multi-source-signal-synthesiser
Skill13 starsChanged 5 months agoMITView on GitHub
Synthesises user signals from multiple research sources into a
In this repository23 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- ambiguity-resolverSkill
- challengeSkill
- deep-thinkSkill
- design-critique-templateSkill
- experiment-designSkill
- ingestSkill
- interview-analysisSkill
- next-advisorSkill
- pipeline-stepsSkill
- post-launch-reviewSkill
- problem-structuringSkill
- product-requirements-docSkill
- setup-initiativeSkill
- solution-scoringSkill
- strategic-narrative-generatorSkill
- system-design-docSkill
- technical-spec-documentSkill
- tracking-and-funnelsSkill
- ui-pattern-librarySkill
- user-persona-builderSkill
- user-story-generatorSkill
- user-testingSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 33
Every signal must be tagged with evidence type and confidence score:
- Distinctive: line 42
Conflict resolution: When REAL contradicts SYNTHETIC, REAL wins.
- Distinctive: line 62
Don't average away disagreement — it usually means different user segments
- Very specific: line 119
Every verdict this synthesis produces MUST land in the hypothesis registry — hypotheses.py set <id> --status confirmed|refuted|testing …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Multi-Source Signal Synthesiser Skill
Purpose
Reconcile user signals from multiple sources — interviews, support tickets, NPS, app reviews, analytics, surveys, synthetic research — into a unified, weighted insight brief that surfaces the underlying need rather than the surface-level request.
Pipeline Context
Primary use: Step 6 (/validate-problems) — when combining:
- Analytics data (from analyst)
- Survey results (from research)
- Interview notes (from PM)
- Synthetic interviews (from step 2)
Also useful at any step where multiple data sources need reconciliation.
Evidence Typing
Every signal must be tagged with evidence type and confidence score:
| Type | Confidence | Source examples |
| REAL | 0.6 - 1.0 | Analytics data, survey results, user interviews, A/B test results |
| SYNTHETIC | 0.2 - 0.4 | AI-generated interviews, synthetic personas |
| INFERRED | 0.3 - 0.5 | Logical deductions, cross-referencing patterns |
| AMBIGUOUS | 0.1 - 0.3 | Contradictory signals, unclear data |
Conflict resolution: When REAL contradicts SYNTHETIC, REAL wins. Document the delta — the gap between what synthetic research predicted and what real data showed is itself an insight.
Source Weighting (default — adapt to your context)
- Direct research (interviews, usability tests): weight 5
- Analytics data (funnels, cohorts, events): weight 5
- Support tickets (unprompted pain signals): weight 4
- Survey results (structured quantitative): weight 4
- NPS verbatims: weight 3
- App store reviews: weight 2
- Sales call summaries (filtered through sales lens): weight 2
- Synthetic research (AI-generated): weight 1
- Anecdote or single report: weight 1
Process
- Accept inputs from any combination of source types
- Tag each signal by source, apply weight, and assign evidence type + confidence
- CONVERGENCE: same underlying need appearing across 3+ sources
- Calculate combined confidence: highest individual confidence × (1 + 0.1 × number of confirming sources)
- Cap at 1.0
- DIVERGENCE: contradictory signals suggesting user segmentation
- Don't average away disagreement — it usually means different user segments
- FREQUENCY RANKING: count how many independent sources mention each insight
- "N out of M sources mention this" (inspired by frequency-based evidence ranking)
- Distinguish surface request from underlying need (e.g., "faster export" may mean "I don't trust the data will be there when I need it")
- Produce ranked insights by weighted frequency
Output Format
User Signal Synthesis — [Date / Period]
Sources included: [list with evidence type for each] Total signals processed: [n] Evidence quality: [% REAL / % SYNTHETIC / % INFERRED]
Insight 1: [Underlying need, not feature request]
- Frequency: [N/M sources] — [list which sources]
- Evidence type: [REAL/SYNTHETIC/INFERRED] combined confidence: [0.0-1.0]
- Evidence:
- Analytics: [specific data point] (REAL, 0.8)
- Survey: [specific finding] (REAL, 0.85)
- Interviews: [quote or pattern] (REAL, 0.9)
- Synthetic: [what AI predicted] (SYNTHETIC, 0.3)
- Conflicting signals: [Any contradicting evidence and how to interpret it]
- REAL vs SYNTHETIC delta: [Where synthetic research got it wrong/right]
- Product implication: [Specific, not generic]
[Repeat for top 3-5 insights, ordered by combined confidence × frequency]
Divergent Signals (Possible Segmentation)
[Where user groups appear to have genuinely different needs]
- Segment A says X (evidence: ...) while Segment B says Y (evidence: ...)
- Implication: [consider separate solutions or prioritize one segment]
What the Data Does NOT Tell Us
[Gaps that require further research before acting]
- [Gap 1]: would need [research method] to resolve
- [Gap 2]: low confidence because only SYNTHETIC evidence exists
Confidence Summary
| Hypothesis | Evidence Sources | Combined Confidence | Recommendation |
| P1: ... | Analytics + Survey + Interviews | 0.93 | Confirmed |
| P2: ... | Synthetic only | 0.35 | Needs validation |
| P3: ... | Analytics + contradicts Survey | 0.55 | Investigate segment split |
Worked example (one insight)
Insight: users abandon at payment because they distrust the embedded form, not because of price. Converges: survey Q7 (41% "не доверяю оплате", n=380, REAL) + 6 of 8 interviews (REAL) + session recordings show form hesitation (REAL). Diverges: NPS verbatims mention price (12 mentions) — likely a different segment (new vs returning); flag for segmentation. Verdict: H3 confirmed, REAL 0.8, three sources.
Output to the registry (mandatory)
Every verdict this synthesis produces MUST land in the hypothesis registry —
hypotheses.py set <id> --status confirmed|refuted|testing --type <T>
--confidence <C> --add-source "research/<file>::<where>" per hypothesis.
Conflicting sources → --flag data_inconsistency instead of silently
averaging. A synthesis that only writes prose is invisible to gates and /next.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

