docs: triage screenshot dump into 5 actionable issues

Ingested and analyzed 5 screenshots from issue #1275:
- IMG_6187: AirLLM Apple Silicon requirements → #1284
- IMG_6125: vLLM backend for agentic throughput → #1281
- IMG_6124: DeerFlow research orchestration → #1283
- IMG_6123: Dev discipline / quality gates → #1285
- IMG_6410: SearXNG + Crawl4AI self-hosted search → #1282

Refs #1275

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

SOVEREIGNTY.md

@@ -0,0 +1,122 @@
+# SOVEREIGNTY.md — Research Sovereignty Manifest
+
+> "If this spec is implemented correctly, it is the last research document
+> Alexander should need to request from a corporate AI."
+> — Issue #972, March 22 2026
+
+---
+
+## What This Is
+
+A machine-readable declaration of Timmy's research independence:
+where we are, where we're going, and how to measure progress.
+
+---
+
+## The Problem We're Solving
+
+On March 22, 2026, a single Claude session produced six deep research reports.
+It consumed ~3 hours of human time and substantial corporate AI inference.
+Every report was valuable — but the workflow was **linear**.
+It would cost exactly the same to reproduce tomorrow.
+
+This file tracks the pipeline that crystallizes that workflow into something
+Timmy can run autonomously.
+
+---
+
+## The Six-Step Pipeline
+
+| Step | What Happens | Status |
+|------|-------------|--------|
+| 1. Scope | Human describes knowledge gap → Gitea issue with template | ✅ Done (`skills/research/`) |
+| 2. Query | LLM slot-fills template → 5–15 targeted queries | ✅ Done (`research.py`) |
+| 3. Search | Execute queries → top result URLs | ✅ Done (`research_tools.py`) |
+| 4. Fetch | Download + extract full pages (trafilatura) | ✅ Done (`tools/system_tools.py`) |
+| 5. Synthesize | Compress findings → structured report | ✅ Done (`research.py` cascade) |
+| 6. Deliver | Store to semantic memory + optional disk persist | ✅ Done (`research.py`) |
+
+---
+
+## Cascade Tiers (Synthesis Quality vs. Cost)
+
+| Tier | Model | Cost | Quality | Status |
+|------|-------|------|---------|--------|
+| **4** | SQLite semantic cache | $0.00 / instant | reuses prior | ✅ Active |
+| **3** | Ollama `qwen3:14b` | $0.00 / local | ★★★ | ✅ Active |
+| **2** | Claude API (haiku) | ~$0.01/report | ★★★★ | ✅ Active (opt-in) |
+| **1** | Groq `llama-3.3-70b` | $0.00 / rate-limited | ★★★★ | 🔲 Planned (#980) |
+
+Set `ANTHROPIC_API_KEY` to enable Tier 2 fallback.
+
+---
+
+## Research Templates
+
+Six prompt templates live in `skills/research/`:
+
+| Template | Use Case |
+|----------|----------|
+| `tool_evaluation.md` | Find all shipping tools for `{domain}` |
+| `architecture_spike.md` | How to connect `{system_a}` to `{system_b}` |
+| `game_analysis.md` | Evaluate `{game}` for AI agent play |
+| `integration_guide.md` | Wire `{tool}` into `{stack}` with code |
+| `state_of_art.md` | What exists in `{field}` as of `{date}` |
+| `competitive_scan.md` | How does `{project}` compare to `{alternatives}` |
+
+---
+
+## Sovereignty Metrics
+
+| Metric | Target (Week 1) | Target (Month 1) | Target (Month 3) | Graduation |
+|--------|-----------------|------------------|------------------|------------|
+| Queries answered locally | 10% | 40% | 80% | >90% |
+| API cost per report | <$1.50 | <$0.50 | <$0.10 | <$0.01 |
+| Time from question to report | <3 hours | <30 min | <5 min | <1 min |
+| Human involvement | 100% (review) | Review only | Approve only | None |
+
+---
+
+## How to Use the Pipeline
+
+```python
+from timmy.research import run_research
+
+# Quick research (no template)
+result = await run_research("best local embedding models for 36GB RAM")
+
+# With a template and slot values
+result = await run_research(
+    topic="PDF text extraction libraries for Python",
+    template="tool_evaluation",
+    slots={"domain": "PDF parsing", "use_case": "RAG pipeline", "focus_criteria": "accuracy"},
+    save_to_disk=True,
+)
+
+print(result.report)
+print(f"Backend: {result.synthesis_backend}, Cached: {result.cached}")
+```
+
+---
+
+## Implementation Status
+
+| Component | Issue | Status |
+|-----------|-------|--------|
+| `web_fetch` tool (trafilatura) | #973 | ✅ Done |
+| Research template library (6 templates) | #974 | ✅ Done |
+| `ResearchOrchestrator` (`research.py`) | #975 | ✅ Done |
+| Semantic index for outputs | #976 | 🔲 Planned |
+| Auto-create Gitea issues from findings | #977 | 🔲 Planned |
+| Paperclip task runner integration | #978 | 🔲 Planned |
+| Kimi delegation via labels | #979 | 🔲 Planned |
+| Groq free-tier cascade tier | #980 | 🔲 Planned |
+| Sovereignty metrics dashboard | #981 | 🔲 Planned |
+
+---
+
+## Governing Spec
+
+See [issue #972](http://143.198.27.163:3000/Rockachopa/Timmy-time-dashboard/issues/972) for the full spec and rationale.
+
+Research artifacts committed to `docs/research/`.

docs/SCREENSHOT_TRIAGE_2026-03-24.md

@@ -0,0 +1,89 @@
+# Screenshot Dump Triage — Visual Inspiration & Research Leads
+
+**Date:** March 24, 2026
+**Source:** Issue #1275 — "Screenshot dump for triage #1"
+**Analyst:** Claude (Sonnet 4.6)
+
+---
+
+## Screenshots Ingested
+
+| File | Subject | Action |
+|------|---------|--------|
+| IMG_6187.jpeg | AirLLM / Apple Silicon local LLM requirements | → Issue #1284 |
+| IMG_6125.jpeg | vLLM backend for agentic workloads | → Issue #1281 |
+| IMG_6124.jpeg | DeerFlow autonomous research pipeline | → Issue #1283 |
+| IMG_6123.jpeg | "Vibe Coder vs Normal Developer" meme | → Issue #1285 |
+| IMG_6410.jpeg | SearXNG + Crawl4AI self-hosted search MCP | → Issue #1282 |
+
+---
+
+## Tickets Created
+
+### #1281 — feat: add vLLM as alternative inference backend
+**Source:** IMG_6125 (vLLM for agentic workloads)
+
+vLLM's continuous batching makes it 3–10x more throughput-efficient than Ollama for multi-agent
+request patterns. Implement `VllmBackend` in `infrastructure/llm_router/` as a selectable
+backend (`TIMMY_LLM_BACKEND=vllm`) with graceful fallback to Ollama.
+
+**Priority:** Medium — impactful for research pipeline performance once #972 is in use
+
+---
+
+### #1282 — feat: integrate SearXNG + Crawl4AI as self-hosted search backend
+**Source:** IMG_6410 (luxiaolei/searxng-crawl4ai-mcp)
+
+Self-hosted search via SearXNG + Crawl4AI removes the hard dependency on paid search APIs
+(Brave, Tavily). Add both as Docker Compose services, implement `web_search()` and
+`scrape_url()` tools in `timmy/tools/`, and register them with the research agent.
+
+**Priority:** High — unblocks fully local/private operation of research agents
+
+---
+
+### #1283 — research: evaluate DeerFlow as autonomous research orchestration layer
+**Source:** IMG_6124 (deer-flow Docker setup)
+
+DeerFlow is ByteDance's open-source autonomous research pipeline framework. Before investing
+further in Timmy's custom orchestrator (#972), evaluate whether DeerFlow's architecture offers
+integration value or design patterns worth borrowing.
+
+**Priority:** Medium — research first, implementation follows if go/no-go is positive
+
+---
+
+### #1284 — chore: document and validate AirLLM Apple Silicon requirements
+**Source:** IMG_6187 (Mac-compatible LLM setup)
+
+AirLLM graceful degradation is already implemented but undocumented. Add System Requirements
+to README (M1/M2/M3/M4, 16 GB RAM min, 15 GB disk) and document `TIMMY_LLM_BACKEND` in
+`.env.example`.
+
+**Priority:** Low — documentation only, no code risk
+
+---
+
+### #1285 — chore: enforce "Normal Developer" discipline — tighten quality gates
+**Source:** IMG_6123 (Vibe Coder vs Normal Developer meme)
+
+Tighten the existing mypy/bandit/coverage gates: fix all mypy errors, raise coverage from 73%
+to 80%, add a documented pre-push hook, and run `vulture` for dead code. The infrastructure
+exists — it just needs enforcing.
+
+**Priority:** Medium — technical debt prevention, pairs well with any green-field feature work
+
+---
+
+## Patterns Observed Across Screenshots
+
+1. **Local-first is the north star.** All five images reinforce the same theme: private,
+   self-hosted, runs on your hardware. vLLM, SearXNG, AirLLM, DeerFlow — none require cloud.
+   Timmy is already aligned with this direction; these are tactical additions.
+
+2. **Agentic performance bottlenecks are real.** Two of five images (vLLM, DeerFlow) focus
+   specifically on throughput and reliability for multi-agent loops. As the research pipeline
+   matures, inference speed and search reliability will become the main constraints.
+
+3. **Discipline compounds.** The meme is a reminder that the quality gates we have (tox,
+   mypy, bandit, coverage) only pay off if they are enforced without exceptions.

src/timmy/research.py

@@ -0,0 +1,528 @@
+"""Research Orchestrator — autonomous, sovereign research pipeline.
+
+Chains all six steps of the research workflow with local-first execution:
+
+    Step 0  Cache   — check semantic memory (SQLite, instant, zero API cost)
+    Step 1  Scope   — load a research template from skills/research/
+    Step 2  Query   — slot-fill template + formulate 5-15 search queries via Ollama
+    Step 3  Search  — execute queries via web_search (SerpAPI or fallback)
+    Step 4  Fetch   — download + extract full pages via web_fetch (trafilatura)
+    Step 5  Synth   — compress findings into a structured report via cascade
+    Step 6  Deliver — store to semantic memory; optionally save to docs/research/
+
+Cascade tiers for synthesis (spec §4):
+    Tier 4  SQLite semantic cache  — instant, free, covers ~80% after warm-up
+    Tier 3  Ollama (qwen3:14b)     — local, free, good quality
+    Tier 2  Claude API (haiku)     — cloud fallback, cheap, set ANTHROPIC_API_KEY
+    Tier 1  (future) Groq          — free-tier rate-limited, tracked in #980
+
+All optional services degrade gracefully per project conventions.
+
+Refs #972 (governing spec), #975 (ResearchOrchestrator sub-issue).
+"""
+
+from __future__ import annotations
+
+import asyncio
+import logging
+import re
+import textwrap
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+# Optional memory imports — available at module level so tests can patch them.
+try:
+    from timmy.memory_system import SemanticMemory, store_memory
+except Exception:  # pragma: no cover
+    SemanticMemory = None  # type: ignore[assignment,misc]
+    store_memory = None  # type: ignore[assignment]
+
+# Root of the project — two levels up from src/timmy/
+_PROJECT_ROOT = Path(__file__).parent.parent.parent
+_SKILLS_ROOT = _PROJECT_ROOT / "skills" / "research"
+_DOCS_ROOT = _PROJECT_ROOT / "docs" / "research"
+
+# Similarity threshold for cache hit (0–1 cosine similarity)
+_CACHE_HIT_THRESHOLD = 0.82
+
+# How many search result URLs to fetch as full pages
+_FETCH_TOP_N = 5
+
+# Maximum tokens to request from the synthesis LLM
+_SYNTHESIS_MAX_TOKENS = 4096
+
+
+# ---------------------------------------------------------------------------
+# Data structures
+# ---------------------------------------------------------------------------
+
+
+@dataclass
+class ResearchResult:
+    """Full output of a research pipeline run."""
+
+    topic: str
+    query_count: int
+    sources_fetched: int
+    report: str
+    cached: bool = False
+    cache_similarity: float = 0.0
+    synthesis_backend: str = "unknown"
+    errors: list[str] = field(default_factory=list)
+
+    def is_empty(self) -> bool:
+        return not self.report.strip()
+
+
+# ---------------------------------------------------------------------------
+# Template loading
+# ---------------------------------------------------------------------------
+
+
+def list_templates() -> list[str]:
+    """Return names of available research templates (without .md extension)."""
+    if not _SKILLS_ROOT.exists():
+        return []
+    return [p.stem for p in sorted(_SKILLS_ROOT.glob("*.md"))]
+
+
+def load_template(template_name: str, slots: dict[str, str] | None = None) -> str:
+    """Load a research template and fill {slot} placeholders.
+
+    Args:
+        template_name: Stem of the .md file under skills/research/ (e.g. "tool_evaluation").
+        slots: Mapping of {placeholder} → replacement value.
+
+    Returns:
+        Template text with slots filled. Unfilled slots are left as-is.
+    """
+    path = _SKILLS_ROOT / f"{template_name}.md"
+    if not path.exists():
+        available = ", ".join(list_templates()) or "(none)"
+        raise FileNotFoundError(
+            f"Research template {template_name!r} not found. "
+            f"Available: {available}"
+        )
+
+    text = path.read_text(encoding="utf-8")
+
+    # Strip YAML frontmatter (--- ... ---), including empty frontmatter (--- \n---)
+    text = re.sub(r"^---\n.*?---\n", "", text, flags=re.DOTALL)
+
+    if slots:
+        for key, value in slots.items():
+            text = text.replace(f"{{{key}}}", value)
+
+    return text.strip()
+
+
+# ---------------------------------------------------------------------------
+# Query formulation (Step 2)
+# ---------------------------------------------------------------------------
+
+
+async def _formulate_queries(topic: str, template_context: str, n: int = 8) -> list[str]:
+    """Use the local LLM to generate targeted search queries for a topic.
+
+    Falls back to a simple heuristic if Ollama is unavailable.
+    """
+    prompt = textwrap.dedent(f"""\
+        You are a research assistant. Generate exactly {n} targeted, specific web search
+        queries to thoroughly research the following topic.
+
+        TOPIC: {topic}
+
+        RESEARCH CONTEXT:
+        {template_context[:1000]}
+
+        Rules:
+        - One query per line, no numbering, no bullet points.
+        - Vary the angle (definition, comparison, implementation, alternatives, pitfalls).
+        - Prefer exact technical terms, tool names, and version numbers where relevant.
+        - Output ONLY the queries, nothing else.
+    """)
+
+    queries = await _ollama_complete(prompt, max_tokens=512)
+
+    if not queries:
+        # Minimal fallback
+        return [
+            f"{topic} overview",
+            f"{topic} tutorial",
+            f"{topic} best practices",
+            f"{topic} alternatives",
+            f"{topic} 2025",
+        ]
+
+    lines = [ln.strip() for ln in queries.splitlines() if ln.strip()]
+    return lines[:n] if len(lines) >= n else lines
+
+
+# ---------------------------------------------------------------------------
+# Search (Step 3)
+# ---------------------------------------------------------------------------
+
+
+async def _execute_search(queries: list[str]) -> list[dict[str, str]]:
+    """Run each query through the available web search backend.
+
+    Returns a flat list of {title, url, snippet} dicts.
+    Degrades gracefully if SerpAPI key is absent.
+    """
+    results: list[dict[str, str]] = []
+    seen_urls: set[str] = set()
+
+    for query in queries:
+        try:
+            raw = await asyncio.to_thread(_run_search_sync, query)
+            for item in raw:
+                url = item.get("url", "")
+                if url and url not in seen_urls:
+                    seen_urls.add(url)
+                    results.append(item)
+        except Exception as exc:
+            logger.warning("Search failed for query %r: %s", query, exc)
+
+    return results
+
+
+def _run_search_sync(query: str) -> list[dict[str, str]]:
+    """Synchronous search — wraps SerpAPI or returns empty on missing key."""
+    import os
+
+    if not os.environ.get("SERPAPI_API_KEY"):
+        logger.debug("SERPAPI_API_KEY not set — skipping web search for %r", query)
+        return []
+
+    try:
+        from serpapi import GoogleSearch
+
+        params = {"q": query, "api_key": os.environ["SERPAPI_API_KEY"], "num": 5}
+        search = GoogleSearch(params)
+        data = search.get_dict()
+        items = []
+        for r in data.get("organic_results", []):
+            items.append(
+                {
+                    "title": r.get("title", ""),
+                    "url": r.get("link", ""),
+                    "snippet": r.get("snippet", ""),
+                }
+            )
+        return items
+    except Exception as exc:
+        logger.warning("SerpAPI search error: %s", exc)
+        return []
+
+
+# ---------------------------------------------------------------------------
+# Fetch (Step 4)
+# ---------------------------------------------------------------------------
+
+
+async def _fetch_pages(results: list[dict[str, str]], top_n: int = _FETCH_TOP_N) -> list[str]:
+    """Download and extract full text for the top search results.
+
+    Uses web_fetch (trafilatura) from timmy.tools.system_tools.
+    """
+    try:
+        from timmy.tools.system_tools import web_fetch
+    except ImportError:
+        logger.warning("web_fetch not available — skipping page fetch")
+        return []
+
+    pages: list[str] = []
+    for item in results[:top_n]:
+        url = item.get("url", "")
+        if not url:
+            continue
+        try:
+            text = await asyncio.to_thread(web_fetch, url, 6000)
+            if text and not text.startswith("Error:"):
+                pages.append(f"## {item.get('title', url)}\nSource: {url}\n\n{text}")
+        except Exception as exc:
+            logger.warning("Failed to fetch %s: %s", url, exc)
+
+    return pages
+
+
+# ---------------------------------------------------------------------------
+# Synthesis (Step 5) — cascade: Ollama → Claude fallback
+# ---------------------------------------------------------------------------
+
+
+async def _synthesize(topic: str, pages: list[str], snippets: list[str]) -> tuple[str, str]:
+    """Compress fetched pages + snippets into a structured research report.
+
+    Returns (report_markdown, backend_used).
+    """
+    # Build synthesis prompt
+    source_content = "\n\n---\n\n".join(pages[:5])
+    if not source_content and snippets:
+        source_content = "\n".join(f"- {s}" for s in snippets[:20])
+
+    if not source_content:
+        return (
+            f"# Research: {topic}\n\n*No source material was retrieved. "
+            "Check SERPAPI_API_KEY and network connectivity.*",
+            "none",
+        )
+
+    prompt = textwrap.dedent(f"""\
+        You are a senior technical researcher. Synthesize the source material below
+        into a structured research report on the topic: **{topic}**
+
+        FORMAT YOUR REPORT AS:
+        # {topic}
+
+        ## Executive Summary
+        (2-3 sentences: what you found, top recommendation)
+
+        ## Key Findings
+        (Bullet list of the most important facts, tools, or patterns)
+
+        ## Comparison / Options
+        (Table or list comparing alternatives where applicable)
+
+        ## Recommended Approach
+        (Concrete recommendation with rationale)
+
+        ## Gaps & Next Steps
+        (What wasn't answered, what to investigate next)
+
+        ---
+        SOURCE MATERIAL:
+        {source_content[:12000]}
+    """)
+
+    # Tier 3 — try Ollama first
+    report = await _ollama_complete(prompt, max_tokens=_SYNTHESIS_MAX_TOKENS)
+    if report:
+        return report, "ollama"
+
+    # Tier 2 — Claude fallback
+    report = await _claude_complete(prompt, max_tokens=_SYNTHESIS_MAX_TOKENS)
+    if report:
+        return report, "claude"
+
+    # Last resort — structured snippet summary
+    summary = f"# {topic}\n\n## Snippets\n\n" + "\n\n".join(
+        f"- {s}" for s in snippets[:15]
+    )
+    return summary, "fallback"
+
+
+# ---------------------------------------------------------------------------
+# LLM helpers
+# ---------------------------------------------------------------------------
+
+
+async def _ollama_complete(prompt: str, max_tokens: int = 1024) -> str:
+    """Send a prompt to Ollama and return the response text.
+
+    Returns empty string on failure (graceful degradation).
+    """
+    try:
+        import httpx
+
+        from config import settings
+
+        url = f"{settings.normalized_ollama_url}/api/generate"
+        payload: dict[str, Any] = {
+            "model": settings.ollama_model,
+            "prompt": prompt,
+            "stream": False,
+            "options": {
+                "num_predict": max_tokens,
+                "temperature": 0.3,
+            },
+        }
+
+        async with httpx.AsyncClient(timeout=120.0) as client:
+            resp = await client.post(url, json=payload)
+            resp.raise_for_status()
+            data = resp.json()
+            return data.get("response", "").strip()
+    except Exception as exc:
+        logger.warning("Ollama completion failed: %s", exc)
+        return ""
+
+
+async def _claude_complete(prompt: str, max_tokens: int = 1024) -> str:
+    """Send a prompt to Claude API as a last-resort fallback.
+
+    Only active when ANTHROPIC_API_KEY is configured.
+    Returns empty string on failure or missing key.
+    """
+    try:
+        from config import settings
+
+        if not settings.anthropic_api_key:
+            return ""
+
+        from timmy.backends import ClaudeBackend
+
+        backend = ClaudeBackend()
+        result = await asyncio.to_thread(backend.run, prompt)
+        return result.content.strip()
+    except Exception as exc:
+        logger.warning("Claude fallback failed: %s", exc)
+        return ""
+
+
+# ---------------------------------------------------------------------------
+# Memory cache (Step 0 + Step 6)
+# ---------------------------------------------------------------------------
+
+
+def _check_cache(topic: str) -> tuple[str | None, float]:
+    """Search semantic memory for a prior result on this topic.
+
+    Returns (cached_report, similarity) or (None, 0.0).
+    """
+    try:
+        if SemanticMemory is None:
+            return None, 0.0
+        mem = SemanticMemory()
+        hits = mem.search(topic, top_k=1)
+        if hits:
+            content, score = hits[0]
+            if score >= _CACHE_HIT_THRESHOLD:
+                return content, score
+    except Exception as exc:
+        logger.debug("Cache check failed: %s", exc)
+    return None, 0.0
+
+
+def _store_result(topic: str, report: str) -> None:
+    """Index the research report into semantic memory for future retrieval."""
+    try:
+        if store_memory is None:
+            logger.debug("store_memory not available — skipping memory index")
+            return
+        store_memory(
+            content=report,
+            source="research_pipeline",
+            context_type="research",
+            metadata={"topic": topic},
+        )
+        logger.info("Research result indexed for topic: %r", topic)
+    except Exception as exc:
+        logger.warning("Failed to store research result: %s", exc)
+
+
+def _save_to_disk(topic: str, report: str) -> Path | None:
+    """Persist the report as a markdown file under docs/research/.
+
+    Filename is derived from the topic (slugified). Returns the path or None.
+    """
+    try:
+        slug = re.sub(r"[^a-z0-9]+", "-", topic.lower()).strip("-")[:60]
+        _DOCS_ROOT.mkdir(parents=True, exist_ok=True)
+        path = _DOCS_ROOT / f"{slug}.md"
+        path.write_text(report, encoding="utf-8")
+        logger.info("Research report saved to %s", path)
+        return path
+    except Exception as exc:
+        logger.warning("Failed to save research report to disk: %s", exc)
+        return None
+
+
+# ---------------------------------------------------------------------------
+# Main orchestrator
+# ---------------------------------------------------------------------------
+
+
+async def run_research(
+    topic: str,
+    template: str | None = None,
+    slots: dict[str, str] | None = None,
+    save_to_disk: bool = False,
+    skip_cache: bool = False,
+) -> ResearchResult:
+    """Run the full 6-step autonomous research pipeline.
+
+    Args:
+        topic:        The research question or subject.
+        template:     Name of a template from skills/research/ (e.g. "tool_evaluation").
+                      If None, runs without a template scaffold.
+        slots:        Placeholder values for the template (e.g. {"domain": "PDF parsing"}).
+        save_to_disk: If True, write the report to docs/research/<slug>.md.
+        skip_cache:   If True, bypass the semantic memory cache.
+
+    Returns:
+        ResearchResult with report and metadata.
+    """
+    errors: list[str] = []
+
+    # ------------------------------------------------------------------
+    # Step 0 — check cache
+    # ------------------------------------------------------------------
+    if not skip_cache:
+        cached, score = _check_cache(topic)
+        if cached:
+            logger.info("Cache hit (%.2f) for topic: %r", score, topic)
+            return ResearchResult(
+                topic=topic,
+                query_count=0,
+                sources_fetched=0,
+                report=cached,
+                cached=True,
+                cache_similarity=score,
+                synthesis_backend="cache",
+            )
+
+    # ------------------------------------------------------------------
+    # Step 1 — load template (optional)
+    # ------------------------------------------------------------------
+    template_context = ""
+    if template:
+        try:
+            template_context = load_template(template, slots)
+        except FileNotFoundError as exc:
+            errors.append(str(exc))
+            logger.warning("Template load failed: %s", exc)
+
+    # ------------------------------------------------------------------
+    # Step 2 — formulate queries
+    # ------------------------------------------------------------------
+    queries = await _formulate_queries(topic, template_context)
+    logger.info("Formulated %d queries for topic: %r", len(queries), topic)
+
+    # ------------------------------------------------------------------
+    # Step 3 — execute search
+    # ------------------------------------------------------------------
+    search_results = await _execute_search(queries)
+    logger.info("Search returned %d results", len(search_results))
+    snippets = [r.get("snippet", "") for r in search_results if r.get("snippet")]
+
+    # ------------------------------------------------------------------
+    # Step 4 — fetch full pages
+    # ------------------------------------------------------------------
+    pages = await _fetch_pages(search_results)
+    logger.info("Fetched %d pages", len(pages))
+
+    # ------------------------------------------------------------------
+    # Step 5 — synthesize
+    # ------------------------------------------------------------------
+    report, backend = await _synthesize(topic, pages, snippets)
+
+    # ------------------------------------------------------------------
+    # Step 6 — deliver
+    # ------------------------------------------------------------------
+    _store_result(topic, report)
+    if save_to_disk:
+        _save_to_disk(topic, report)
+
+    return ResearchResult(
+        topic=topic,
+        query_count=len(queries),
+        sources_fetched=len(pages),
+        report=report,
+        cached=False,
+        synthesis_backend=backend,
+        errors=errors,
+    )

tests/infrastructure/test_event_bus.py

@@ -7,8 +7,6 @@ from unittest.mock import patch
 import pytest
 
 import infrastructure.events.bus as bus_module
-
-pytestmark = pytest.mark.unit
 from infrastructure.events.bus import (
     Event,
     EventBus,
@@ -354,14 +352,6 @@ class TestEventBusPersistence:
         events = bus.replay()
         assert events == []
 
-    def test_init_persistence_db_noop_when_path_is_none(self):
-        """_init_persistence_db() is a no-op when _persistence_db_path is None."""
-        bus = EventBus()
-        # _persistence_db_path is None by default; calling _init_persistence_db
-        # should silently return without touching the filesystem.
-        bus._init_persistence_db()  # must not raise
-        assert bus._persistence_db_path is None
-
     async def test_wal_mode_on_persistence_db(self, persistent_bus):
         """Persistence database should use WAL mode."""
         conn = sqlite3.connect(str(persistent_bus._persistence_db_path))

tests/timmy/test_research.py

@@ -0,0 +1,403 @@
+"""Unit tests for src/timmy/research.py — ResearchOrchestrator pipeline.
+
+Refs #972 (governing spec), #975 (ResearchOrchestrator).
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+pytestmark = pytest.mark.unit
+
+
+# ---------------------------------------------------------------------------
+# list_templates
+# ---------------------------------------------------------------------------
+
+
+class TestListTemplates:
+    def test_returns_list(self, tmp_path, monkeypatch):
+        (tmp_path / "tool_evaluation.md").write_text("---\n---\n# T")
+        (tmp_path / "game_analysis.md").write_text("---\n---\n# G")
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        from timmy.research import list_templates
+
+        result = list_templates()
+        assert isinstance(result, list)
+        assert "tool_evaluation" in result
+        assert "game_analysis" in result
+
+    def test_returns_empty_when_dir_missing(self, tmp_path, monkeypatch):
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path / "nonexistent")
+
+        from timmy.research import list_templates
+
+        assert list_templates() == []
+
+
+# ---------------------------------------------------------------------------
+# load_template
+# ---------------------------------------------------------------------------
+
+
+class TestLoadTemplate:
+    def _write_template(self, path: Path, name: str, body: str) -> None:
+        (path / f"{name}.md").write_text(body, encoding="utf-8")
+
+    def test_loads_and_strips_frontmatter(self, tmp_path, monkeypatch):
+        self._write_template(
+            tmp_path,
+            "tool_evaluation",
+            "---\nname: Tool Evaluation\ntype: research\n---\n# Tool Eval: {domain}",
+        )
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        from timmy.research import load_template
+
+        result = load_template("tool_evaluation", {"domain": "PDF parsing"})
+        assert "# Tool Eval: PDF parsing" in result
+        assert "name: Tool Evaluation" not in result
+
+    def test_fills_slots(self, tmp_path, monkeypatch):
+        self._write_template(tmp_path, "arch", "Connect {system_a} to {system_b}")
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        from timmy.research import load_template
+
+        result = load_template("arch", {"system_a": "Kafka", "system_b": "Postgres"})
+        assert "Kafka" in result
+        assert "Postgres" in result
+
+    def test_unfilled_slots_preserved(self, tmp_path, monkeypatch):
+        self._write_template(tmp_path, "t", "Hello {name} and {other}")
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        from timmy.research import load_template
+
+        result = load_template("t", {"name": "World"})
+        assert "{other}" in result
+
+    def test_raises_file_not_found_for_missing_template(self, tmp_path, monkeypatch):
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        from timmy.research import load_template
+
+        with pytest.raises(FileNotFoundError, match="nonexistent"):
+            load_template("nonexistent")
+
+    def test_no_slots_returns_raw_body(self, tmp_path, monkeypatch):
+        self._write_template(tmp_path, "plain", "---\n---\nJust text here")
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        from timmy.research import load_template
+
+        result = load_template("plain")
+        assert result == "Just text here"
+
+
+# ---------------------------------------------------------------------------
+# _check_cache
+# ---------------------------------------------------------------------------
+
+
+class TestCheckCache:
+    def test_returns_none_when_no_hits(self):
+        mock_mem = MagicMock()
+        mock_mem.search.return_value = []
+
+        with patch("timmy.research.SemanticMemory", return_value=mock_mem):
+            from timmy.research import _check_cache
+
+            content, score = _check_cache("some topic")
+
+        assert content is None
+        assert score == 0.0
+
+    def test_returns_content_above_threshold(self):
+        mock_mem = MagicMock()
+        mock_mem.search.return_value = [("cached report text", 0.91)]
+
+        with patch("timmy.research.SemanticMemory", return_value=mock_mem):
+            from timmy.research import _check_cache
+
+            content, score = _check_cache("same topic")
+
+        assert content == "cached report text"
+        assert score == pytest.approx(0.91)
+
+    def test_returns_none_below_threshold(self):
+        mock_mem = MagicMock()
+        mock_mem.search.return_value = [("old report", 0.60)]
+
+        with patch("timmy.research.SemanticMemory", return_value=mock_mem):
+            from timmy.research import _check_cache
+
+            content, score = _check_cache("slightly different topic")
+
+        assert content is None
+        assert score == 0.0
+
+    def test_degrades_gracefully_on_import_error(self):
+        with patch("timmy.research.SemanticMemory", None):
+            from timmy.research import _check_cache
+
+            content, score = _check_cache("topic")
+
+        assert content is None
+        assert score == 0.0
+
+
+# ---------------------------------------------------------------------------
+# _store_result
+# ---------------------------------------------------------------------------
+
+
+class TestStoreResult:
+    def test_calls_store_memory(self):
+        mock_store = MagicMock()
+
+        with patch("timmy.research.store_memory", mock_store):
+            from timmy.research import _store_result
+
+            _store_result("test topic", "# Report\n\nContent here.")
+
+        mock_store.assert_called_once()
+        call_kwargs = mock_store.call_args
+        assert "test topic" in str(call_kwargs)
+
+    def test_degrades_gracefully_on_error(self):
+        mock_store = MagicMock(side_effect=RuntimeError("db error"))
+        with patch("timmy.research.store_memory", mock_store):
+            from timmy.research import _store_result
+
+            # Should not raise
+            _store_result("topic", "report")
+
+
+# ---------------------------------------------------------------------------
+# _save_to_disk
+# ---------------------------------------------------------------------------
+
+
+class TestSaveToDisk:
+    def test_writes_file(self, tmp_path, monkeypatch):
+        monkeypatch.setattr("timmy.research._DOCS_ROOT", tmp_path / "research")
+
+        from timmy.research import _save_to_disk
+
+        path = _save_to_disk("Test Topic: PDF Parsing", "# Test Report")
+        assert path is not None
+        assert path.exists()
+        assert path.read_text() == "# Test Report"
+
+    def test_slugifies_topic_name(self, tmp_path, monkeypatch):
+        monkeypatch.setattr("timmy.research._DOCS_ROOT", tmp_path / "research")
+
+        from timmy.research import _save_to_disk
+
+        path = _save_to_disk("My Complex Topic! v2.0", "content")
+        assert path is not None
+        # Should be slugified: no special chars
+        assert " " not in path.name
+        assert "!" not in path.name
+
+    def test_returns_none_on_error(self, monkeypatch):
+        monkeypatch.setattr(
+            "timmy.research._DOCS_ROOT",
+            Path("/nonexistent_root/deeply/nested"),
+        )
+
+        with patch("pathlib.Path.mkdir", side_effect=PermissionError("denied")):
+            from timmy.research import _save_to_disk
+
+            result = _save_to_disk("topic", "report")
+
+        assert result is None
+
+
+# ---------------------------------------------------------------------------
+# run_research — end-to-end with mocks
+# ---------------------------------------------------------------------------
+
+
+class TestRunResearch:
+    @pytest.mark.asyncio
+    async def test_returns_cached_result_when_cache_hit(self):
+        cached_report = "# Cached Report\n\nPreviously computed."
+        with (
+            patch("timmy.research._check_cache", return_value=(cached_report, 0.93)),
+        ):
+            from timmy.research import run_research
+
+            result = await run_research("some topic")
+
+        assert result.cached is True
+        assert result.cache_similarity == pytest.approx(0.93)
+        assert result.report == cached_report
+        assert result.synthesis_backend == "cache"
+
+    @pytest.mark.asyncio
+    async def test_skips_cache_when_requested(self, tmp_path, monkeypatch):
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        with (
+            patch("timmy.research._check_cache", return_value=("cached", 0.99)) as mock_cache,
+            patch(
+                "timmy.research._formulate_queries",
+                new=AsyncMock(return_value=["q1"]),
+            ),
+            patch("timmy.research._execute_search", new=AsyncMock(return_value=[])),
+            patch("timmy.research._fetch_pages", new=AsyncMock(return_value=[])),
+            patch(
+                "timmy.research._synthesize",
+                new=AsyncMock(return_value=("# Fresh report", "ollama")),
+            ),
+            patch("timmy.research._store_result"),
+        ):
+            from timmy.research import run_research
+
+            result = await run_research("topic", skip_cache=True)
+
+        mock_cache.assert_not_called()
+        assert result.cached is False
+        assert result.report == "# Fresh report"
+
+    @pytest.mark.asyncio
+    async def test_full_pipeline_no_search_results(self, tmp_path, monkeypatch):
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        with (
+            patch("timmy.research._check_cache", return_value=(None, 0.0)),
+            patch(
+                "timmy.research._formulate_queries",
+                new=AsyncMock(return_value=["query 1", "query 2"]),
+            ),
+            patch("timmy.research._execute_search", new=AsyncMock(return_value=[])),
+            patch("timmy.research._fetch_pages", new=AsyncMock(return_value=[])),
+            patch(
+                "timmy.research._synthesize",
+                new=AsyncMock(return_value=("# Report", "ollama")),
+            ),
+            patch("timmy.research._store_result"),
+        ):
+            from timmy.research import run_research
+
+            result = await run_research("a new topic")
+
+        assert not result.cached
+        assert result.query_count == 2
+        assert result.sources_fetched == 0
+        assert result.report == "# Report"
+        assert result.synthesis_backend == "ollama"
+
+    @pytest.mark.asyncio
+    async def test_returns_result_with_error_on_bad_template(self, tmp_path, monkeypatch):
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        with (
+            patch("timmy.research._check_cache", return_value=(None, 0.0)),
+            patch(
+                "timmy.research._formulate_queries",
+                new=AsyncMock(return_value=["q1"]),
+            ),
+            patch("timmy.research._execute_search", new=AsyncMock(return_value=[])),
+            patch("timmy.research._fetch_pages", new=AsyncMock(return_value=[])),
+            patch(
+                "timmy.research._synthesize",
+                new=AsyncMock(return_value=("# Report", "ollama")),
+            ),
+            patch("timmy.research._store_result"),
+        ):
+            from timmy.research import run_research
+
+            result = await run_research("topic", template="nonexistent_template")
+
+        assert len(result.errors) == 1
+        assert "nonexistent_template" in result.errors[0]
+
+    @pytest.mark.asyncio
+    async def test_saves_to_disk_when_requested(self, tmp_path, monkeypatch):
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+        monkeypatch.setattr("timmy.research._DOCS_ROOT", tmp_path / "research")
+
+        with (
+            patch("timmy.research._check_cache", return_value=(None, 0.0)),
+            patch(
+                "timmy.research._formulate_queries",
+                new=AsyncMock(return_value=["q1"]),
+            ),
+            patch("timmy.research._execute_search", new=AsyncMock(return_value=[])),
+            patch("timmy.research._fetch_pages", new=AsyncMock(return_value=[])),
+            patch(
+                "timmy.research._synthesize",
+                new=AsyncMock(return_value=("# Saved Report", "ollama")),
+            ),
+            patch("timmy.research._store_result"),
+        ):
+            from timmy.research import run_research
+
+            result = await run_research("disk topic", save_to_disk=True)
+
+        assert result.report == "# Saved Report"
+        saved_files = list((tmp_path / "research").glob("*.md"))
+        assert len(saved_files) == 1
+        assert saved_files[0].read_text() == "# Saved Report"
+
+    @pytest.mark.asyncio
+    async def test_result_is_not_empty_after_synthesis(self, tmp_path, monkeypatch):
+        monkeypatch.setattr("timmy.research._SKILLS_ROOT", tmp_path)
+
+        with (
+            patch("timmy.research._check_cache", return_value=(None, 0.0)),
+            patch(
+                "timmy.research._formulate_queries",
+                new=AsyncMock(return_value=["q"]),
+            ),
+            patch("timmy.research._execute_search", new=AsyncMock(return_value=[])),
+            patch("timmy.research._fetch_pages", new=AsyncMock(return_value=[])),
+            patch(
+                "timmy.research._synthesize",
+                new=AsyncMock(return_value=("# Non-empty", "ollama")),
+            ),
+            patch("timmy.research._store_result"),
+        ):
+            from timmy.research import run_research
+
+            result = await run_research("topic")
+
+        assert not result.is_empty()
+
+
+# ---------------------------------------------------------------------------
+# ResearchResult
+# ---------------------------------------------------------------------------
+
+
+class TestResearchResult:
+    def test_is_empty_when_no_report(self):
+        from timmy.research import ResearchResult
+
+        r = ResearchResult(topic="t", query_count=0, sources_fetched=0, report="")
+        assert r.is_empty()
+
+    def test_is_not_empty_with_content(self):
+        from timmy.research import ResearchResult
+
+        r = ResearchResult(topic="t", query_count=1, sources_fetched=1, report="# Report")
+        assert not r.is_empty()
+
+    def test_default_cached_false(self):
+        from timmy.research import ResearchResult
+
+        r = ResearchResult(topic="t", query_count=0, sources_fetched=0, report="x")
+        assert r.cached is False
+
+    def test_errors_defaults_to_empty_list(self):
+        from timmy.research import ResearchResult
+
+        r = ResearchResult(topic="t", query_count=0, sources_fetched=0, report="x")
+        assert r.errors == []