Repository navigation
protocol
Purpose: use SDK0.2.40's actual research artifact through the new compiler adapter. This is an integration observation. The example has already appeared in the canonical study; its outcomes are not new learned-quality evidence.
Freeze a clean local compiler commit after its checks finish and record that SHA before building. Use Go1.27.2 and the existing, once-verified SDK0.2.40 release download. Do not download, fit or run the original study again.
Inputs:
-
example.goooSHA2568775f9e0f88813e3342faa0fa9c2132e32f70c11f36956e2984853cfba3cef62. - Canonical model SHA25674abb7c3d8bbc1fa90be22f48cf26cef5e00cbd72cbb8f8689293ce6888a5d37.
- Reuse its eight authored output pairs as both caller construction and native evaluation cases for entry Choose. Preserve its three source condition cases. There are no new independent evaluation inputs in this observation.
Execute once in this order:
- Build a source-bound local CLI and record its version/build metadata and hash.
- Inspect the source with the actual canonical model, expecting zero predictions and candidate executions and the canonical source arrays.
- Construct natively with the model, eight caller cases, the same eight evaluation cases and at most four caller attempts. Retain all JSON and saved source/results.
- Construct the same source with fixed ordering and the same budget/cases.
- Replay the saved model construction without supplying a model. If step3 fails, record the dependency failure and do not invent a replay result.
Use separate new output directories. Capture each command, exit status, complete stdout/stderr and macOS time resource output. Do not retry a failed observation. The model's prediction count/timing and the full CLI cost have different scopes. The chosen order and host load are uncontrolled, so elapsed values cannot establish a model speedup. There is no instantaneous computer-wide CPU-utilization sample. Read-only audits and normal regression tests can inspect these retained records.
- 문서 홈
- 현재 상태와 지원 범위
- 실패를 기록하며 조립을 이어가는 0.6.30
- 업무 규칙 예제와 실제 도입 기준
- 요구를 바꾼 뒤 결과 비교하기
- 결과 조회와 자체 모델의 0.6.28 배포
- 저장 결과를 비교하는 0.6.27 배포
- 작은 모델과 조립하는 0.6.26 배포
- 중간 조건을 확인하는 0.6.25 배포
- 분기 구성·재실행의 0.6.24 개발 후보
- 작은 도구로 발전시키는 Gooo
- 흐름도로 보는 Gooo
- 시작하기
- 언어 둘러보기
- 사용 흐름
- 작은 모델과 메타프로그래밍
- 우리 PC의 자체 경로 decision 모델
- 조건을 읽는 자체 모델과 실제 탐색
- SDK 0.2.39 자체 모델 받기
- 모델 선택 점수와 실제 실패
- 설명·변수 이름 변화의 27회 관측
- 작은 모델의 학습 범위를 넓힌 결과
- 같은 계산을 같은 모델 입력으로 읽기
- 사례를 통과해도 의도가 다를 때
- 사례를 모아 읽는 방식을 바꿔 보기
- 자체 결정 모델 받기 · Hugging Face
- 비교와 분기를 따로 배우면 나아질까
- 소스에 적는 중간 조건과 선택
- 지표와 증거
- 참고하고 감사하는 연구
- 최근 연구와 다음 작은 기능