> ## Documentation Index > Fetch the complete documentation index at: https://wuweism.com/llms.txt > Use this file to discover all available pages before exploring further. # Hybrid Synthesis > Surface genuine novelty by reconciling conflicting claims across a literature corpus and company data. Hybrid Synthesis is designed for one specific problem: finding the gap where existing knowledge fails to agree. Rather than summarizing what is known, it actively searches for conflicts between sources — papers, filings, reports — and builds novel hypotheses in the space where those conflicts live. The approach is inspired by Hong-style recombinant synthesis: take heterogeneous, sometimes contradictory epistemic units and find what emerges when you force them to reconcile. Navigate to Hybrid Synthesis by selecting **Hybrid Synthesis** in the sidebar, or by opening `/hybrid` directly. ## What Hybrid Synthesis produces A completed synthesis returns: * **A structured approach**: the methodological framing for investigating the identified novelty * **Novel ideas with confidence scores**: candidate hypotheses ranked by how strongly they pass the novelty gate * **Thesis statements**: falsifiable claims derived from the synthesis, ready to record as governed claims The top claim from each synthesis run is automatically recorded in your [Claim Ledger](/concepts/claim-ledger). ## Providing sources Hybrid Synthesis accepts two kinds of inputs: PDF documents and company names. You can use either or both. Upload up to **6 PDF files** per synthesis run. These can be journal articles, preprints, reports, or any text-bearing document. In the source panel, click **Upload PDFs** or drag your files directly onto the dropzone. Choose your PDF files from your local filesystem. Files are processed in parallel during ingestion. Each uploaded file appears in the source list with its filename. Remove any file by clicking the **×** next to it before running synthesis. Each PDF is run through scientific extraction to identify numeric evidence — figures, measurements, effect sizes, confidence intervals — before synthesis begins. This extraction feeds the novelty gate's evaluation of evidence quality. Enter up to **5 company names** as sources. Wu-Weism retrieves publicly available information associated with each company (filings, publications, announcements) and treats this corpus as additional source material. In the source panel, click **Add Company** or type directly into the company name field. Type the company name exactly as it appears in public records (e.g., "Recursion Pharmaceuticals", "DeepMind"). Press **Enter** to add it. Repeat for each company. The panel shows your current list and prevents adding more than 5. Company-sourced data is retrieved at synthesis time from public sources. The recency and completeness of that data depends on what is publicly indexed. For proprietary or internal data, upload a PDF export instead. ## Research focus (optional) You can optionally specify a **research focus** — a short phrase or question that guides the synthesis engine toward the region of the source material most relevant to your investigation. Without a research focus, the synthesis engine explores the full conflict space across all sources. With a focus, it prioritizes conflicts and novelty that bear on your specified question. Examples of useful research focus inputs: * "mechanisms linking neuroinflammation to synaptic pruning" * "causal pathways between ESG disclosure and cost of capital" * "interventions that reduce antibiotic resistance in hospital settings" A vague focus (e.g., "anything interesting") is treated the same as no focus. Be specific. ## The synthesis pipeline Once you start synthesis, Wu-Weism runs a multi-stage pipeline. Progress is shown live in the UI as each stage completes. ```mermaid theme={null} flowchart LR A[ingestion] --> B[pdf_parsing] B --> C[entity_harvest] C --> D[synthesis] D --> E[novelty_gate] E -->|pass| F[completed] E -->|fail| G[recovery_plan] G --> F ``` | Stage | What happens | | ---------------- | --------------------------------------------------------------------------------------------------------- | | `ingestion` | Sources are received and queued for processing | | `pdf_parsing` | PDFs are parsed; text, structure, and numeric evidence are extracted | | `entity_harvest` | Named entities, claims, and causal assertions are harvested from all sources | | `synthesis` | Conflicting claims are identified; recombinant hypothesis generation runs across the conflict space | | `novelty_gate` | The generated hypothesis is evaluated against a novelty criterion (see below) | | `recovery_plan` | If the novelty gate fails, the system attempts a recovery synthesis targeting a different conflict region | | `completed` | Final output is ready | Progress events stream in real time. You can read intermediate status messages as the pipeline advances. ## The novelty gate The novelty gate is the quality filter that separates genuine insight from restatement. Before returning output, Wu-Weism evaluates whether the synthesized hypothesis: 1. Is not already explicit in any of the source documents 2. Is not a trivial combination of existing claims 3. Has at least one point of tension with received consensus that it resolves or reframes If the hypothesis passes, it proceeds directly to `completed`. If it fails, the system enters `recovery_plan` — a second synthesis pass that targets a different conflict region in the corpus. The recovery output is also evaluated by the novelty gate before being returned. The novelty gate is conservative by design. It is better to return a narrower, well-supported novel claim than a broader, weakly-supported one. A recovery plan result is not a failure — it simply reflects that the first synthesis pass produced something the system judged as insufficiently novel. ## Claim recording After synthesis completes, the highest-confidence novel claim is automatically written to your [Claim Ledger](/concepts/claim-ledger). You can review, annotate, and export it from there. All other generated claims appear in the synthesis output panel and can be manually added to the Claim Ledger by clicking **Record claim** next to each one. ## Related pages Single-document causal extraction — a lighter-weight alternative when you have one source. Understand how recorded claims are governed, annotated, and exported. Use the Causal Workbench to interrogate a synthesized hypothesis with SCM-backed dialogue. Track the epistemic quality of your synthesis outputs over time.