Roadmap
The roadmap is a set of stages, not dates.
| Stage | Scope | Status |
|---|---|---|
| 0 | Best-practices framework | Done |
| 1 | Scope, partition tree, segment planner, survey | Done |
| 2 | SQLite coverage ledger | Done |
| 3 | Full-read orchestrator and briefs | Done |
| 4 | Deterministic labels, FTS, coverage-aware query | Done |
| 5 | Polling watcher, delta, audit | Done |
| 6 | Raw-read enforcement in Hermes | Done |
| 7 | Workspace virtual repository survey | Done |
| 8 | Documentation baseline | Done |
| 9 | Validation on a model and a quality set | Done |
| 10 | Native watcher and workspace full read | Done |
Stage gates
- Stage 1: a survey on three repositories with different sizes.
- Stage 2: a ledger sync from the JSONL index and a query over FTS.
- Stage 3: a plan, read, done, and verify cycle on a temporary repository.
- Stage 4: a query result that reports read, unread, and stale state.
- Stage 5: a watch scan and an audit repair.
- Stage 6: a blocked Hermes read and an override.
- Stage 7: a workspace survey with a multi-repo plan.
- Stage 8: a light README and a docs tree with resolving links.
- Stage 9: a fixed question set and a measured answer quality.
- Stage 10: a design review before code.
Stage 9 has a fixed eight-question set on Auwai itself, an Auwai arm at 30/32,
a vanilla control arm at 28/32, and a full-window segment validation at 64K
with resolved citations. The evidence is under benchmarks/quality/results/.
The full-window run used a local endpoint; no public endpoint is required.
Stage 10 has a design review in stage-10-design.md, an inotify backend with a polling fallback, an FSEvents backend, and a workspace full-read runner with a workspace coverage ledger.
Working plan
The current results are in Results. The measures are in Metrics.