consolidate-research

Consolidate Research

Synthesize research outputs from multiple sources into a unified, confidence-tiered, pattern-aware report with provenance tracking and quality metrics. Works standalone or manifest-driven from create-research-brief.

Shared references: consolidation-manifest-schema.md (schema), pattern-registry.md (patterns + defaults)

Workflow Overview

1. RECEIVE INPUTS ─── Manifest detection, source identification
2. NORMALIZE INPUTS ── Model-specific extraction → common claim format
3. TRIAGE CLAIMS ───── Atomic decomposition, dependency graphs, provenance
4. RECONCILE ───────── Provenance-weighted disagreement, citation dedup
5. SYNTHESIZE ──────── Cross-domain synthesis, structural artifact merging
6. GENERATE OUTPUT ─── Pattern-specific template, quality metrics, freshness
7. SELF-REVIEW ─────── 7 mandatory checks before finalizing

---

Step 1: Receive & Validate Inputs

Manifest Detection

Input contains YAML block with `consolidation_manifest:` label?
│
├─ YES (manifest-driven)
│   ├─ Parse manifest fields: research_id, pattern, models, coverage_matrix
│   ├─ Extract consolidation config: mode, verification_priorities
│   ├─ Load pattern-specific output template from Pattern Registry
│   ├─ Note research chain context (upstream_id, inherited_constraints)
│   └─ Use model identifiers from manifest — do NOT guess sources
│
└─ NO (standalone mode)
    ├─ Ask for or infer: topic, sources, consolidation mode
    ├─ Derive pattern from content using Pattern Registry decision tree
    ├─ Use pattern defaults from Quick Reference table
    └─ Confirm inferred pattern + mode with user before proceeding

Source Identification Heuristics

When manifest is absent and source is unlabeled, apply capability-aware signals:

Signal	Likely Source	Strength
Deep reasoning chains, cross-domain analogies, self-corrections, hedged nuance	Claude Opus 4.6	primary_researcher
Structured tables with dense citations, systematic catalogs, source appendix	Gemini 3.1 Pro Deep Research	structured_cataloger
Site-specific citations, temporal progression ("as of [date]"), intervention markers	GPT-5.2 Deep Research	targeted_investigator
Quick facts, recent dates, sentiment language, short-form responses	GPT-5.2 Chat	recency_validator
Proprietary data, specific methodologies, organizational context	Human analyst	domain_expert

If uncertain after heuristic check, ask user to confirm. When manifest is present, always use manifest.models[].model_id.

Validation

Research outputs provided?
├─ NO → Ask: "Please share the research outputs to consolidate."
└─ YES → Sources identifiable?
         ├─ NO → Ask: "Which source produced each output?"
         └─ YES → ≥2 sources?
                  ├─ NO → Warn: single-source consolidation has limited value.
                  │        Offer: proceed as structured review, or add sources.
                  └─ YES → Proceed to Step 2.

---

Step 2: Normalize Inputs

Convert heterogeneous model outputs into a common claim format before triaging.

Extraction Patterns by Source

Source	Extract	Tag
Claude Opus 4.6	Reasoning chains (analytical steps, not just conclusions)	type: reasoning_chain
	Conclusions linked to their supporting chains	type: factual/causal
	Web search findings with source URLs	channel: web_search
	Cross-domain insights and analogies	type: cross_domain_synthesis
	Self-review corrections (higher-confidence refinements)	type: factual
Gemini 3.1 Pro	Tables — preserve structure, do not flatten	type: structural_artifact
	Prose claims with inline citation mapping	type: factual/causal
	Source appendix — map citations to claims, compute quality scores	provenance metadata
	Comparison matrices with per-cell provenance	type: structural_artifact
	file_search-grounded claims	channel: internal_document
GPT-5.2 Deep	Assertions with temporal markers ("as of [date]")	provenance.temporal_marker
	Site-specific findings with source domain	channel: site_restricted
	Intervention-adjusted findings (higher targeted confidence)	boosted weight
	Timeline/progression narratives	type: structural_artifact
GPT-5.2 Chat	Quick facts with recency dates	channel: quick_validation
	Sentiment signals	type: recommendation
	All Chat claims default to lower provenance weight	validation role

Normalized Claim Format

claim: {id: "C-{seq}", text: "...", type: factual|causal|quantitative|recommendation|unique|reasoning_chain|structural_artifact,
  source_model: claude-opus-4-6|gemini-3.1-pro|gpt-5.2-deep|gpt-5.2-chat|human,
  provenance: {channel: web_search|file_search|site_restricted|quick_validation|internal_document,
    site_restrictions: [], context_documents: [], citation_quality: 0-5, temporal_marker: "YYYY-MM-DD"|null},
  depends_on: [claim IDs]}

Citation Quality Scale

Score	Source Type	Examples
5	Primary	SEC filings, peer-reviewed, official stats, vendor docs
4	High-quality secondary	Gartner, Forrester, named-source journalism
3	General secondary	News, press releases, industry publications
2	Tertiary	Wikipedia, blogs, aggregators
1	Unsourced assertion	No citation trail
0	Unverifiable	Contradicts known facts or cites non-existent sources

---

Step 3: Triage Claims

Decompose normalized claims into an atomic claims matrix with dependency tracking.

Claim Types

Type	Handling
Factual assertions	Cross-validate; trace to primary sources
Causal claims	Map reasoning chains; note mechanism divergence
Quantitative data	Flag discrepancies >10%; verify primary source agreement
Recommendations	Tag as interpretation; link to supporting facts
Unique insights	Preserve with provenance flag; do not discard
Reasoning chains	Preserve structure; validate logical steps; compare across sources
Structural artifacts	Preserve format (tables, matrices, timelines); merge in Step 5

Claims Matrix

Rows = claims, columns = sources. (qN) = citation quality. Flag same-primary-source convergence separately.

             │ Claude │ Gemini │ GPT-5.2D│ GPT-5.2C│ Status
─────────────┼────────┼────────┼─────────┼─────────┤
Claim A      │ ✓ (q4) │ ✓ (q5) │  ✓ (q3) │         │ Convergent (diff sources)
Claim B      │ ✓ (q4) │ ✓ (q4) │         │         │ Convergent (same primary?)
Claim C      │ ✓ (q3) │        │         │         │ Unique
Claim D      │ ✓ (q4) │ differ │  differ │         │ Contested

Claim Dependency Graphs

Claim A (factual) ──supports──► Claim B (causal) ──supports──► Claim C (recommendation)

Rules: Recommendation confidence CANNOT exceed supporting facts. If a fact drops tiers, propagate to ALL dependents. Document chains in output.

---

Step 4: Reconcile

Provenance-Weighted Disagreement Protocol

Execute in order — stop at first resolution:

Step	Test	Resolution
1	Coexist? Different scope/timeframe/definition?	Preserve both with clarifying context
2	Provenance? Higher channel? (site_restricted > web_search > quick_validation; internal_document > web_search for proprietary)	Favor higher provenance; note alternative
3	Citation dedup? Same primary source(s)?	Treat as SINGLE-SOURCE (Tier 2 cap), not independent
4	Specificity? One more specific or better sourced?	Favor specificity; note alternative
5	Majority? Most sources agree?	Lead with majority; preserve dissent
6	All diverge	Flag "unresolved." Agentic mode: web search. Else: present all views

Confidence Tier Assignment

Tier	Threshold	Criteria
1 (High)	>75%	Cross-model from DIFFERENT primaries + avg quality ≥4 + falsifiable
2 (Moderate)	50-75%	Same-primary cross-model (cap) OR single quality ≥4 OR majority w/ avg 3
3 (Low)	<50%	Single-source quality <4 OR contested OR unsourced

Key rule: Same-primary-source agreement caps at Tier 2 — prevents "3 models citing same blog" inflation.

Dependency Propagation

After reconciling individual claims: (1) Walk dependency graph root-to-leaf. (2) Supporting claim dropped → dependent drops too (floor: Tier 3). (3) Recommendations on Tier 3 facts → flag: "requires independent verification." (4) Document all propagation in Methodology Notes.

False Confidence Audit

Apply to all Tier 1 candidates (mandatory in Adversarial mode):

Check	Question
Citation diversity	Different primary sources, or all citing same 2-3?
Specificity test	Falsifiable claim, or vague enough to be unfalsifiable?
Recency check	Could this have changed? Source publication dates?
Contrarian search	Credible dissent? What would skeptics say?
Mechanism check	Same mechanism = shared bias risk. Different mechanisms converging = higher confidence.

If concerns → downgrade to Tier 2 with explicit note.

---

Step 5: Synthesize

Transforms reconciled claims into integrated insights. Mandatory — not optional.

Cross-Domain Synthesis Pass (4 Required Prompts)

#	Prompt	Seeks
1	"Which findings, when combined, imply something neither source stated?"	Emergent insights
2	"Which constraints interact with findings from another domain?"	Constraint interactions
3	"Which consensus views look different through an unrelated domain's lens?"	Frame-breaking
4	"What would need to be true for the consensus to be wrong?"	Contrarian check

Tag all outputs claim_type: cross_domain_synthesis. Document pass execution even if yield is zero.

Structural Artifact Merging

Artifact	Merge Strategy
Tables	Union rows/columns; per-cell provenance; highlight conflicts
Timelines	Interleave chronologically; flag disputed dates with both versions
Decision trees	Merge branches; note divergent recommendations at same decision point
Matrices	Union dimensions; per-cell provenance; highlight scoring disagreements

---

Step 6: Generate Output

Template Selection

Manifest present with pattern → use directly. No manifest → infer from content via Pattern Registry decision tree; confirm with user.

Universal Sections (All Patterns)

#	Section	Content
1	Executive Summary	2-3 paragraphs: key findings, implications, overall confidence
2	Tier 1 Findings	High-confidence claims: claim + support + implication
3	Tier 2 Findings	Moderate-confidence: claim + support + caveat
4	Contested Areas	Per-source views + assessment + resolution path
5	Coverage Gaps	Expected vs actual coverage table with recommended actions
6	Unique Insights	Per-source single-source findings with provenance
7	Cross-Domain Synthesis	Emergent insights, constraint interactions, frame-breaking observations
8	For Downstream	Pattern-specific actionable outputs
9	Quality Metrics	Computed metrics block
10	Freshness Model	Staleness detection + refresh recommendations
11	Methodology Notes	Mode, sources, conflicts resolved, chain context, limitations
12	Self-Review Results	7-check outcomes summary

Pattern-Specific Sections

In addition to universal sections, each pattern adds:

Pattern	Additional Sections
landscape_mapping	Taxonomy + Player Inventory per category, White Space Map
comparative_evaluation	Weighted Decision Matrix (options x criteria), Sensitivity Analysis, Recommendation with flip conditions
implementation_pattern	Architecture Decision Catalog, Pattern Catalog by phase, Anti-Pattern Register
best_practices	8-Dimension Knowledge Base sections, Quick Reference Card
competitive_intelligence	Per-Competitor Strategic Profile, Competitive Dynamics Analysis
market_research	TAM/SAM/SOM with ranges, Segmentation Framework, Demand Drivers/Inhibitors
user_research	Persona Cards, JTBD Map (functional/emotional/social), Unmet Needs Hierarchy
economic_analysis	Financial Model (cost + value + ROI), Sensitivity Table, Benchmark Comparison
compliance_requirements	Requirements Register, Constraint Map, Governance Recommendations

"For Downstream" Actionability

Pattern	Downstream Content	Primary Target
landscape_mapping	Shortlist criteria + recommended evaluation set	comparative_evaluation
comparative_evaluation	Decision recommendation + selection rationale + ADR draft	Foundry / implementation_pattern
implementation_pattern	Architecture decision log + implementation checklist	Foundry (requirements)
best_practices	Skill-ready KB structure + Quick Reference Card	Claude Code skills
competitive_intelligence	Positioning strategy + differentiation matrix	Ignite (GTM)
market_research	Segment prioritization + entry strategy recommendation	Spark (ideation)
user_research	Persona cards + JTBD map + ranked unmet needs	Foundry + Spark
economic_analysis	Financial model summary + key assumptions + sensitivity ranges	Vantage
compliance_requirements	Requirements register + constraint map	Foundry

Quality Metrics

Compute and include in every output:

Metric	Formula	Meaning
coverage_ratio	claims addressed / total claims	Input coverage completeness
conflict_resolution_rate	resolved / identified	Disagreement handling
provenance_depth	% Tier 1 with multi-source from DIFFERENT primaries	Independent corroboration
actionability_score	% findings with downstream action	Output usefulness
staleness_risk	% claims with sources older than volatility threshold	Temporal reliability
dependency_chain_integrity	% recommendations with validated chains	Logical soundness
cross_domain_synthesis_yield	count of cross-domain insights	Synthesis value-add

Freshness Model

freshness_model:
  research_id: "{from manifest or generated}"
  research_pattern: "{pattern_id}"
  topic_volatility: high|medium|low   # from Pattern Registry
  estimated_half_life: 3_months|6_months|12_months
  recommended_refresh: "YYYY-MM-DD"   # execution date + half-life
  staleness_indicators: ["signal 1", "signal 2", "signal 3"]  # 3-5 observable triggers
  source_dates: {oldest: "YYYY-MM-DD", newest: "YYYY-MM-DD"}

---

Step 7: Self-Review (Mandatory)

Execute ALL checks before finalizing. Report results in the Self-Review Results section.

#	Check	Procedure	If Failed
1	Contradiction scan	Do claims in one section contradict another?	Reconcile or flag explicitly.
2	Confidence tier audit	Did any claim's tier shift during writing?	Update tier + propagate dependencies.
3	Dependency chain validation	Are all recommendation → fact chains consistent?	Flag broken chains; downgrade recommendations.
4	Coverage check	Compare actual coverage vs. manifest coverage_matrix (or inferred).	Document gaps in Coverage Gaps section.
5	Cross-domain synthesis check	Was the mandatory synthesis pass executed? Insights generated?	If skipped, execute now. If zero yield, document.
6	Quality metrics validation	Are metrics internally consistent? (e.g., denominator matches actual count)	Recompute.
7	Freshness check	Any source dates older than topic volatility threshold?	Flag in staleness_risk metric + Freshness Model.

---

Consolidation Modes (7)

Mode Quick Reference

Mode	Best For	Leads With	Key Differentiator
Standard	General synthesis	Convergent findings	Balanced tiering
Adversarial	High-stakes decisions	Stress-tested claims	False confidence audit on ALL Tier 1
Gap-Driven	Coverage completeness	Missing areas	Coverage matrix as primary frame
Confidence-Weighted	Executive / financial decisions	Evidence quality	Citation quality drives tiers
Depth-First	Strategic / analytical topics	Deepest reasoning	Reasoning chains featured
Breadth-First	Landscape / domain mapping	Comprehensive coverage	Completeness over depth
Agentic	Minimal-intervention consolidation	Full autonomous pipeline	Web search for gap-filling + verification

Mode Selection Logic

Priority: (1) User explicit > (2) Manifest recommended_mode > (3) Infer from context > (4) Pattern default.

Context inference signals: "high stakes"/"executive" → confidence_weighted | "due diligence"/"comprehensive" → gap_driven | suspiciously aligned outputs → adversarial | "quick synthesis" → standard | strategic/nuanced → depth_first | new domain/landscape → breadth_first | "agentic"/"hands-off" → agentic.

When inferred, confirm: "I'll use [Mode] based on [reasoning]. This will [brief description]. Proceed?"

Pattern x Mode Defaults

Pattern	Default Mode	Override Trigger
landscape_mapping	breadth_first	"deep-dive on key players" → depth_first
comparative_evaluation	confidence_weighted	"quick comparison" → standard
implementation_pattern	depth_first	"comprehensive catalog" → breadth_first
best_practices	gap_driven	"focus on anti-patterns only" → depth_first
competitive_intelligence	adversarial	"just the facts" → standard
market_research	standard	"high-stakes investment" → confidence_weighted
user_research	depth_first	"broad needs survey" → breadth_first
economic_analysis	confidence_weighted	"rough estimate" → standard
compliance_requirements	gap_driven	"highest-risk areas" → depth_first

Mode Procedures

Standard: (1) Identify convergent claims. (2) Flag divergent for reconciliation. (3) Apply disagreement protocol. (4) Tier by confidence (same-primary-source caps at Tier 2).

Adversarial: (1) Flag suspicious alignment on contested topics. (2) False Confidence Audit on ALL Tier 1. (3) Identify cross-source critique points. (4) Stress test: what falsifies aligned claims? (5) Downgrade where audit reveals concerns.

Gap-Driven: (1) Derive expected coverage from manifest or pattern template. (2) Audit actual vs expected. (3) Document total gaps (no coverage) + partial gaps (single-source). (4) Prioritize by downstream impact.

Confidence-Weighted: (1) Citation Quality Scale on every claim. (2) Per-claim confidence = (sum scores) / (max possible). (3) Weight by quality, not source count. (4) Single quality-5 > three quality-1.

Depth-First: (1) Find deepest source (usually Claude/human). (2) Lead with reasoning chains. (3) Validate key chain claims via other sources. (4) Fill breadth gaps from cataloger (Gemini). (5) Feature strategic insights prominently.

Breadth-First: (1) Find most comprehensive source (usually Gemini). (2) Lead with coverage completeness. (3) Add depth on high-importance areas. (4) Update with recency from GPT-5.2. (5) Completeness is primary metric.

Agentic: Autonomous end-to-end with minimal user intervention. (1) Ingest all + manifest. (2) Execute Steps 2-5 autonomously. (3) Web search to resolve conflicts. (4) Targeted searches for coverage gaps. (5) Draft with cross-domain synthesis. (6) Full self-review with web verification. (7) Compute metrics; revise if needed. (8) Final output. Requires: claude-opus-4-6, thinking: adaptive, effort: max. Use when: all models done, topic bounded, manifest present, user wants hands-off.

---

Research Chain Protocol

When manifest.research_chain.upstream_id is set: (1) Request upstream consolidated output if not provided. (2) Validate downstream vs upstream consistency. (3) Flag contradictions. (4) Propagate inherited_constraints. (5) Apply chain validations (below). (6) Note in Methodology: "Part of chain: [upstream] → [this]". If no upstream → standard consolidation.

Chain-Specific Validations

Chain Link	Validation
Landscape → Comparative	Do all evaluated options appear in upstream landscape? Flag new ones.
Comparative → Implementation	Does selected option match upstream recommendation? If different, document why.
Market → Landscape	Are market segments consistent with landscape scope? Flag scope drift.
Market → Competitive	Are market structure assumptions consistent? Flag divergence.
Compliance → Implementation	Do all patterns satisfy upstream constraint map? Flag violations.
User Research → Comparative	Do evaluation criteria map to identified user needs? Flag orphan criteria.

Constraint Propagation Rules

Upstream constraints are HARD by default — cannot contradict without explicit justification. If downstream evidence challenges an upstream constraint, flag as "constraint tension" (do not silently override). Document all propagated constraints in Methodology Notes.

---

Integration Notes

Manifest-driven (after create-research-brief): Manifest provides ALL config — skip derivation. Pattern template pre-selected. Coverage matrix enables precise gap detection. Chain context available. Model capabilities known — provenance tagging is precise. Verification priorities pre-identified.

Standalone: Derive config from content + user input. User may paste a manifest manually. Pattern detection via Registry decision tree. Source ID via heuristic table. Coverage from pattern defaults. Works with any source types.