intelligence-synthesis

name	intelligence-synthesis
description	Correlates signals across market research, customer feedback, and usage analytics to produce scored insights. Clusters related signals into themes, calculates evidence strength, and feeds into DF-017's scoring framework.
agents	{"primary":"business-analyst","supporting":["research-agent"]}
inputs	[".factory/discovery/feedback-digest-*.md (latest)",".factory/discovery/competitive-update-*.md (latest)",".factory/discovery/analytics-digest-*.md (latest, if exists)",".factory/discovery/product-research-*.md (latest DF-017 scan)"]
outputs	[".factory/discovery/insights-YYYY-MM-DD.md"]

Intelligence Synthesis

Correlates signals across all intelligence sources -- market research, customer feedback, competitive monitoring, and usage analytics -- to produce scored insights. This is the layer that turns raw signals into actionable intelligence by extracting themes, calculating evidence strength, and formatting insights for DF-017's scoring framework.

Trigger

• Scheduled -- weekly or as configured in discovery-config.yaml schedule.full_synthesis
• Manual -- human requests "Synthesize intelligence for [product]"
• Automatic -- runs after all ingestion steps complete in discovery.lobster

Prerequisites

• At least one input source must exist (market research at minimum)
• Business-analyst agent available
• Previous ingestion steps completed (or their outputs exist from prior runs)

Input Sources

Source	File Pattern	Required?	Producer
Market Research	.factory/discovery/feature-research-*-YYYY-MM-DD.md	YES	DF-017 discovery-engine
Customer Feedback	.factory/discovery/feedback-digest-YYYY-MM-DD.md	NO	customer-feedback-ingestion
Competitive Update	.factory/discovery/competitive-update-YYYY-MM-DD.md	NO	competitive-monitoring
Analytics Digest	.factory/discovery/analytics-digest-YYYY-MM-DD.md	NO	analytics-integration

The synthesis layer works with partial data. If only market research exists, it still produces insights (with lower evidence strength scores). As more sources come online, evidence strength increases.

Synthesis Workflow

Step 1: Collect Latest Inputs

Read the most recent file matching each input pattern. Track which sources are available for evidence strength scoring.

sources_available:
  market_research: true    # always required
  customer_feedback: true  # if digest exists
  competitive_intel: true  # if update exists
  usage_analytics: false   # if digest exists
source_count: 3            # out of 4 possible

Step 2: Extract Signals

From each source, extract individual signals:

From Market Research:

• Competitive features shipped
• Technology opportunities
• Industry trend shifts
• User signal patterns

From Customer Feedback:

• Top feature requests (by frequency)
• Top pain points (by severity)
• Emerging signals (new this period)

From Competitive Update:

• Competitor releases overlapping our roadmap
• Pricing changes affecting our positioning
• New entrants in our space

From Analytics:

• Features with declining adoption
• Error hotspots
• Funnel drop-off points
• Unused features

Step 3: Theme Extraction

Cluster related signals across sources into themes:

1. Semantic clustering: Group signals that refer to the same concept

   • Example: "dark mode" request (GitHub) + competitor ships dark mode (competitive) + 40% users on dark OS themes (analytics) = one theme

2. Cross-source correlation: A signal appearing in multiple sources is stronger than a single-source signal

3. Theme naming: Each cluster gets a descriptive theme title

4. Signal mapping: Each theme lists which signals from which sources contribute to it

Step 4: Evidence Strength Scoring

Score each theme's evidence strength based on source diversity and quality:

Evidence Level	Sources	Score Range
Market scan only	deep research (perplexity_research)	0.3 - 0.5
Market + one customer signal	Research + feedback OR analytics	0.5 - 0.6
Market + multiple customer signals	Research + feedback + analytics	0.6 - 0.8
Market + customer + competitive	Research + feedback + competitive	0.7 - 0.8
All three customer sources	Feedback + competitive + analytics	0.8 - 0.9
All sources + revenue impact	All + payment/revenue data	0.9 - 1.0

Within each range, score higher when:

• More independent signals confirm the theme
• Signals are recent (within 30 days)
• Signals come from diverse channel types (not just one GitHub repo)

Step 5: Priority Signal Analysis

For each theme, assess priority dimensions:

Dimension	Question	Weight
Pain frequency	How many independent reports?	HIGH
Revenue signal	Does this affect paying customers?	HIGH
Competitive urgency	Did a competitor just ship this?	HIGH
Trend direction	Is the signal growing or fading?	MEDIUM
User impact	How many users affected (from analytics)?	MEDIUM
Effort estimate	Quick win or major investment?	LOW (informational)

Step 6: Score Insights on 7 Dimensions

For each theme that merits an insight, apply DF-017's scoring framework with the new evidence_strength dimension:

Dimension	Weight	Description
Value	0.25	How much value does this create for users?
Feasibility	0.15	Can we build this with current tech/team?
Alignment	0.15	Does this fit our product vision?
Novelty	0.10	Is this a new capability (not incremental)?
Time-Criticality	0.10	Is there a window of opportunity?
Effort	0.10	How much work to implement? (inverse)
Evidence Strength	0.15	How many independent sources confirm this?

Step 7: Determine Recommended Action

Based on composite scores and evidence strength:

Composite	Evidence Strength	Recommended Action
>= 0.7	>= 0.6	Brief -> Planning (auto-generate brief)
0.5 - 0.7	any	Backlog (resurface in next report)
any	0.4 - 0.6	Backlog (may promote as evidence grows)
< 0.5	< 0.4	Registry (log for future, re-evaluate)
any (competitive HIGH)	>= 0.7	URGENT -> Immediate human notification

Step 8: Produce Insights Report

Write .factory/discovery/insights-YYYY-MM-DD.md:

---
document_type: intelligence-insights
date: YYYY-MM-DD
product: [product-name]
insights_count: [N]
sources_available: [N of 4]
---

# Intelligence Insights: YYYY-MM-DD

## Source Coverage
| Source | Available | Freshness |
|--------|----------|-----------|
| Market Research | YES/NO | [date] |
| Customer Feedback | YES/NO | [date] |
| Competitive Intel | YES/NO | [date] |
| Usage Analytics | YES/NO | [date] |

## Insight 1: [Theme Title]

### Signal Correlation
| Source | Signal | Date | Strength |
|--------|--------|------|----------|
| Customer Feedback | [signal] | [date] | strong/moderate/weak |
| Competitive | [signal] | [date] | strong/moderate/weak |
| Analytics | [signal] | [date] | strong/moderate/weak |
| Market Research | [signal] | [date] | strong/moderate/weak |

### Scores
| Dimension | Score | Rationale |
|-----------|-------|-----------|
| Value | [0.XX] | [why] |
| Feasibility | [0.XX] | [why] |
| Alignment | [0.XX] | [why] |
| Novelty | [0.XX] | [why] |
| Time-Criticality | [0.XX] | [why] |
| Effort | [0.XX] | [why] |
| Evidence Strength | [0.XX] | [source count + quality] |
| **Composite** | **[0.XX]** | |

### Recommended Action
[Brief -> Planning | Backlog | Registry | Monitor | URGENT]

## Insight 2: [Theme Title]
...

Working with Partial Data

The synthesis layer degrades gracefully:

• Market research only: Insights generated with evidence_strength 0.3-0.5. All themes noted as "unvalidated by customer data."
• Market + feedback: Evidence strength up to 0.6-0.7. Customer feedback confirms or contradicts market research findings.
• Market + competitive: Evidence strength 0.5-0.7. Competitive pressure validates urgency but lacks customer confirmation.
• All sources: Full evidence scoring up to 0.9-1.0.

Quality Gate

• All available signal sources collected and correlated
• Evidence strength scored for every theme based on source diversity
• Themes clustered from cross-source signals (not just listed per-source)
• Insights report written to .factory/discovery/insights-YYYY-MM-DD.md

Quality Criteria

• All available sources collected and referenced
• Theme extraction clusters related signals (not just lists)
• Evidence strength scored based on source count and quality
• Each insight scored on 7 dimensions (6 existing + evidence_strength)
• Recommended actions match the routing rules in discovery-engine SKILL.md
• Works with partial data (some sources may not be configured)
• Cross-source correlations identified and documented
• URGENT insights flagged for immediate notification