Evidence

This is the bounded record behind the acceptance-layer thesis. The public projection is a signed calibration summary, not the primary governed receipts or a public rerun of the acceptance verdict. The underlying receipts and replay tooling remain private, not publicly accessible; access requires separate authorization.

The code mechanisms, historical PR references and limits are indexed on the proof file.

Signed by the producer and verified here against the configured public key. That check proves these exact bytes are unchanged since signing. It does not prove key provenance beyond this site's configuration, an independent audit of the run, or a re-run of the evaluation - it is a calibration summary frozen on September 11, 2026. Anything missing or invalid renders as unavailable, never as a claim. The static snapshots below remain the reported historical record.

Source revision
85442b8032cb7bae72866dfd7782234a98dd2fb7
Baseline
e4a6c73c0f4298b3bf06caf55b318a48becaaa91
Target dimension
performance.nesting_ceiling
Result
FAIL → PASS
Protected dimensions
api.default_changed, api.export_removed, api.serialized_field_break, api.signature_break, operations.external_call_timeout, operations.retry_bound, performance.io_in_loop, security.secret_to_observable_sink, security.untrusted_to_command_sink, test.changed_symbol_link_coverage, test.high_complexity_link_coverage
Native tests
PASS (26854 passed, 0 failed, 0 errors, 0 skipped)
Verdict
COMPLETED; floor ADMITTED; acceptance PASS
Internal status
AUTHORITATIVErecorded authorityState; the verdict does not name a person
Timestamp
2026-09-11T11:18:34Z
Redaction manifest
none
Lane
CALIBRATION
Freshness
frozen historical record

Signature verified against the configured public key - these signed bytes pass the integrity check. This does not independently establish key provenance, audit the run or reproduce its acceptance decision.

The projection above is data, not a screenshot. Download it, recompute its digest, and check the signature yourself against the projection public key this site is configured to trust. You need jq, sha256sum and openssl - preinstalled or a package on any macOS or Linux box.

Projection: signed-projection-2026-09-11.json
Projection public key (Ed25519, base64 SPKI): MCowBQYDK2VwAyEA4SewVRRH0cHh4Ib5yCIh+G8l5f0fhb33d3S5dsjboOE=

recompute the digest, verify the signature
# Save the projection (also linked above)
curl -sSLO https://madebyapex.dev/evidence/signed-projection-2026-09-11.json

# Recompute the evidence digest: canonical JSON (keys sorted, no
# whitespace) with the signature and evidenceDigest fields removed.
# The result must equal the evidenceDigest field.
jq -cjS 'del(.signature, .evidenceDigest)' signed-projection-2026-09-11.json | sha256sum
jq -r .evidenceDigest signed-projection-2026-09-11.json

# Verify the Ed25519 signature over the signed bytes: canonical JSON
# with only the signature field removed.
jq -cjS 'del(.signature)' signed-projection-2026-09-11.json > canonical.json
jq -r .signature signed-projection-2026-09-11.json | base64 --decode > signature.bin
echo 'MCowBQYDK2VwAyEA4SewVRRH0cHh4Ib5yCIh+G8l5f0fhb33d3S5dsjboOE=' | base64 --decode > projection-key.der
openssl pkeyutl -verify -pubin -inkey projection-key.der -keyform DER -rawin -in canonical.json -sigfile signature.bin
# Expected: Signature Verified Successfully

What this proves: these exact bytes were signed by the holder of the projection private key this site is configured to trust, and the recomputed digest is the key under which this site files the projection against its receipt record - so the filing provably covers these exact bytes. What it does not prove: key provenance beyond this site's configuration, an independent audit of the run, or the receipt itself. It does not re-run the evaluation. Fresh-clone replay is reported in a separate private account, and this projection is a signed calibration-lane summary, frozen historical. The development-only primary receipts, mirrors and replay tooling are not publicly accessible; access requires separate authorization.

The public signed projection, verifier and annotated explanation are downloadable. The underlying governed receipts and mirror remain private, not publicly accessible; access requires separate authorization.

  • The offline verifier kit - a single file that recomputes the evidence digest and checks the Ed25519 signature with the published projection key. Runs entirely in your browser, works offline, sends nothing anywhere. Check the published summary yourself; this does not verify the private primary receipts.
  • The annotated summary - an annotated explanation of the signed fields and intended workflow, including the reported 26,854-test PASS. It is not an authenticated chronology or a replay of private stage receipts.
  • The signed projection itself - the frozen calibration record as raw JSON. Feed it to the verifier kit, or to your own tooling.
  • Reproduce the check with Node or download the kit. The zip holds the offline verifier, this signed projection, the pinned public key, and a README. Node only. No network.
ClaimLaneVerify nowPrivate · access not established
The private calibration account reports an earlier operations.retry_bound floorCalibrationReported account below; no public primary replaySealed receipt, governed mirror
The signed summary: performance.nesting_ceiling FAIL → PASS, 11 protected dimensions listed, source revision as recorded in the projection. The earlier candidate reference eb13c30a is not authenticated by that artifact.CalibrationSigned projection above - recompute the digest and verify the signature yourself (timestamped 2026-09-11)Governed pull requests; evaluator definitions
Native acceptance in the signed summary: 26,854 passed, zero failuresCalibrationStated inside the signed projectionFull run logs
Private same-engine re-evaluation reported a match on the recorded verdicts. Not reproducible from this site.CalibrationNothing public yetReplay tooling and sealed receipts
Full campaign loop against a live distributed system (Astronomy Shop)Historical feasibilityNothing public - narrative onlyCampaign records
UpliftA reported calibration uplift on a mirror of an external repository: on a private governed mirror of pypa/packaging, the signed-summary candidate moved performance.nesting_ceiling from FAIL to PASS and lists 11 protected dimensions. The list itself does not independently demonstrate their preservation. Externally owned codebase - not an externally accepted improvement, and not custody of anything.
RatchetThe private calibration account reports an earlier operations.retry_bound floor and reversal checks. The public projection records floor ADMITTED for its reported result; it is not an authenticated chronology of those private events.
ReplayFresh-clone re-derivation, reported in a private record: a same-engine re-evaluation on a fresh clone reproduced the evaluation digest and the recorded verdicts. Not reproducible from this site.
AcceptanceTarget-native test result: 26,854 passed, 0 failures, errors or skips, as reported by the signed summary. Its acceptance PASS field does not establish native Apex Review ACCEPT or independent reproduction of the run.
$ apex cycles --target pypa/packaginglane: development_only (calibration)
earlier run
[sealed]floor raised: operations.retry_bound
later run [sealed]nesting_ceiling FAIL→PASS · 11 protected dimensions listed
replay [private]same-engine re-evaluation · recorded verdicts (private record)

A signed sample fixture showing the rendering path only. Not a verified outcome.

SAMPLE FIXTURE - demonstrates the rendering path only; not a verified outcome

Source revision
sample-revision-0000000
Baseline
sample-baseline
Target dimension
sample-dimension
Result
sample-before → sample-after
Protected dimensions
sample-protected-a, sample-protected-b
Native tests
sample-native-tests-pass
Verdict
sample-verdict
Internal status
SAMPLE
Redaction manifest
internalRunbook
Freshness
sample fixture

Sample fixture verified against the quarantined sample key. Demonstration only - not a verified outcome.

CalibrationThe pypa/packaging cycles are calibration-lane runs; their receipts are marked development-only. They demonstrate the floor and ratchet machinery on a private mirror of an externally owned repository - not a sealed fix of a real deficiency or acceptance by that repository's maintainers.
HistoricalThe OpenTelemetry Astronomy Shop campaigns are historical feasibility evidence: the full campaign loop against a live distributed system, not yet a reproducible public proof campaign.
AccessThe governed mirrors, sealed receipts and replay tooling are private. No authorized access procedure is established here. Any future disclosure requires its own custody and authorization decision.

Narrative version: Governed calibration on pypa/packaging →