# Scalar identity candidate — 2026-10-09

Public candidate: [PR 1418](https://github.com/kimjooyoon/meta-ontology-go/pull/1418),
source `28b84915a322b86a2ea0676c8b12c34f1ac1a947`. Canonical CI
`37901769992` passed. Verified normal merge to dev
`832dbf3684b104b03101f41beb8f2c5efc517ec9` preserves the exact tested tree, with
sole previous-dev parent. This feature is separate from public release 0.6.20.

`final-proof/` retains original canonical artifact `11603048841`, 11,186 bytes,
SHA256 `ef918d997ded8a75f9dea0b81716c086f521d107e4c4020048e2f8e45a82333a`.
The verifier requires PASS for all six current checks, exact head/base refs, the
frozen PR body and original provenance. Release20 publication and main promotion
completed before this dev merge. Post-merge CI is a separate observation.

The clean release20 source rejected a renamed Integer with the unchanged canonical
ID at code generation, although its semantic plan stayed the same. The original
error, JSON and exit status remain in `canonical-integer-alias-red.*`.

The implementation resolves three recognized scalar IDs before native codegen,
retaining authored type names. Full affected race passed for bodycodegen,
bodyexecution and workspaceexecution (36.561s, 166.426s, 63.164s), with affected vet.
Those logs cover the implementation before the final example/regression additions.
The final typed predicate regression race and full repository build also passed.

`readiness/` retains successful PR run `37901770176`, four platforms of the
existing release profile plus aggregate/eligibility. Original artifact
`11603100482` is 1,046 bytes, SHA256
`e8a753a9eb0b53bf85a7e2305212bb8a8c10d4e8a4d9a8ca90f73d193c58d43f`;
eligibility is exact/closed 7/7. The new scalar-name native examples below were
observed locally; the release profile exercises its existing public examples.

## One field: a representation decline

`one-field-declined/` retains clean source
`88f2c1286314e64816bc329f701f1e38dd3e3b54`. Both modes executed four native
cases and saved replay successfully. The model was loaded but made zero inference
calls: `model_context.reason` was `THREE_FIELD_CHOICES_REQUIRED`. The dogfood
script's assertion requiring one call failed, so its later after-hash file was
never created. This observation is preserved as a model contract decline.

## Three fields: actual own-model construction

`three-field/` retains clean final source, the authored fixture, cases, raw mode
reports, saved replay and timing. Three source-owned choices supply text, active
and exact count values. Four authored cases provide 12 field checks, followed by
four distinct native input tuples, including +/-9007199254740993.

| Observation | Fixed order | Own model |
| --- | ---: | ---: |
| First candidate field checks | 4/12 | 7/12 |
| Candidates tried | 8 | 6 |
| Selected field checks | 12/12 | 12/12 |
| Distinct native cases | 4/4 | 4/4 |
| New assembly inference calls | 0 | 1 |
| Saved replay inference calls | 0 | 0 |

Generated Go SHA and native traces agree. The reports' input-separation receipt
confirms zero overlap with source selection cases; model-training exposure remains
unknown. These observations describe this supplied program, not general accuracy.
The unchanged model metadata SHA is
`3c68205a660695103712e2115ce90accd298b406dced8ad5ad5c9b84bab0a202`;
weights SHA `76f68845a03ed8c8bc261a57c919e96dcdd352d3c35bcdaeb3d9b51a88975c6f`.

Model setup was 0.448958ms, prediction 67,833ns, resident tensors 2,096 bytes.
Whole commands observed 0.50s fixed and 0.32s model, approximately 0.25s process
CPU each. Maximum RSS was 85,983,232 / 86,114,304 bytes respectively. These are
single sequential observations including native builds and runs, with cache/order
effects. No host CPU delta or speedup benchmark was measured.

`SHA256SUMS` covers this retained bundle. Binaries, executable build folders and
model weights are omitted. Original report content and exact integer values are
preserved. The original raw reports carry local execution paths.