> ## Documentation Index
> Fetch the complete documentation index at: https://docs.autosearch.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Delivery Status

> Module-by-module status of the v2 rewrite

# v2 Delivery Status

Snapshot of which modules from `plan v2.3 § 13.5` have landed. All shipped modules have unit or integration tests and a 1:1 source comment on each file.

## Pipeline (M0–M8)

| Module                                             | Status                                                | Source                                                                                |
| -------------------------------------------------- | ----------------------------------------------------- | ------------------------------------------------------------------------------------- |
| M0 Knowledge Recall                                | ✅ shipped                                             | Self-written (node-deepresearch semantics)                                            |
| M1 Clarify                                         | ✅ shipped                                             | open\_deep\_research/src/open\_deep\_research/prompts.py:L3-L41                       |
| M2 Search Strategy                                 | ✅ shipped                                             | gpt-researcher/gpt\_researcher/prompts.py:L213-L255                                   |
| M3 Iteration Controller                            | ✅ shipped                                             | open\_deep\_research/src/legacy/graph.py:L235-L354                                    |
| M4 Trafilatura Cleaner                             | ✅ shipped                                             | storm/knowledge\_storm/utils.py:L685-L711                                             |
| M4 per-source cleaner framework                    | 🟡 Protocol only; real cleaners pending channel layer | Self-written                                                                          |
| M5 Evidence Processor (URL dedup + BM25 + SimHash) | ✅ shipped                                             | deer-flow infoquest\_client.py:L183-L230 + rank-bm25 + simhash                        |
| M7 Report Synthesizer                              | ✅ shipped                                             | storm outline\_generation / article\_generation + open\_deep\_research citation rules |
| M8 Quality Gate                                    | ✅ shipped                                             | open\_deep\_research/src/legacy/prompts.py:L168-L198                                  |

`Pipeline` orchestrator composes the above with a clarification early-exit and one quality-gate retry. Phase events are emitted through a user-supplied callback.

## Presentation

| Surface                                             | Status    | Notes                                                                 |
| --------------------------------------------------- | --------- | --------------------------------------------------------------------- |
| CLI `autosearch query`                              | ✅ shipped | `--mode fast\|deep`, `--top-k`, `--stream`, `--json`                  |
| CLI `autosearch mcp`                                | ✅ shipped | Alias for `autosearch-mcp` stdio server                               |
| CLI `autosearch serve`                              | ✅ shipped | `--host`, `--port` for FastAPI SSE                                    |
| HTTP `/health`                                      | ✅ shipped | JSON liveness                                                         |
| HTTP `/search` SSE                                  | ✅ shipped | Streams `phase` / `iteration` / `gap` / `quality` / `finished` events |
| MCP server (FastMCP stdio)                          | ✅ shipped | Tools: `research(query, mode)`, `health()`                            |
| Claude Code `/autosearch` slash command             | ✅ shipped | `commands/autosearch.md`                                              |
| Citation rendering                                  | ✅ shipped | Inline `[n]` + `## References` + `## Sources` breakdown               |
| Progress streaming                                  | ✅ shipped | Callback → stderr NDJSON (CLI) + SSE (HTTP)                           |
| OpenAI-compat `/v1/chat/completions` + `/v1/models` | ✅ shipped | node-deepresearch port; SSE chunked stream (role + content + DONE)    |

## Observability & Persistence

| Module                                | Status                                            | Notes                                                     |
| ------------------------------------- | ------------------------------------------------- | --------------------------------------------------------- |
| Cost Tracker                          | ✅ shipped + wired into `LLMClient` and `Pipeline` | gpt-researcher utils/costs.py                             |
| Session SQLite store (3-table schema) | ✅ shipped + wired into `Pipeline`                 | crawl4ai async\_database.py pattern + self-written schema |

## Channels

Deferred — real adapters will land after the channel layer unpauses. `DemoChannel` ships today as a placeholder implementing the `Channel` Protocol so the pipeline runs end-to-end.

Planned roadmap (plan § 5):

* Overseas P0 (12–15): arxiv / DDGS / GitHub repos+issues / Reddit / HackerNews / StackOverflow / YouTube / ProductHunt / Semantic Scholar / HuggingFace / npm / PyPI
* Chinese P0 (6): 小红书 / B 站 / 微信公众号 / 知乎 / 抖音 / 微博
* Chinese P1 (4–6): CSDN / 掘金 / 36kr / 雪球 / 小宇宙 / 快手

## Known Deferred Items

* `autosearch init` dependency orchestrator (plan § 4.5) — channel-dep-heavy, deferred until channel layer unpauses
* Per-source HTML cleaners for specific Chinese sites — deferred (channel layer owns the fetch path)
* Plan v2.3 itself — spike 2 findings suggested the plan initially tested the wrong fetch paths (HTML direct scrape for zhihu/csdn/juejin where the plan always intended API + cookie per § 5 中文 P0). No structural plan revision needed; the spike retrospectively validated plan § 5's path choices.

## Test Coverage

* 131 tests across 4 tiers: unit + integration (default CI), smoke (push-to-main), real\_llm (nightly, needs secrets), perf (on-demand)
* 0 ruff issues, consistent ruff format
* Pre-commit hook enforces author identity + PII/codename scan
* Pre-push rebases on main and runs the default selector (`not real_llm and not perf and not slow and not network`) in \~2 s
* See [`docs/testing/TEST_PLAN.md`](testing/TEST_PLAN.md) for the full pyramid

## Related Docs

* Plan: `~/.claude/plans/autosearch-v2-dev.md` (v2.3)
* Research mapping: `~/.claude/plans/research/module-map-*.md`
* Spike results: `docs/spikes/`