/research-synthesis

Role: Principal Investigator (synthesis) Pipeline position: mid-chain

Reconciles the collected sources into a state-of-the-art map + literature matrix and states the explicit research gap the work addresses. Every load-bearing claim is adversarially verified against two or more independent sources, refute-by-default.

Canonical invocationCopy Anchor Link

/research-synthesis

With arguments:

/research-synthesis [--suite-name NAME]

InputsCopy Anchor Link

_inputs/source-ledger.md (the ranked, screened source index) plus the per-source extractions under sources/.

OutputsCopy Anchor Link

_inputs/synthesis.md — the SOTA map, the literature matrix, the contested-findings ledger, and the explicit one-sentence gap statement with the search coverage it rests on. It reports both the report count and the study count.

DownstreamCopy Anchor Link

/research-proposal (downstream consumer)

Workflow phasesCopy Anchor Link

The command follows the standard /research pipeline workflow:

1. Load context — Read the source ledger and the per-source extractions.
2. Verify the Sequence Gate — Confirm the source ledger exists; otherwise halt with Blocked: run /research-sources first.
3. Synthesize — Build the SOTA map + literature matrix, carrying each source's study id, evidence class, and integrity status from the ledger; reconcile agreements and contradictions; state the explicit gap and name the search coverage it rests on.
4. Adversarially verify — Every load-bearing claim is refuted-by-default against two or more independent studies.
5. Update the handoff manifest — Record the synthesis path and the contested-claims ledger.
6. Emit the synthesis — _inputs/synthesis.md.

Rigor floorCopy Anchor Link

Before the gap statement is finalized, the synthesis reconciles its terminology, findings, and recommendations against related sibling corpora (cross-corpus coherence); a conflict is recorded in the contested-findings ledger, never silently harmonized. The gap must resolve to a distinct, parsimonious contribution — not one already closed by a corroborated claim, and not decomposable into overlapping non-distinct elements.

Counting evidenceCopy Anchor Link

• Studies, not reports. Reports that share a study id in the ledger, studies that share a collected sample, and sources that cite each other count as one source; two papers from one author group are checked for shared data first. Certainty ratings count independent studies the same way.
• Integrity. A retracted or withdrawn source does not count toward corroboration, and a source whose integrity status is concern is never the sole support for a claim.
• Contrary evidence and coverage. A prior synthesis found in the search locates primary studies and does not count beside them. When the contrary-evidence class retained nothing, the finding is recorded as none found in the searches recorded, never as consensus, and the gap sentence names its scope in the same terms: the indexes, the terms and date range, the date last run, and any pass flagged incomplete.

Failure modesCopy Anchor Link

ExamplesCopy Anchor Link

# Synthesize the collected sources into a SOTA map + gap
/research-synthesis --suite-name <name>

Cross-referencesCopy Anchor Link

• Pipeline overview
• Commands reference