Skip to content

docs: the handoff after P5.3 merged, with CI deferred - #91

Merged
richardkiene merged 1 commit into
mainfrom
docs/handoff-post-p53-merge
Aug 21, 2026
Merged

docs: the handoff after P5.3 merged, with CI deferred#91
richardkiene merged 1 commit into
mainfrom
docs/handoff-post-p53-merge

Conversation

@richardkiene

Copy link
Copy Markdown
Contributor

Brings docs/status/START-HERE.md current after PR #90 merged, and records the maintainer's 2026-08-21 decision to defer CI until there is a working, usable version.

  • P5.3 is merged: fe27646, parents b0a8ea1 + ac365a6, a true merge. All 13 branch commits verified reachable from main, which is what keeps this file's and the evidence document's SHA citations valid.
  • CI is deferred, and it outranks the queue. The bar is a green local run plus actual local usage. Queue item 1 (flake mechanisms) is now explicitly marked as not what to do next.
  • What "usable" means is named: P5's second exit clause, gascan-e2e on arca, with the ignore-attribute counts re-derived today and labelled as attribute counts rather than test counts.
  • A derivation became an observation. The claim that neither conformance result is reproducible in CI was reasoned; PR P5.3: the backend conformance suite, and what it measured on each backend #90's run 32437752256 now shows it directly, and also shows THE TENTH MECHANISM is flaky rather than constant.
  • Two anchors corrected rather than copied forward, including the design's off-by-one on backend_selection.

Docs only — no code touched.

…nd clause

Four updates to the cold-start block, all re-derived on 2026-08-21.

P5.3 is merged. PR #90, merge commit fe27646, parents b0a8ea1 + ac365a6 --
a true merge, not a squash. `git merge-base --is-ancestor` confirms all 13
branch commits are reachable from main, which is why merge-only matters here:
this file and docs/evidence/2026-08-20-backend-conformance.md cite them.

The maintainer deferred CI on 2026-08-21, and that outranks the queue. The bar
is a green local run plus actual local usage. Item 1 of the queue -- the flake
mechanisms -- is explicitly not what to do next, and the queue now says so.

What "usable" means is named concretely: P5's second exit clause, the
product-level gascan-e2e suite on arca. MEASURED by `grep -c '#\[ignore'` over
crates/gascan-e2e/tests/*.rs -- apple 11 attributes (apple_apply 8,
apple_lifecycle 1, apple_recovery 1, apple_security 1), arca 3 (arca_engine 2,
arca_startup 1); labelled attribute counts, not test counts, because the P5.3
design's §5 cites 24 and 6 from an earlier revision. Both `command()` builders
it names were verified present: apple_common/mod.rs:961, arca_common/mod.rs:303.

A derivation became an observation. The claim that neither real-backend
conformance result is reproducible in CI was reasoned, not seen. PR #90's run
32437752256 now shows it: the engine job's live-tier step reported
`0 passed; 26 failed; 3 filtered out; finished in 0.06s`, and
conformance::backend_contract_holds_on_arca panicked at
crates/gascan-arca/tests/live/common/mod.rs:148:48 on GASCAN_ARCA_BASE_OCI_LAYOUT
-- the environment, before reaching any backend. That run also passed the Swift
step that is red on main, so THE TENTH MECHANISM is flaky rather than constant
and a red `engine` job now has two possible steps behind it.

Two anchors corrected rather than copied. The P5.3 design's §6 cites
crates/gascan-core/src/backend.rs:167 for backend_selection defaulting to Apple;
:167 is the Arca arm and the (false, false) => Apple arm is :168, in the function
opening at :156. That trap does not apply to conformance and applies fully to the
product-e2e work, which is why it is recorded where that work is assigned.
@richardkiene
richardkiene merged commit 899a29f into main Aug 21, 2026
1 check passed
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