Last updated: 2026-05-21
This document defines the current candidate baseline for freezing CodeWell V1.
It is not a public launch note.
It exists to separate three things that should not be conflated:
- a stable internal V1 baseline
- a public open-source release decision
- later V2 research work
Current working branch:
feature/precision-governor-v1
Current preserved pre-V1 snapshot:
- branch:
snapshot/pre-precision-governor - tag:
pre-precision-governor-2026-05-19
Current V1 freeze candidate should be understood as:
- the state after the recent retrieval-policy refinements
- the current TS/JS real-project evaluation coverage
- the completed
strong_multi_goalrepeated-run protocol
The current candidate now has:
- a lightweight local-first product direction
- no required LLM API, vector DB, hosted service, or reranker
- maintained fixture, self-eval, package, and real-project evaluation paths
- internal TS/JS agent-eval evidence with bucketed interpretation
- a completed
3 x 2repeated-run read for thestrong_multi_goaltask slice - a product-facing V1 evidence summary in
docs/V1_EVALUATION_NOTE.md
Reasonable V1 claim:
CodeWell is a selective navigation aid for coding agents on repository tasks with real path ambiguity.
Reasonable V1 non-claim:
CodeWell is not yet a universal speedup layer across all coding tasks or all prompt shapes.
Freezing V1 should mean:
- preserve this state as the comparison baseline for V2
- preserve the current evaluation interpretation alongside the code
- stop changing the core product thesis for V1
- allow later V2 work to branch cleanly without rewriting the V1 evidence story
It should not automatically mean:
- publicly launch immediately
- stop improving docs or release packaging
- claim benchmark-level external validation
Before declaring the baseline frozen in practice, the remaining work is mostly packaging and release-facing clarity:
- update release-facing notes to reference the V1 evidence summary directly
- decide whether to create a dedicated V1 branch or tag such as
v1-open-source-baseline - run or re-run the full release gate when the working tree is ready
- capture the final freeze summary alongside the chosen branch/tag
The next concrete action is:
- create the final freeze branch/tag for this V1 candidate
- store the final summary docs beside that snapshot
- start V2 work on a separate branch
docs/V1_EVALUATION_NOTE.mddocs/RELEASE_READINESS.mddocs/AGENT_EVAL_SUMMARY.mddocs/AGENT_EVAL_CONCLUSION.md