Skip to content

Promote nightly → main: wire-42 bundles, example re-pins, taxonomy docs - #10

Merged
scott-wyatt merged 3 commits into
mainfrom
nightly
Nov 18, 2026
Merged

scott-wyatt merged 3 commits into
mainfrom
nightly

Conversation

@scott-wyatt

Copy link
Copy Markdown
Contributor

What

Three commits on nightly, promoted together:

  • b2d7cdc snapp: re-emit aph@0.1.0 and aph_guardrails@0.1.0 on nlang 18b7564e3 (install 14, wire 42) and re-pin the 102 N Lang examples. Install 14 refuses the wire-41 bundles with a schema-hash mismatch, which made every example package fail nlang check. Both bundles were rebuilt in place under one named epoch, two runs each, byte-identical across the pair. Neither bundle declares @deps, so the engine's empty-Mod defect for dependent Snapps does not apply; both binaries grew (230139 → 259035 and 504768 → 576700 bytes). Swept after the re-pin on install 14: 102 of 102 check, 307 tests, 0 failures. The engine's new per-package .n/last_diagnostics.json joins the ignore list.
  • 6bf0477 elixir: refresh the NIF Cargo.lock for aph-core's sha2 edge (one line; mix test kept regenerating it).
  • 5dac57d docs: the error taxonomy is twenty-four codes, not sixteen (AGENTS.md invariant 4 and the spec skill's heading).

Gates this PR relies on

The conformance crate's pin weld (nlang_examples_pin_test) hashes the committed bundle bytes and compares every example's integrity; it runs in the Rust job here. snapp_freshness_test needs nlang on PATH and skips in CI, so the byte-for-byte two-run check above is the freshness evidence for this change.

Follow-up, owner step

The live vocabulary digest record for aph_guardrails still carries the bundle bytes as of 2026-11-08; every re-emit since has changed them. One re-bless after this lands.

Scott Wyatt added 3 commits November 17, 2026 09:34
AGENTS.md invariant 4 still said the taxonomy was a closed set of sixteen,
APH_E001 through APH_E016. The §11 table in spec/aph-0.1.md has carried
twenty since RFC 0003 and RFC 0005 landed (E017 AudienceMismatch, E018
EnvelopeAlreadySpent, E019 EnvelopeWindowInvalid, E020
RecipientClassNotAllowed), and the v0.2 delta declares four more (E021
through E024, the seal and rotation-attestation refusals) that aph-core
implements. An agent reading the old sentence would refuse valid codes.

The spec skill had the same drift in miniature: its heading said fifteen
over a table of twenty rows.
… wire 42) and re-pin the examples

Install 14 refuses the wire-41 bundles outright: a schema-hash mismatch
(0x6986c710f96240d9 written, 0x7e52c38c82af89ec current) on every
`nlang check` that reads them, which is every one of the 102 example
packages. Both bundles are rebuilt in place with `nlang build`, run from
inside each package so source spans stay package-relative, under one
named epoch (SOURCE_DATE_EPOCH 1794935521), two runs each, byte-identical
across the pair; `nlang export` refreshed the JSON faces (abi 41 -> 42).
No copy of the wire-41 files was kept.

  aph@0.1.0.nbin             230139 -> 234435 bytes  sha256 125e2d3a...
  aph_guardrails@0.1.0.nbin  504768 -> 509067 bytes  sha256 ef9a2303...

Neither bundle declares @deps, so install 14's empty-Mod defect for
dependent Snapps does not apply, and the sizes agree: both grew.

The 102 examples pin the aph bundle by sha256, so their `integrity`
moves in this same change (examples/nlang/README.md, "When the bundle is
re-emitted"); the conformance weld fails otherwise. Swept on install 14
after the re-pin: 102 of 102 check, 307 tests, 0 failures.

The engine now writes `.n/last_diagnostics.json` beside each package
config on every check; it is regenerable and joins the ignore list.
aph-core already depends on sha2; the NIF's committed Cargo.lock lacked
the edge, so every `mix test` regenerated the file and left the checkout
dirty. One line, produced by cargo during today's binding runs.
@scott-wyatt
scott-wyatt merged commit d14335a into main Nov 18, 2026
9 checks 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