Research Synthesis Engine

Research Synthesis Engine — Knowledge Generation Dual to Review Council

                    ┌─────────────────────────┐
                    │   FORMALIZATION PIPELINE │
                    └─────────────────────────┘

INPUT SIDE                                           OUTPUT SIDE
═══════════                                          ═══════════
Research Synthesis Engine                            Review Council
   (Α,Β,Γ,Δ,Ε)                                       (Σ,Φ,Ν,Λ,Ω)
   "What should we prove?"                            "Is the proof good?"
   Knowledge → Specifications                        Proofs → Verdicts
   SYNTHESIZE loop                                   RALPH loop

The Research Synthesis Engine is the generative counterpart to the Review Council. It transforms raw research findings, literature, cross-domain knowledge, and epistemic-discovery-engine outputs into structured, actionable specification packages that feed the formalization pipeline.

Routing

• USE FOR / DO NOT USE FOR / TRIGGERS — see the description field in the YAML frontmatter above. Same dispatch contract is restated here for in-skill discovery.
• USE FOR: running the 5-role synthesis engine (Α-S Scanner / Β-S Clusterer / Γ-S Distiller / Δ-S Verifier / Ε-S Publisher) — Scan-Cluster-Distill-Verify-Publish (SYNTHESIZE) loop, four product tiers (Quick / Standard / Deep / Comprehensive), duality with the review-council RALPH loop, domain cross-references, Mathlib coverage tracking.
• DO NOT USE FOR: single-source literature review (use @lean-research); proof or formalization work (use @lean-proof or @lean-ai-formalization); review of a finished synthesis (use @lean-review-council); Zettelkasten card emission only (use @lean-zettelkasten).
• TRIGGERS: synthesis, SYNTHESIZE loop, synthesis council, Α-S, Β-S, Γ-S, Δ-S, Ε-S, scanner, clusterer, distiller, verifier, publisher, quick synthesis, deep synthesis.

Workflow

1. Decide the synthesis tier (Quick / Standard / Deep / Comprehensive) per handbook Part 3.
2. Convene the 5-role SYNTHESIZE loop (handbook Part 2) over the input corpus — Α-S scans, Β-S clusters, Γ-S distills, Δ-S verifies, Ε-S publishes.
3. Run handbook Part 4 to pair with @lean-review-council whenever the synthesis touches a proof artifact.
4. Emit the synthesis product (handbook Part 3 templates), and hand off Zettel cards to @lean-zettelkasten.
5. Update Mathlib coverage status (handbook Part 9) if the synthesis revealed a new gap or closed one.

Recovery & STOP

• STOP if 2+ roles report contradictory distillations on the same input — escalate to @lean-research for evidence widening before re-running Δ-S.
• STOP if the SYNTHESIZE loop iterates 3+ times without convergence — switch tier (e.g., Quick → Standard) per handbook Part 6.
• STOP if the published product would over-claim Mathlib coverage — defer to handbook Part 9 cross-check.

Handoffs

• Predecessors: agent:gateway, skill:lean-research, skill:research-council, skill:epistemic-discovery-engine.
• Successors: skill:lean-zettelkasten, skill:lean-review-council, skill:lean-doc-feedback, skill:lean-report.

Detailed reference

Full methodology content (Parts 1–9) lives in references/research-synthesis-engine-handbook.md. Load that file when the skill is convened; the SKILL.md only carries the dispatch contract and a parts index.

See also

• ../../references/research-synthesis-engine-handbook.md — Full handbook (extracted from this skill)
• ../lean-zettelkasten/SKILL.md — Successor
• ../lean-review-council/SKILL.md — Successor
• ../lean-doc-feedback/SKILL.md — Successor
• ../lean-report/SKILL.md — Successor