Skip to content

fix(hotpatch): resolve bucket-D prod/repo divergences - #4

Open
alpha1122x wants to merge 7 commits into
fix/prod-hotpatch-reconciliationfrom
fix/hotpatch-bucket-d-resolution
Open

fix(hotpatch): resolve bucket-D prod/repo divergences#4
alpha1122x wants to merge 7 commits into
fix/prod-hotpatch-reconciliationfrom
fix/hotpatch-bucket-d-resolution

Conversation

@alpha1122x

Copy link
Copy Markdown
Owner

Summary

Per-file resolution table

Hotpatch / file Resolution Notes
ac_selfdeploy_bundle/shapes.py Land prod Disk bounds, eval tdx.xlarge default, disk cost helpers
ac_selfdeploy_bundle/lifecycle.py Land prod Disk-aware combined budget (blocked on shapes)
ac_selfdeploy_bundle/phala.py Land prod delete_cvm HTTP DELETE + unique app_id teardown resolve
ac_benchmarks.py Land prod resolve_dataset_digest_path for site-packages / /app/golden
ac_eval_wire.py Merge Eval path. Land n_concurrent, progress validators, optional hydration_digest. Keep guest_artifact_proof + fail-closed match is not True, optional app_identity, RA-TLS KR authority validation
ac_authorization.py Merge Eval path. Land resolve_plan_n_concurrent, plan field, active/terminal prepare recovery, digest path helper. Keep dual residual / settings-gated admit_eval_cvm_launch_from_assignment (prod tried to drop it — attestation risk)
ac_sdk_config.py Merge Land eval_max_attempts 1–256. Keep raw_weight_push settings + internal_token() + AliasChoices
ac_canonical_compose.py Merge Land EVAL_PROGRESS_* + DSTACK_DOCKER_* allowed envs. Keep artifact URL/token envs + key_release_endpoint module integration. Compose hash changed → pin constant updated; ops must re-measure live tee-pin-pack
ac_selfdeploy_bundle/eval.py Merge Add disk_size_gb on discovery path only. Keep multi-account discovery, no nonce, artifact grant minting (repo is correct vs prod pin/nonce)
ac_selfdeploy_bundle/review.py Merge Add disk_size_gb on discovery path. Keep discovery / no nonce
ac_selfdeploy_bundle/cli.py Merge Land HTTP teardown, disk flags, token handoff, measurement pin assert. Keep artifact grant + discovery deploy
ac_review_runtime.py Merge Land submit_verdict tool-call prompt. Keep joinbase REVIEW_API_BASE_URL pin (prod weakened it — not landed)
prism_plagiarism_llm/queue.py Land prod Worker-plane no-op on master + OpenRouter plagiarism adjudicator

Prod-only drift intentionally not committed

  • Review runtime free REVIEW_API_BASE_URL override (weakens pin) — remains prod-only if still mounted.
  • Selfdeploy eval/review app_id pin + provision nonce path still on prod mounts — do not re-apply; repo discovery path is the live-verified 422 fix. Prod mounts for those files should be refreshed from this branch after merge.

Still unresolved / residual risk

  • Live tee-pin-pack under mission evidence still carries pre-env-change hash 04011776…; generator is now 9a550b2d…. Ops must re-measure eval compose pin before next production pin cut (test skips stale pack when present).
  • Prod ac_authorization residual-gate removal was rejected as attestation weakening; if prod mount still has the weak path, keep the mount until this lands and image rebuilds.
  • Attestation surfaces (compose_hash byte equality, measurement allowlist, KMS digest, signature verification) were not loosened.

Eval path callouts

Touched directly: eval_wire.py, authorization.py, benchmarks.py, selfdeploy/eval.py, canonical/compose.py (allowed envs / measured compose). Guest artifact proof and dual residual AGATE gate preserved.

Test plan

  • UV_CACHE_DIR=/var/tmp/uv-cache uv run pytest -q -p no:randomly from packages/challenges/agent-challenge
  • Gate: 57 failed, 3276 passed, 9 skipped, 6 errors (baseline 57/3262/9/6; passed ≥3262, skipped=9)
  • Blackbox tests/test_product_blackbox_attestation_chain.py: 17 failed / 19 passed
  • tests/test_bucket_d_hotpatch_resolution.py behavioral coverage (RED→GREEN for n_concurrent, digest path, disk, progress, dual residual signature)
  • ruff clean on touched files
  • Secret-scan on staged diffs: no phak_, sk-, Bearer, private-key headers, mnemonics

CI

  • Waiting for required checks to go green on this head SHA.

Notes

  • Read-only on master 86.38.238.235 (no restarts, mount changes, or deploys).
  • Target base is fix/prod-hotpatch-reconciliation (not main).

echobt added 7 commits July 28, 2026 07:20
… delete

Prod bucket-D shapes/lifecycle/phala fixes: stage disk bounds, eval
tdx.xlarge default, disk-aware budget, and PhalaCloudClient.delete_cvm
so validators can tear down without a phala binary.
…nstalls

Prefer env/manifest and /app|/opt golden layouts over Path(__file__).parents[3]
so eval/prepare no longer 503s when the package is installed under site-packages.
…d compose envs

Land miner n_concurrent plan binding, optional hydration_digest, mid-run
progress validators, and EVAL_PROGRESS/DSTACK_DOCKER allowed envs. Keep
guest_artifact_proof, RA-TLS KR authority checks, dual residual review
gate, and raw-weight-push settings from the repo side.
…handoff

Provision disk_size without re-pinning app_id or reintroducing nonce.
CLI gains HTTP teardown, disk flags, and eval run-token handoff while
keeping multi-account discovery and artifact grant minting.
… prompt

Land prod tool-call contract wording so Grok tool_choice=auto emits a
closed submit_verdict call. Keep the joinbase REVIEW_API_BASE_URL pin.
When worker_plane is enabled the master queue no longer claims GPU work.
Borderline/attach plagiarism pairs go through the OpenRouter adjudicator
instead of terminal-rejecting the quarantine band.
…tures

Add behavioral tests for digest path, n_concurrent, progress wire, disk
guards, and dual residual signature. Update compose pin hash and
lifecycle/disk/teardown expectations for the landed product changes.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants