Learning roadmap
Every non-obvious concept this project runs on, with one canonical home each.
Every non-obvious concept this project runs on, with one canonical home each.
Entries follow one shape: what (one line), why it matters here,
gotcha from this codebase, external source. Deep explanations live at
the sources; project behavior lives in docs/ — these pages only bridge the two.
Pages
| Page | Covers |
|---|---|
| new-things.md | Platform + stack: Tauri 2, IPC, CLI-agent subprocesses, ast-grep, agent talks, Rust traits/rusqlite, provider APIs, auto-updater, PostHog, DORA, calibration, workspaces, CF Pages, GEO, release chaining |
| telemetry-and-indexing.md | The usage pipeline: JSONL transcript parsing, usage dedup, incremental byte-offset cursors, day bucketing, API-equivalent pricing, rolling quota windows, background QoS |
| verification-and-judgment.md | The verification stack: pairwise judging + order effects, staged verification, audience provenance, deterministic taste verdict, worktree isolation, FTS5, synthetic QA, spec-driven workflow |
Coverage map (subsystem → learning entry → canonical doc)
| Subsystem | Learning entry | Canonical doc |
|---|---|---|
| Tauri shell + IPC | new-things: Tauri 2, Tauri IPC | docs/architecture/overview.md |
| Review engine (CLI agents) | new-things: CLI-agent subprocess | docs/architecture/overview.md |
| Evidence patterns | new-things: ast-grep | — |
| Session indexing + usage telemetry | telemetry-and-indexing (all) | docs/development/performance.md |
| Provider quota cards | telemetry-and-indexing: rolling windows | — |
| Synthetic user QA | verification-and-judgment: synthetic QA | docs/product/synthetic-user-qa.md |
| Audience validation / taste | verification-and-judgment: pairwise judging, provenance, taste verdict | openspec/specs/ (audience-validation, taste-verdict) |
| Fix attempts / sandbox | verification-and-judgment: worktree isolation | — |
| Repo Unpacked | new-things: outcome calibration | docs/architecture/repo-unpacked.md |
| Benchmarking claims | — (read the doc directly) | docs/development/benchmark.md |
| Testing stacks | verification-and-judgment: synthetic QA | docs/development/testing.md |
| Release pipeline | new-things: auto-updater, Actions chaining | .github/workflows/*.yml comments |
| T-Rex PR watchers | new-things: PR polling watchers | — |
| Work (conversation over local PTY execution) | new-things: PTY-backed execution | openspec/specs/agent-panel/ |
| Landing / discoverability | new-things: GEO | LANDING_STANDARD.md (fleet root) |
Update rule: when a change introduces a concept with a non-obvious gotcha, add one entry to the matching page (or a new page past ~300 lines) and a row here. Don’t re-explain what a linked source already explains.