Skip to content

Worked example: Case D (post-commit divergence) for horizon-shield#51 pilot - #139

Merged
giskard09 merged 1 commit into
mainfrom
worked-example/horizon-shield-pilot-case-d
Oct 18, 2026
Merged

giskard09 merged 1 commit into
mainfrom
worked-example/horizon-shield-pilot-case-d

Conversation

@giskard09

Copy link
Copy Markdown
Owner

Summary

Proposed fourth case for the live pilot scoped in ogasurfproject-jpg/horizon-shield#51 (aeoess, ogasurfproject-jpg with HORIZON SHIELD real, babyblueviper1 — one paid audit call, three sets of receipts: authority+payment, APS, acceptance). The issue explicitly invites this: "Any other project that wants to run its approach on the same workflow is welcome; the inputs below do not change for that."

The gap

The pilot's own cases A/B/C are all about the moment of the chain of receipts:

  • A: within/paid/accepted
  • B: acceptance fails (mismatched deliverable)
  • C: budget exceeded

None answer: what if the JCCDB cost database an estimate was audited against changes after A already closed validly? That's exactly post_commit_consistency/POST_COMMIT_DIVERGENCE, shipped earlier the same day in a different domain (execution-join-ref-v1, #137, file-write approvals for astrogilda).

What this is

examples/conformance/execution-join-ref/horizon-shield-pilot-case-d/ composes that invariant onto the request_occurrence/task/execution-receipt object shapes frozen in COMPANION_v0.md/PILOT_v0.md (commit 4d720ace, MIT — cited, not vendored, no code of theirs is imported).

Explicit scope limit (this is the point, not a shortcut): authority_and_payment (MUSUBI/APS) and acceptance (babyblueviper1's work-acceptance-record) are declared, not reimplemented. This is a fourth, complementary check, not a competing implementation of the three roles the pilot's own participants already cover.

Vectors

  • case-a-within-paid-accepted-001: analogue of the pilot's own Case A, shows the new invariant doesn't interfere with the happy path.
  • case-d-jccdb-snapshot-divergence-001: the new case. Same request/task/execution-receipt as A — the original call was genuinely within, paid, accepted. ~24h later an independent observation of the JCCDB snapshot gets a different response_sha256. canonical_envelope, chain_integrity, authority_and_payment, acceptance all stay true; only post_commit_consistency fails — POST_COMMIT_DIVERGENCE.

Also adds a cross-reference from docs/spec/execution-join-ref-v1.md to this worked example.

No PR to their repos — same pattern this repo always uses for cross-project worked examples.

Reproduce

python3 examples/conformance/execution-join-ref/horizon-shield-pilot-case-d/verify.py

Test plan

  • verify.py → 2/2 pass
  • Mutation tests: fixing Case D's observed_hash flips post_commit_consistency to true (and correctly flags the now-stale declared false); breaking Case A's task_sha256 binding correctly fails canonical_envelope/chain_integrity
  • execution-join-ref's own 14-vector suite re-verified unaffected (14/14)

… pilot

Proposed fourth case for the live pilot scoped in
ogasurfproject-jpg/horizon-shield#51 (aeoess, ogasurfproject-jpg,
babyblueviper1 -- one paid audit call, three sets of receipts). The
issue explicitly invites this: "Any other project that wants to run its
approach on the same workflow is welcome."

The pilot's own cases A/B/C are all about the moment of the chain of
receipts (paid+executed+accepted; acceptance fails; budget exceeded).
None answer: what if the JCCDB cost database an estimate was audited
against changes AFTER acceptance already closed validly? That's exactly
post_commit_consistency/POST_COMMIT_DIVERGENCE, shipped earlier the same
day in a different domain (execution-join-ref-v1 PR#137, file-write
approvals for astrogilda).

examples/conformance/execution-join-ref/horizon-shield-pilot-case-d/
composes that invariant onto the request_occurrence/task/
execution-receipt shape frozen in COMPANION_v0.md/PILOT_v0.md (commit
4d720ace, MIT, cited not vendored -- no code of theirs is imported).
Explicit scope limit: authority_and_payment (MUSUBI/APS) and acceptance
(babyblueviper1's work-acceptance-record) are DECLARED, not
reimplemented -- this is a fourth, complementary check, not a competing
implementation of the three roles the pilot's participants already
cover.

2/2 vectors pass: case-a (happy path, shows the new invariant doesn't
interfere) and case-d (the new case -- every other invariant stays true,
only post_commit_consistency fails).

No PR to their repos -- own worked example citing their frozen inputs,
same pattern this repo always uses.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@giskard09
giskard09 merged commit 9e28cdd into main Oct 18, 2026
4 checks passed
@giskard09
giskard09 deleted the worked-example/horizon-shield-pilot-case-d branch October 18, 2026 15:28
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.

1 participant