# Gooo 0.6.10: published, installed and used to edit Gooo

Published [v0.6.10-dev](https://github.com/kimjooyoon/meta-ontology-go/releases/tag/v0.6.10-dev)
on 2026-10-08, exact source `eb0dc4704705147f9ba944db3df2ba5e3225cbff`.
Merged-dev readiness run 37762608339 and publication run 37763172067 succeeded.

## Public file checks and installation

The public release has nine assets and is a non-draft prerelease. Its annotated
tag resolves to the exact dev source. Seven manifest payload records and eight
checksum entries matched the public asset names, digests and sizes.

The macOS arm64 archive and five metadata/checksum files were downloaded and
their actual bytes matched the public metadata. The other three platform archives
were compared by metadata in this local observation; the readiness and publication
workflows checked their native payloads. `downloads/verification.txt` and
`downloads/verify-publication.sh` record this scope. To rerun the verifier, download
the arm64 archive into `downloads/` alongside the retained files, then run the script
from that directory. `downloaded-file-sha256.txt` retains the archive and binary hashes.

The downloaded executable reported version 0.6.10-dev, clean VCS source,
Go 1.27.1 and decision runtime v0.2.26-experimental. Its exact bytes replaced the
local installed compiler. `downloaded-build.json` and `installed-build.json` retain
the two observations.

## Installed-binary source-splice regression

Using Workbench source `8fab0f6f4a2835b639afa2f9a6fa7e4fa281dd8b`, the installed
public binary reran the frozen `experiments/dependent-splice` study. This workbench
commit updates the compiler pin and usage documentation; experiment code, fixtures
and model weights remain unchanged from the original published study.

| Attempt budget | Fixed-order input cases | Graph-model input cases |
| --- | ---: | ---: |
| 1 | 321/542 | 321/542 |
| 2 | 321/542 | 321/542 |
| 4 | 321/542 | 542/542 |
| 8 | 542/542 | 542/542 |

All saved replays preserved the corresponding actual outputs, including partial
results. Complete programs matched 1,626/1,626 output fields. This is one already
observed program with 542 input tuples, repeated across modes and budgets. It is a
release regression, not new evidence of generalization. No model was retrained.

The two full-budget tool runs also returned edited Gooo source, changing
`input + 1` to `input * 2`. Compiling that returned source and running -5, 0 and 9
produced -10, 0 and 18 in both modes (3/3 each). Native execution and saved replay
made zero new model calls. The compressed original study includes source, actual
values, saved execution records and edited-program runs; it contains no compiler
binary. `study-summary.json` retains the model and source hashes.

The runner's independent reference function and original frozen plan are in
the public Workbench repository. Reproduce from that repository with:

```sh
go run ./experiments/dependent-splice --compiler /path/to/installed/gooo --out out/release-0610-study
```

At this observation, main promotion PR 1380 remains under its own CI. Release,
installation and main promotion are tracked as separate operations.

Later update: main promotion and Workbench adoption completed. See
[the subsequent completion record](../release-0610-completed-20261008/README.md).