented Jun 15, 2026 •

edited by gunbai-bot Bot

Loading