# Public Gooo 0.6.11 and installed use

Observed 2026-10-08. Public prerelease:
https://github.com/kimjooyoon/meta-ontology-go/releases/tag/v0.6.11-dev

Source `27594824ad628ef0e8362bbc1e0e034aee88e07c`, Go 1.27.1,
decision-runtime SDK v0.2.26-experimental, clean source identity.

Readiness run 37775782496 executed all four native platforms on this merged
source. Publication run 37776480077 completed successfully. The annotated tag
points to that source. The nine public asset names, seven manifest payloads and
eight checksum entries were checked. Six downloaded files, including Darwin
arm64, matched their public hashes and byte sizes. The other three platform
archives were checked by public metadata only in this local observation.

The downloaded Darwin arm64 binary was installed as the local `gooo`; its bytes
matched the unpacked release. `installed-build.json` retains its identity.
`verify-publication.sh` is the actual verifier; to rerun it, download the six
files named by the script and rename `release-SHA256SUMS` to `SHA256SUMS`.

## Actual use with unchanged models

The separate Gooo standard library was run with the installed public compiler.
Its 17 functions matched 79 function-input cases and saved replay. The same
importing program ran under fixed and model ordering at budgets 1, 2, 4 and 8.
All 25 function/mode/budget summaries and 52 raw observations matched the clean
0.6.11 candidate: delivered inputs, actual/expected values, finite counts,
generated code and replay flags. Timing and platform binary hashes were excluded.

Both orderings still needed eight candidates. At that budget each matched all
10 consumer inputs and 30 named fields. Lower-budget partial results remain in
the archive. This is an existing-program compiler regression; model weights,
Gooo functions and fixtures were not changed.

The direct preview command used the graph QAT model with title `한글 예제`,
budget 20, and returned `{available:true,bytes:13,title:"한글 예제"}`. Saved replay
with an empty title and budget 1 returned
`{available:false,bytes:1,title:"untitled"}` with zero new predictions. These
input-only uses carry no finite correctness score.

The workbench's existing `verify --model builtin` also succeeded with the public
binary: standard functions 91/91 named outputs; diagnostics 24/24 in each mode;
starter 8/8 in each mode; invoice example 6/6; CI-plan example 24/24. Each saved
replay verified. The observer ran while only the compiler pin/documentation
edits were uncommitted, and its retained `repository_clean:false` is preserved.
This is a separate suite from the 17-function standard-library repository.

`installed-observations.tar.gz` retains full records for both suites and the
direct preview uses. Compiler binaries and copied model weights are excluded.
The summary files and comparison output make the denominators inspectable.

Main promotion and downstream adoption have their own PR/CI records. Public
release completion does not imply those separate merges are complete.