refactor(orchestrator): extract _base_run_state (shared book/article run-state) [Tier 1]

start_book/start_article now spread a shared _base_run_state (~18 identical keys); mode-specific keys (id, phase, unit counter, per-mode flags) added by each caller. Resulting state dict is identical to before (both full-run tests green). Deliberately did NOT merge _commit/_commit_section - their canon-vs-citation handling makes a shared helper read worse than the small dup. plan.md §20 updated.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: vikast908

plan.md

@@ -1016,15 +1016,15 @@ The book (chapter) and article (section) pipelines run near-parallel code in `or
 down **incrementally and test-gated**: these paths have a history of *silent drift* (the revise-parity
 bug), so behavior-preserving extraction + the full suite (and ideally a live run) between steps is
 mandatory. Already shared (do not re-extract): `_pick_variant`, `_save_version`, `_record_preference`,
-`_length_note`, `_merge_fix_notes`, `_escalate`. **Done:** `_run_learner` (shared learner tail).
+`_length_note`, `_merge_fix_notes`, `_escalate`. **Done:** `_run_learner` (shared learner tail);
+`_base_run_state` (shared run-state keys for `start_book`/`start_article`).
 
 Prioritized, by risk:
 
-- **Tier 1 (LOW - do next):** `_init_project_state` - a shared run-state dict builder for
-  `start_book`/`start_article` (mode-specific keys as kwargs; mind the `current_chapter`/`current_section`
-  + `num_*` name differences). `_commit`/`_commit_section` - extract the shared humanize+summarize+
-  version+skill-record scaffold; keep canon-extraction (book) vs citation-renumbering (article) as a
-  `finalize` callback.
+- **Tier 1:** ✅ `_base_run_state` (the shared run-state dict; mode-specific keys spread in by each
+  caller) - **done**. ❌ `_commit`/`_commit_section` - **evaluated and deliberately NOT merged**: the
+  paths differ structurally (canon-extraction + Store updates vs citation-renumber-before-gather), so a
+  shared helper would be callback-soup that reads worse than the ~8 duplicated lines. Leave separate.
 - **Tier 2 (MEDIUM - schema/strategy callbacks):** `_chapter_fetch`/`_section_fetch` - shared
   `concurrency.gather` shell, research/images/skills passed as strategy fns (article also returns
   `source_text` for claim-verify). The divergent-draft + revision attempt loop inside

resume.md

@@ -284,11 +284,12 @@ redundancy items are large refactors, left as recommendations.
   source docstrings (`cli`, `orchestrator`, `brain`, `prompts`, …) + the CLAUDE workflow - moving them
   would be high-churn for no gain. `requirements.txt` stays (a documented convenience mirror; pyproject
   is canonical).
-- **Book↔article de-dup STARTED (safe, test-gated):** mapped the duplication (Explore agent), then
-  extracted the leaf-level **`_run_learner`** shared by `_learn`/`_learn_article` (suite green). Committed
-  a prioritized, risk-rated **refactor plan in plan.md §20** for the rest (Tier 1: `_init_project_state`,
-  `_commit` scaffold; Tier 2: fetch + attempt-loop with callbacks; Tier 3 defer the run-loop). Each tier
-  = its own PR, suite + both-mode fake run between steps - deliberately NOT bulldozed (drift-bug history).
+- **Book↔article de-dup IN PROGRESS (safe, test-gated; plan.md §20):** mapped it (Explore agent), then
+  extracted `_run_learner` (learner tail) and **Tier 1 `_base_run_state`** (shared run-state keys for
+  `start_book`/`start_article`; mode-specific keys spread in). Suite green after each. **Deliberately
+  did NOT merge `_commit`/`_commit_section`** - the paths differ structurally (canon vs
+  citation-renumber), so a shared helper reads worse than the small dup (good judgment ≠ blind DRY).
+  Remaining (own PRs): Tier 2 (fetch shell + attempt-loop via callbacks), Tier 3 (defer run-loop).
 - **Still flagged:** giant `shell.py` (144KB) / `orchestrator.py` (115KB) - split AFTER the dedup (do
   §20 first, since the dup is intertwined). Optional: consolidate `SampleRun/`→`examples/`;
   collapse `requirements.txt`→pyproject+lock.

src/book_agent/orchestrator.py

@@ -347,6 +347,29 @@ def _record_author(uid: str, author: str | None) -> None:
 
 
 # ── Setup (human picks a direction, then autonomous) ─────────────────────────
+def _base_run_state(uid, abstract, *, intake, author, max_revisions, autonomous, humanize,
+                    settings) -> dict:
+    """Run-state keys shared by every project (book + article). Callers spread this and add
+    the mode-specific keys (the id, phase, unit counter, and any per-mode quality flags)."""
+    return {
+        "uid": uid, "abstract": abstract,
+        "intake": intake or "", "author": author or "",
+        "max_revisions": max_revisions, "committed": 0, "pending_review": False,
+        "use_researcher": settings.use_researcher,
+        "deep_research": settings.deep_research,
+        "autonomous": autonomous,
+        "humanize": settings.humanize if humanize is None else humanize,
+        "use_images": settings.use_images,
+        "diagram_engine": settings.diagram_engine,
+        "use_embeddings": settings.use_embeddings,
+        "divergent_drafts": settings.divergent_drafts,
+        "tournament_judge": settings.tournament_judge,
+        "min_insight": settings.min_insight,
+        # Autonomous runs never pause on low confidence.
+        "escalate_below_confidence": 0.0 if autonomous else settings.escalate_below_confidence,
+    }
+
+
 def start_book(
     cfg: ModelConfig, settings: Settings, uid: str, abstract: str,
     chosen: S.Direction, book_id_override: str | None,
@@ -369,23 +392,14 @@ def start_book(
     brain.write_text(paths.toc, render.render_toc_md(toc))
 
     state = {
-        "uid": uid, "book_id": book_id, "abstract": abstract,
-        "intake": intake or "", "author": author or "",
+        **_base_run_state(uid, abstract, intake=intake, author=author,
+                          max_revisions=max_revisions, autonomous=autonomous,
+                          humanize=humanize, settings=settings),
+        "book_id": book_id,
         "phase": "chapters", "num_chapters": len(toc.chapters),
-        "max_revisions": max_revisions, "consolidate_every": settings.consolidate_every,
-        "current_chapter": 1, "committed": 0, "pending_review": False,
-        "use_researcher": settings.use_researcher,
-        "deep_research": settings.deep_research,
-        "autonomous": autonomous,
-        "humanize": settings.humanize if humanize is None else humanize,
-        "use_images": settings.use_images,
-        "diagram_engine": settings.diagram_engine,
-        "use_embeddings": settings.use_embeddings,
-        "divergent_drafts": settings.divergent_drafts,
-        "tournament_judge": settings.tournament_judge,
-        "min_insight": settings.min_insight,
-        # Autonomous runs never pause: no confidence/contradiction escalation.
-        "escalate_below_confidence": 0.0 if autonomous else settings.escalate_below_confidence,
+        "consolidate_every": settings.consolidate_every,
+        "current_chapter": 1,
+        # Autonomous runs never pause on contradictions either.
         "escalate_on_contradiction": False if autonomous else settings.escalate_on_contradiction,
     }
     brain.write_json(paths.run_state, state)
@@ -1593,30 +1607,19 @@ def start_article(
     brain.write_text(paths.root / "thesis.md", nodes.render_thesis(thesis))
 
     state = {
-        "uid": uid, "article_id": article_id, "abstract": abstract,
-        "intake": intake or "", "author": author or "",
-        "mode": "article",
+        **_base_run_state(uid, abstract, intake=intake, author=author,
+                          max_revisions=max_revisions, autonomous=autonomous,
+                          humanize=humanize, settings=settings),
+        "article_id": article_id, "mode": "article",
         "phase": "sections", "num_sections": len(outline.sections),
-        "max_revisions": max_revisions, "current_section": 1,
-        "committed": 0, "pending_review": False,
-        "use_researcher": settings.use_researcher,
-        "deep_research": settings.deep_research,
-        "autonomous": autonomous,
-        "humanize": settings.humanize if humanize is None else humanize,
-        "use_images": settings.use_images,
-        "diagram_engine": settings.diagram_engine,
-        "use_embeddings": settings.use_embeddings,
+        "current_section": 1,
         "article_cohesion": settings.article_cohesion,
         "strip_inline_citations": settings.strip_inline_citations,
         "rank_references": settings.rank_references,
         "table_read": settings.table_read,
         "table_read_revise": settings.table_read_revise,
-        "divergent_drafts": settings.divergent_drafts,
         "divergent_skeletons": settings.divergent_skeletons,
-        "tournament_judge": settings.tournament_judge,
         "verify_claims": settings.verify_claims,
-        "min_insight": settings.min_insight,
-        "escalate_below_confidence": 0.0 if autonomous else settings.escalate_below_confidence,
         "escalate_on_contradiction": False,
     }
     brain.write_json(paths.run_state, state)