Skip to content

P2C10: reproduce measured correction over BLS #13

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
eamirian wants to merge 1 commit into from
Closed

P2C10: reproduce measured correction over BLS #13

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 20 additions & 4 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -261,6 +261,20 @@ fixture inputs, not ACE outputs. This is exact forecast/result scoring, not a hi
contemporaneous forecast, model-skill finding, empirical calibration curve, population reliability
estimate, causal claim, or human-benefit finding.

P2C10 independently reproduces correction-quality measurement over the U.S. Bureau of Labor
Statistics public errata family. The recorded July 1, 2025 JOLTS release `USDL-25-1087` and BLS's
July 2 correction are admitted as distinct immutable Observations. Treatment and control preserve
both source identities, the exact correction link, two reviewed Actions each, and the same matched
conditions; treatment renders the required minus sign in `−39,000`, while control retains the
reported pre-correction form. Exact reviews score `1.0` and `0.0`, so the unchanged domain-neutral
contract classifies the bounded two-pair difference as useful and emits only a non-effective
proposal.

The historical wrong form is derived from BLS's explicit statement that the sentence required a
missing minus sign; the currently archived release is already corrected. This is recorded replay
over one correction, not live monitoring, statistical validation of JOLTS, population correction
performance, causality, or human benefit. No Domain Pack, connector, or Core contract changes.

## What the public World proof demonstrates

Generate a self-contained visual Reality Brief and its exact machine-readable backing data:
Expand Down Expand Up @@ -361,6 +375,7 @@ $PY -m scripts.p2c6_contradiction_attention_outcome "$WORKSPACE"
$PY -m scripts.p2c7_correction_detection_delay_outcome "$WORKSPACE"
$PY -m scripts.p2c8_correction_revision_stability_outcome "$WORKSPACE"
$PY -m scripts.p2c9_forecast_calibration_outcome "$WORKSPACE"
$PY -m scripts.p2c10_independent_correction_reproduction "$WORKSPACE"
```

The released 0.9.0 gates are reproducible through the locked environment, as CI does. The
Expand Down Expand Up @@ -397,8 +412,8 @@ The complete governed product-journey evidence is recorded in
[`docs/audits/world-intelligence-p2c2-governed-reality-brief-2026-08-10.md`](docs/audits/world-intelligence-p2c2-governed-reality-brief-2026-08-10.md).
The source-checkout measured-feedback candidate is recorded in
[`docs/audits/world-intelligence-p2c3-measured-feedback-2026-08-10.md`](docs/audits/world-intelligence-p2c3-measured-feedback-2026-08-10.md).
The latest withheld-result forecast-scoring candidate is recorded in
[`docs/audits/world-intelligence-p2c9-forecast-calibration-outcome-2026-08-10.md`](docs/audits/world-intelligence-p2c9-forecast-calibration-outcome-2026-08-10.md).
The latest independent-source correction candidate is recorded in
[`docs/audits/world-intelligence-p2c10-independent-correction-reproduction-2026-08-10.md`](docs/audits/world-intelligence-p2c10-independent-correction-reproduction-2026-08-10.md).

Release-level scope and evidence are recorded in
[`docs/releases/world-intelligence-p2c2-v0.9.0.md`](docs/releases/world-intelligence-p2c2-v0.9.0.md),
Expand Down Expand Up @@ -435,8 +450,9 @@ detection delay against a product target. P2C8 measures exact unaffected-claim i
preservation across real Brief contracts while holding correction semantics, source coverage, and
claim count constant. P2C9 adds an exact withheld-result forecast record and derives a single-event
Brier contribution while explicitly withholding any population-calibration or model-skill claim.
The next bounded measurement work is another independently sourced correction event or independent
Market reproduction.
P2C10 reproduces correction-quality measurement over an independently sourced BLS erratum without
changing the shared contracts. The next bounded measurement work is independent Market
reproduction.
Separately reviewed opt-in network transport and P2D multi-source conflict/correction with LIVE
inputs remain independent work. None of these steps may add autonomous publishing, delivery,
persuasion, or action authority to a Domain Pack.
Expand Down
11 changes: 10 additions & 1 deletion ROADMAP.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ source code into the platform.

See the [0.9.0 release record](docs/releases/world-intelligence-p2c2-v0.9.0.md).

## Candidate — P2C3–P2C9 measured feedback and product-owned outcomes
## Candidate — P2C3–P2C10 measured feedback and product-owned outcomes

- The exact 0.9.0 Brief and a World-owned source-only control pass through two matched reviewed
export pairs under one frozen structural citation-coverage criterion.
Expand Down Expand Up @@ -51,6 +51,11 @@ See the [0.9.0 release record](docs/releases/world-intelligence-p2c2-v0.9.0.md).
resolves the binary event; exact reviews derive single-event Brier quality of `0.9375` and
`0.4375`. This proves forecast/result scoring and leakage-resistant record order, not historical
contemporaneity, probability generation by ACE, model skill, or population calibration.
- A seventh frozen criterion repeats correction-quality measurement over the independent BLS public
errata source family. Treatment and control retain the same release, correction, correction link,
source coverage, reviewed workflow, and matched conditions; only treatment renders the exact
corrected minus sign. The unchanged shared contract derives `1.0` versus `0.0`, classifies the
bounded two-pair difference useful, and leaves the proposal non-effective and unapplied.
- This is source-checkout evidence against stacked Core candidates, not a released World capability, human
benefit finding, causal claim, network-freshness proof, or applied governance change.

Expand All @@ -72,6 +77,10 @@ performance, or a general semantic-equivalence engine. The stacked
[P2C9 work packet](docs/design/world-intelligence-p2c9-forecast-calibration-outcome-work-packet-v1.md)
freezes an exact forecast/result scoring boundary without claiming a historical forecast, model
skill, or population calibration.
The stacked
[P2C10 work packet](docs/design/world-intelligence-p2c10-independent-correction-reproduction-work-packet-v1.md)
freezes independent BLS correction reproduction without claiming live monitoring, statistical
validity, population correction performance, causality, or human benefit.

## Next — trustworthy live orientation

Expand Down
137 changes: 137 additions & 0 deletions ...dits/world-intelligence-p2c10-independent-correction-reproduction-2026-08-10.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,137 @@
# World Intelligence P2C10 independent correction reproduction audit — 2026-08-10

Status: **stacked candidate evidence only; not a release, SI4 pass, live-monitoring result,
population-performance finding, or applied governance change**

## Source identity

- World base: P2C9 commit `53feadb40fcc93d23f326b16979ed6640471c4cf`
- World branch: `codex/world-independent-correction-reproduction`
- Core dependency: exact observed-result candidate commit
`433e3d16c5458c975557dcd1552824fb959d4d12`
- Released identity intentionally unchanged: `ace-domain-world-intelligence==0.9.0`,
`ace-core>=0.5.0,<0.6`

## Exact point-in-time result

The frozen BLS fixture is
`sha256:981183a2464f74f4421bd0a6470f0342a5abae8cfebc1b5a8da1562df26babb1`.
It binds release `USDL-25-1087` to correction `bls-errata-2025-07-01-jolts`, with recorded
Observation identities:

```text
original: observation:faf25d26cc88802368cabf3e17538a7d
correction: observation:3a1351d6ac306374b8a5b472c192d2b9
```

The point-run treatment artifact was
`official_correction_artifact:34d538fbd2c3a6ef6d3e79d0fdb8a344` with material
`sha256:34d538fbd2c3a6ef6d3e79d0fdb8a3440c06b02a1e156a6001bfc1491fd86e21`.
The matched control was
`official_correction_artifact:ab6cfedd1b76602e51c0c8d3c1dbd6d2` with material
`sha256:ab6cfedd1b76602e51c0c8d3c1dbd6d22e5c76b01b1eb59350a8e1bdaa7d8207`.

Both artifacts name the same exact release and correction Observations, `corrects` relation,
source-policy digest, and reviewed workflow. Treatment renders `−39,000`; control renders the
reported pre-correction `(39,000)` form. All four reviews independently recorded complete source
coverage, visible correction linkage, and preserved prior history:

```text
treatment scores: 1.0, 1.0
control scores: 0.0, 0.0
matched pairs: 2
mean effect: 1.0
classification: useful
proposal action: promote
proposal live effect: false
historical replay: true
replay reauthorization: false
```

The exact point-run evaluation was `impact_evaluation:5490b5c680940d88d25d9dacca67103b`
with material
`sha256:5490b5c680940d88d25d9dacca67103b6fb33b8d8c98e0e756488f45dece964e`.
The exact non-effective proposal was
`impact_governance_proposal:da09bf45bb9f48d6cefbb3fb49122e67` with material
`sha256:da09bf45bb9f48d6cefbb3fb49122e679d010cd39140950a57c447e2465f7f86`.

Point-run artifact and evaluation identities include exact record-availability and reviewed-action
times. Historical replay in the same durable store returns those exact identities without
reauthorization. Fresh hosts reproduce the fixture digest, source content identities, scores,
classification, and proposal semantics without pretending independent wall-clock availability
coordinates are identical.

## Verification

The frozen World dependency versions plus the stacked Core candidate and separately packaged
reference action adapter produced:

```text
python -B -m pytest domain_packs/tests/test_p2c3_measured_feedback.py \
domain_packs/tests/test_p2c4_reviewed_impact_disposition.py \
domain_packs/tests/test_p2c5_citation_correctness_outcome.py \
domain_packs/tests/test_p2c6_contradiction_attention_outcome.py \
domain_packs/tests/test_p2c7_correction_detection_delay_outcome.py \
domain_packs/tests/test_p2c8_correction_revision_stability_outcome.py \
domain_packs/tests/test_p2c9_forecast_calibration_outcome.py \
domain_packs/tests/test_p2c10_independent_correction_reproduction.py -q --tb=short
30 passed in 11.01s

python -B -m pytest -q --tb=short
113 passed in 24.20s

python -B -m pytest adapters/federal_register_source/tests -q --tb=short
26 passed in 0.25s

python -B -m pytest tests/test_release_contract.py -q --tb=short
7 passed in 0.05s

# Installed public ace-core==0.5.0; candidate tests skip explicitly.
python -B -m pytest -q --tb=short -rs
83 passed, 30 skipped in 14.10s

ruff check --no-cache <P2C10 changed Python files>
PASS

ruff format --check --no-cache <P2C10 changed Python files>
3 files already formatted

UV_CACHE_DIR=/tmp/ace-p2c10-uv-cache uv build --out-dir /tmp/ace-p2c10-dist-20260810
Successfully built unchanged 0.9.0 source distribution and inert data-only wheel

git diff --check
PASS
```

The thirty public-Core skips are explicit candidate boundaries: P2C3 through P2C10 require
unreleased stacked Core measured-impact contracts. The public P2C2 journey and every released
boundary remain green. The wheel contains 45 inert Domain Pack JSON/metadata files, no Python or
entry points, and retains `ace-core>=0.5.0,<0.6`.

Repository-wide Ruff remains an inherited release-hygiene blocker. With the same locked Ruff
version, both the P2C9 parent and this P2C10 worktree report exactly 14 lint findings and 12 format
targets. Scoped P2C10 checks and `git diff --check` are green. This packet does not rewrite
unrelated history, but release closeout must reconcile the repository-wide gate before publication.

## Claim boundary

BLS's public errata states that the sentence required a missing minus sign; the current archived
release already exposes the corrected form. The fixture explicitly derives the historical original
form from that erratum. It does not claim to preserve original response bytes.

The World review exact-loads the artifact and both immutable source Observations, derives source
coverage, correction linkage, prior-record preservation, corrected-form equality, and stale-form
absence, then records an exact observed result. Core and Intelligence receive only domain-neutral
records, conditions, scores, and classification. Historical replay requires no new authority, and
the proposal is non-effective, non-selectable, unapplied, and subject to separate human review.

This is one hermetic recorded BLS correction. It is not live monitoring, network-arrival evidence,
statistical validation of JOLTS, population correction performance, a general source-independence
claim, causality, general Brief quality, or human benefit.

## Remaining work

Independent Market reproduction, combined-main review/CI, public Core artifacts,
repository-wide lint/format reconciliation, security/release checks, and opt-in live transport
remain future bounded work. Core issue #49 F1, F3, and F5 still require explicit 0.6 release-owner
disposition; this World packet neither implements nor re-dates them.
117 changes: 117 additions & 0 deletions .../world-intelligence-p2c10-independent-correction-reproduction-work-packet-v1.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,117 @@
# World Intelligence P2C10 independent correction reproduction work packet (v1)

**Status:** stacked source-checkout candidate; this packet does not release World Intelligence,
apply a governance proposal, close ACE Core issue #38, pass SI4, or complete ACE 0.6.0.

**Frozen:** 2026-08-10 from World P2C9 commit
`53feadb40fcc93d23f326b16979ed6640471c4cf`, stacked on the Core exact observed-result
provenance candidate `433e3d16c5458c975557dcd1552824fb959d4d12`.

## Objective

Falsify source-family coupling by reproducing the measured correction journey over a materially
different real publisher and source policy while leaving the shared Core + Intelligence contracts
unchanged:

```text
BLS release Observation + later BLS erratum Observation
-> treatment corrected artifact / stale-form control
-> Decision -> reviewed Action -> independent review -> exact Outcome
-> useful / harmful / unproven evaluation -> proposal only
```

The source pair is the BLS Job Openings and Labor Turnover release `USDL-25-1087` and its public
erratum. The [archived release](https://www.bls.gov/news.release/archives/jolts_07012025.htm) now
contains the corrected `−39,000` form. The [BLS errata page](https://www.bls.gov/errata/) states
that the July 1, 2025 sentence required a missing minus sign and that corrections were made July 2.
The fixture therefore labels the pre-correction form as derived from that explicit erratum; it does
not pretend the current archive still exposes the superseded bytes.

## Frozen product policy and matched control

World owns `world_independent_official_correction_statement_quality` version `candidate-1`. A
review scores `1.0` only when all of the following are exact and inspectable:

1. both the original-release and correction Observations are present;
2. the correction names the exact release it corrects;
3. the original immutable record remains loadable and unchanged;
4. the rendered statement equals the public corrected statement; and
5. the reported pre-correction form is absent.

Otherwise the score is `0.0`. Treatment and control share both source references, correction
linkage, policy digest, two reviewed Action workflows, observation window, and matched conditions.
Treatment renders `The number of job openings decreased in federal government (−39,000).`; control
retains the missing-sign form `(39,000)`. Source coverage and action volume therefore cannot explain
the score difference.

The criterion requires two matched pairs and an effect of `1.0`. Intelligence derives useful,
harmful, or unproven under that frozen rule. Core appends exact reviews, Outcomes, evaluation, and
proposal history. A useful result maps only to a non-effective, non-selectable `promote` proposal
requiring separate human review.

## Exact acceptance

P2C10 must:

1. rerun P2C2 through P2C9 and preserve all prior immutable results and proposals;
2. admit the recorded BLS release and erratum as distinct exact Observations without network use;
3. retain BLS source vocabulary and historical-original derivation policy only in World;
4. append treatment and stale-form control artifacts that name the same exact source pair and
correction relation;
5. complete two reviewed treatment Actions and two reviewed control Actions;
6. exact-load each artifact and both source Observations before deriving the product score;
7. append four authenticated reviews and four Core Outcomes naming those reviews as observed
results;
8. classify `1.0` versus `0.0` over two pairs as useful and emit proposal-only `promote`;
9. replay without reauthorization and reproduce the fixture digest, scores, classification, and
proposal semantics across fresh hosts; and
10. reject drifted source linkage, duplicate source identity, changed statement forms, invented
scores, and fixture-policy drift.

Stacked Core tests remain authoritative for missing attribution, condition mismatch, cutoff
leakage, unavailable Outcomes, duplicate/replayed evidence, interruption, restart, and denied
authority. This packet exercises those unchanged contracts through a second real source family
rather than duplicating their lower-level tests in World.

## Ownership boundary

World owns BLS vocabulary, source URLs, the historical-original derivation disclosure, correction
statement policy, fixture, matched control, review contract, and product evidence. Core owns durable
identities, append-only records, provenance, Decisions, reviewed Actions, Outcomes, authority, and
replay. Intelligence owns domain-neutral conditions, matched evaluation, uncertainty,
classification, and proposal contracts. No BLS, JOLTS, errata, or minus-sign noun moves into Core or
Intelligence.

The installable Domain Pack remains inert JSON and unchanged. The Federal Register connector is
unchanged and is not used to fetch BLS. The fixture is recorded, hermetic, and network-free.

## Files, rollback, and deletion criteria

This packet owns:

- `domain_packs/tests/fixtures/p2c10_bls_correction_pair.json`;
- `scripts/p2c10_independent_correction_reproduction.py`;
- `domain_packs/tests/test_p2c10_independent_correction_reproduction.py`;
- the additive P2C9 state handoff; and
- this work packet, its audit, and restrained README/roadmap references.

It changes no shipped Domain Pack, connector, package version, dependency range, lockfile, release
record, Core contract, or public artifact. Rollback removes the harness, tests, handoff, fixture,
and candidate documentation. Records already persisted by a host remain immutable history.

Delete or replace this fixture only if BLS removes the public correction evidence, the exact URLs
cannot be independently verified, or a stronger redistributable snapshot supersedes it. Such a
change requires a new fixture identity and cannot rewrite prior evidence.

## Non-claims and next packet

This packet does not establish live monitoring, network freshness, the statistical validity of the
JOLTS estimate, general correction quality, population performance, source independence beyond the
two exercised families, causality, general Brief quality, or human benefit. It does not apply a
proposal or grant authority to a Domain Pack.

The next bounded falsification packet is an independent Market Intelligence reproduction of the
same public Core + Intelligence contracts. Combined-main review/CI, public Core artifacts,
compatibility and security checks, repository-wide hygiene, and separately reviewed opt-in live
transport remain release work. Core issue #49 F1, F3, and F5 still require explicit 0.6
release-owner disposition; this packet does not implement, defer, or re-date them.
Loading
Loading