# Gooo 0.6.10 candidate: four native platforms

Observed on 2026-10-08. Source `63a39565213e2293468dc354bcc26f022657e78a`,
[PR 1379](https://github.com/kimjooyoon/meta-ontology-go/pull/1379),
[readiness run 37759864792](https://github.com/kimjooyoon/meta-ontology-go/actions/runs/37759864792).
All four native jobs, aggregation and eligibility completed successfully.
This record describes a PR candidate; it does not establish publication or installation.

## What ran

Each hosted platform built and executed its own candidate: Linux amd64, Windows amd64,
macOS amd64 and macOS arm64. Receipts bind Go 1.27.1, clean source identity and the
0.6.10-dev CLI version. Repeated binary and archive builds matched on each platform.

| Observation | Inputs in each pass | Passes per platform |
| --- | ---: | ---: |
| Integer division | 8 | construction and saved replay |
| Retry policy | 12 | construction and saved replay |
| Filename operations | 12 | construction and saved replay |
| Actual package text input | 3 | construction and saved replay |

`native-recount.json` independently compares actual and expected values from the
24 native composition records: 256/256 case observations. These are 32 inputs
repeated across four platforms and two passes, not 256 different inputs.
Integer comparisons preserve JSON numbers without converting them to floating point.
Saved replay preserved the generated program and used zero new model calls.

The package path additionally returned the three exact records checked by
`verify-packages.sh`, including Korean text and an emoji. It preserved generated
code across saved replay. This is an input-only usage path: the CLI correctly keeps
its own finite score at 0/0 and reports `OBSERVED`. The external release check compares
the actual values; it does not turn these inputs into CLI-supplied expectations.

All four graph exports have the same source and context hashes, 115 nodes, zero
model predictions and zero candidate evaluations. This checks the source-to-model
input representation; it does not measure model selection quality.

Eligibility closes seven release evidence cells, including the separate structural
corpus of 26 cases and 39 indicators. These counts describe release checks, not
language intent accuracy or additional program families.

## Retained evidence

The four native artifact ZIPs were downloaded and their bytes matched the API
SHA-256 digests. `artifact-download-verification.json` records those comparisons;
`artifacts.json` and `run.json` retain their source metadata. Only raw observations,
receipts and small verification files are retained here, without duplicate binaries.
The aggregate binary bundle was not independently downloaded for this record.

From this directory:

```sh
shasum -a 256 -c SHA256SUMS
sh verify-packages.sh
```

`observe-composition.go` is the independent native-output recount tool. The JSON
records retain the original values and execution metadata for further inspection.
No model training or new program-generalization experiment was performed here.