Repository navigation
Conversation
pipeline/diff_uk_event_runs.py compares two complete replay namespaces after re-verifying each side's staging, comparison and artifact receipts. It flags, at source-row, measure-year and PE-head level: numerical moves above max(£1m, 1% of the larger magnitude); sign reversals and zero crossings; coverage, mapping, support, inertness and construction changes regardless of size. A year one side cannot support is a support flag, never coverage, and only on rows that could carry a counterpart. Ratio bins are reported as transitions. Flags are investigation prompts; nothing passes or fails. Each flagged (event, measure, year) gets a packet with the OBR source cells, both execution identities, artifact digests, executed reform dictionaries, focused rerun commands and the sources the registry cites. --check re-renders and compares bytes for CI. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A rebuilt environment for the same three wheels resolved other pydantic, pydantic-core and filelock versions and left out pypdf, and a raw RECORD digest also bound the installer (uv_cache.json, console-script shebangs that name the venv). Now: - identity.requirements_sha256 binds the exact freeze; verify_installed checks every pinned distribution at its version (extras allowed); - record_sha256 excludes installer bookkeeping, bytecode and generated scripts outside site-packages, so pip and uv installs of one wheel agree; - a freeze with local-wheel or commit installs is normalised to exact pins; - a variant id takes an identity-digest suffix only when another identity already holds its version slug; - compute_uk_event --variants-file selects from another variants document (a qualification run adds its candidate there). Schemas: requested_by.workflow_run_id is optional (a catch-up run started outside Actions has none); a failed receipt may have no execution block (a step failed before the candidate's identity was known). Both relax v1. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
.github/workflows/uk-replay.yml receives W4's uk-replay-qualification-request dispatches (and manual requests). The enqueue job writes each request to the replay-state branch first (pipeline/uk_replay_queue.py), so GitHub cancelling an older pending run never loses one; the drain job runs one request at a time, two simulation workers at most, and a schedule every three hours drains what is left. A request still running seven hours after its last update is retried. pipeline/uk_replay_qualify.py run-request takes one request end to end: schema check; the dataset from Hugging Face, checked by size and digest; the model and core wheels downloaded and checked against the request, then installed with the wrapper and, for a certified pair, that pair's exact freeze (a new candidate resolves within the newest certified freeze); a probe of the installed identity, the release bundle under packaged certification, the certification and the data years; the corpus; the diff against the newest other certified variant; and a sealed receipt (NAMESPACE_MANIFEST.json, REPLAY_RECEIPT.json). Every outcome, including a rejection or a crash, produces a receipt. The terminal STATUS.json names the receipt digest and the commit that added it. ci.yml now re-stages and re-compares every committed namespace, re-renders every committed diff (--check) and verifies every seal; and a replay engine job per certified variant installs its exact freeze, verifies the installed environment, runs the construction guards and regenerates the committed premises and diagnostics, and with the HUGGING_FACE_TOKEN secret replays one real measure-year and compares bytes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A rebuilt environment for the same three wheels resolved other pydantic, pydantic-core and filelock versions and left out pypdf, and a raw RECORD digest also bound the installer (uv_cache.json, console-script shebangs that name the venv). Now: - identity.requirements_sha256 binds the exact freeze; verify_installed checks every pinned distribution at its version (extras allowed); - record_sha256 excludes installer bookkeeping, bytecode and generated scripts outside site-packages, so pip and uv installs of one wheel agree; - a freeze with local-wheel or commit installs is normalised to exact pins; - a variant id takes an identity-digest suffix only when another identity already holds its version slug; - compute_uk_event --variants-file selects from another variants document (a qualification run adds its candidate there). Schemas: requested_by.workflow_run_id is optional (a catch-up run started outside Actions has none); a failed receipt may have no execution block (a step failed before the candidate's identity was known). Both relax v1. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> (cherry picked from commit e983891)
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
…ruction On the first real premise diff (2.89.2 to 2.125.1) every reversal's reform world, which is current law, moved with current law: the employer-NIC secondary threshold (96 to 96.153846), the Class 2 small-profits threshold, float noise in the blended CGT rates. Those were raised as construction flags, which release qualification proposes to fail on. A baseline or reform world value change is now a construction flag only where one side's world overrides that side's current law; otherwise the move is recorded in current_law_changed context. Premise-check changes (the CGT annualisation premise) still flag. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The variants file is the index (no INDEX.json); a sealed namespace holds NAMESPACE_MANIFEST.json and REQUEST.json; reruns of a sealed variant use an explicit output directory rather than a rerun-<n> suffix. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #162. It also carries #161, merged in, and #160's latest commit, until those land. WIP: W6 of the Autumn Budget 2026 CI plan.
What this adds
Receiver.
.github/workflows/uk-replay.ymlreceives W4'suk-replay-qualification-requestdispatches (interface:docs/uk_replay/QUALIFICATION_INTERFACE.md) and manual requests.replay-statebranch before anything else (pipeline/uk_replay_queue.py). GitHub's cancellation of older pending runs therefore cannot lose a request. Enqueueing is idempotent.uk-replay-drain, two simulation workers at most) runs.github/scripts/uk_replay_drain.shfor the oldest request, throughpipeline/uk_replay_qualify.py run-request:HUGGING_FACE_TOKENand check its size and SHA-256.schema_invalid,data_access_failed,artifact_hash_mismatch,release_bundle_mismatch,incompatible_pair,no_supported_years.failedreceipt, so no request loops. The receipt is committed first. The terminalSTATUS.jsonthen names its digest and commit.runningafter seven hours is retried.Replay in CI (
ci.yml):--check;replay enginejob, per certified variant:Verified end to end, outside Actions
Using the actual subcommands against a
released_bundlerequest for policyengine.py 6.2.6 built from its PyPI wheel hashes:preparedownloaded and verified the three wheels and installed the certified freeze exactly.proberecognised the environment as the committed certified variant (no new record).corpusrecomputed all five events in about 18 minutes with no failures. Artifact numbers were byte-identical to the earlier run except the identity digest.Needs before it can run in Actions
HUGGING_FACE_TOKENrepo secret, requested from the coordinating session.repository_dispatch(W4 asked the coordinating session).replay-statebranch is created by the first enqueue (orphan).Invariants and tests (
tests/test_uk_replay_qualify.py, 18 tests)-r) versus constrained (-c) installs work as specified.axiom: n/a: replay CI and qualification plumbing in scorecard, no policy encoding.
🤖 Generated with Claude Code