Mutate sources in place, giving each a unique cite_key.

Pass 1: tree + structural gaps + current literature digest.

Pass 2: structural gaps + analytical sketches + targeted sources.

Build search query from theme keywords.

Convert a plain keyword query into Scopus TITLE-ABS-KEY syntax. Handles multi-word phrases and simple keyword combinations.

Basic consistency check: does the pipeline's claim about a work align with the work's actual abstract? Uses keyword overlap as a heuristi

Returns (exists, note). Crossref 200 means the DOI is registered.

Fallback: search OpenAlex by title.

Check if the local ConceptNet SQLite database exists and has data.

Map activated disciplines to theme_ids in config.json. Uses explicit override map first, then fuzzy matching as fallback. Explicit map pr

Re-check key in case it was set after initial import — lazy init.

Register with Consensus ourselves (follow_redirects=True) so we get a real server-issued client_id. Stored in token file so the SDK skips

Walk the rendered text, find sentences that carry one or more valid [CiteKey] markers, and return [{"claim": sentence, "cite_keys": [...]}].

Extract the instructions section from a review document.

Extract the instructions section from a review document.

Extract meaningful terms from problem statement. Removes stopwords and very short tokens.

Direct SQL fallback if database.py doesn't expose a helper.

Fetch ConceptNet relations for a term from local SQLite. Returns list of {rel, target, weight}. Checks pipeline.db cache first — queries

Pull relevant current sources per analytical gap from DB.

First initial from 'Last, First' or 'First Last'.

APA 7th author list: 1 author: Freire, P. 2 authors: Freire, P., & hooks, b. 3-20: Smith, A., Jones, B., & Lee, C.

Lazy import agents to give clear error if one is missing.

LLM synthesis — prunes false positives and catches what static map missed. candidate_themes is the list of theme_ids that cluster matching propos

Match expanded concepts against cluster trigger_concepts. Returns (activated_cluster_ids, activated_disciplines, bridge_concepts). A clu

authors column may be JSON array, semicolon list, or plain string.

Produce Break 0 review document.

Produce Break 0 review document.

Produce Break 1 review document — Grounder + Historian + Gaper outputs.

Produce Break 1 review document — Grounder + Historian + Gaper outputs.

Produce Break 2 review document — full synthesis for trajectory evaluation.

Produce Break 2 review document — full synthesis for trajectory evaluation.

Surgically remove hallucinated keys while preserving valid ones in the same bracket. [Good, Bad] -> [Good]. [Bad] alone -> [unsupported].

Dedicated Scribe path for the Understanding Map. Stages: 1. Build citable manifest from this run's sources. 2. Generate Understa

Attempt to salvage implications from a truncated JSON response. Extracts all complete { ... } implication objects found before the trunca

Semantic search via Consensus MCP — 200M+ peer-reviewed papers. Particularly valuable for Grounder because Consensus finds conceptually r

Search Google Books API for foundational books.

Search Open Library for foundational books — no key needed.

Search OpenAlex with contextual query.

Search Semantic Scholar with contextual query.

Use Claude's web search tool via Anthropic API to get broader coverage. Returns results as pseudo-sources for the synthesis prompt.

Write seminal_references.csv — one row per paper, sorted by category then year.

Write seminal_references.csv — one row per paper, sorted by category then year.

Write one Jekyll post per seminal paper. Filename: YYYY-MM-DD-slug.md (using date_collected or today)

Write one Jekyll post per seminal paper. Filename: YYYY-MM-DD-slug.md (using date_collected or today)

Write seminal_references.json with papers grouped by primary category. Structure: { "generated_at": "...", "total": N,

Write seminal_references.json with papers grouped by primary category. Structure: { "generated_at": "...", "total": N,

Mode 2: Pipeline feeder. If selected_themes provided (from concept mapper), uses those directly. Otherwise falls back to keyword matching

Context for Gaper — minimal. Gaper builds its own two-pass context internally from the DB. We just pass the problem and break instructions.

Context for Grounder — problem + Social intelligence.

Context for Historian — problem + seminal works from Grounder.

Context for Scribe — synthesis + directions + output spec.

Format a CitableSource as an APA 7th entry (text string, no URL).

Generate a formatted reference section. If tree_only=True (default), only includes sources referenced in the argument tree. Otherwise in

Generate the evaluation Markdown report.

Create the argument_tree table if it doesn't exist.

Return all seminal sources from the pipeline DB as a list of dicts. Optionally filter by run_id.

Return all seminal sources from the pipeline DB as a list of dicts. Optionally filter by run_id.

Load sources from the pipeline database for a given run.

Load full source records by their IDs.

Get all source_ids from the argument tree for a run.

Simple keyword-based theme matching. Returns (selected_themes, excluded_themes). Each theme has keywords — if any keyword seed appea

Extract weight from the JSON metadata field.