collecting-evidence - SKILL.md Agent Skill

name: collecting-evidence description: Use when researching a specific pillar and need to create traceable evidence objects. Guides creation of YAML evidence files with semantic IDs, confidence scores, and assumptions. context: fork agent: general-purpose

Evidence Collection

This skill guides the creation of structured Evidence Objects for a single research pillar.

Prerequisites

Ledger workspace initialized (/ledger-init completed)
Pillar assignment (which pillar to research)
Research scope from 01-pillars/PILLARS.md

Workflow

Use TodoWrite to track these mandatory steps:

1. Load pillar scope and research questions 2. Identify evidence sources 3. Collect raw evidence 4. Create Evidence Objects with semantic IDs 5. Validate evidence quality 6. Check evidence gate (minimum 5 per pillar)

Step 1: Load Pillar Scope

Read 01-pillars/PILLARS.md to understand:

Pillar priority level
Specific research questions for this pillar
Any scope restrictions

See references/research-protocols.md for pillar-specific protocols.

Step 2: Identify Evidence Sources

For each research question, identify potential sources:

Source Type	Examples	Typical Confidence
`url`	Research reports, documentation	60-90
`pdf`	Academic papers, whitepapers	70-95
`interview`	User interviews, expert calls	50-80
`internal-doc`	Company data, prior research	60-85
`experiment`	A/B tests, prototypes	70-95
`dataset`	Analytics, survey results	65-90

Step 3: Collect Raw Evidence

For each source:

Extract the key claim(s)
Note supporting quotes/data
Assess confidence level
List assumptions

Web research protocol:

Use WebSearch for discovery
Use WebFetch to retrieve and analyze content
Track source URLs and retrieval dates
Note if sources agree or contradict

Step 4: Create Evidence Objects

Write YAML files to 02-evidence/<pillar>/.

Naming: Use semantic IDs per references/id-generation-rules.md.

Schema: See references/evidence-object-schema.md.

```yaml id: EV-market-pricing-smb-wtp pillar: market source: type: url ref: "https://example.com/pricing-research" retrieved_at: 2026-01-21 claim: "SMB segment willingness-to-pay peaks at $29/mo for productivity tools." quote: "Our survey of 500 SMBs found median WTP of $29/month..." confidence: 0.75 assumptions: - "Survey sample representative of target market" - "WTP for 'productivity tools' applies to our specific category" notes: "Sample skewed toward US companies. May need regional validation." tags: - pricing - smb - wtp ``` - Semantic ID describes content (market-pricing-smb-wtp) - Falsifiable claim with specific number ($29/mo) - Honest confidence (0.75, not inflated) - Explicit assumptions documented - Source fully traceable

```yaml id: EV-001 pillar: market source: type: url ref: "some website" claim: "People like our product" confidence: 0.95 assumptions: [] ``` - Non-semantic ID (EV-001 tells nothing about content) - Vague, unfalsifiable claim ("people like") - Overconfident (0.95) without strong source - No assumptions documented - Untraceable source reference

Step 5: Validate Evidence Quality

Each Evidence Object must pass:

Check	Requirement
Falsifiable claim	Claim can be proven wrong
Confidence assigned	0.0-1.0 value present
Assumptions listed	At least 1 assumption
Source traceable	Can revisit the source
ID is semantic	Follows ID scheme

Quality warnings:

Confidence > 0.9 without peer-reviewed source
Multiple evidence objects with identical claims
Assumptions that can't be tested

Step 6: Check Evidence Gate

Before synthesis, verify minimum 5 Evidence Objects per pillar.

Evidence Gate Check: market
├── EV-market-tam-b2b-saas ✓
├── EV-market-pricing-smb-wtp ✓
├── EV-market-growth-remote-tools ✓
├── EV-market-segment-priorities ✓
└── EV-market-competitive-density ✓
Total: 5/5 minimum ✓ GATE PASSED

If gate fails, continue research until threshold met.

User Interaction

Use the AskUserQuestion tool when:

Source prioritization needed

Question: "Multiple sources available for [topic]. Which to prioritize?"
Options:
- "Academic/peer-reviewed sources (higher confidence)"
- "Recent industry reports (more current)"
- "Direct user research (more specific)"
- "Research all and compare"

Confidence assessment uncertain

Question: "How confident should I rate this claim: '[claim]'?"
Options:
- "High (0.8-0.9) - Strong source, well-supported"
- "Medium (0.5-0.7) - Reasonable source, some uncertainty"
- "Low (0.3-0.5) - Weak source or significant assumptions"
- "Help me assess the source quality"

Contradictory evidence found

Question: "Sources disagree on [topic]. Source A says X, Source B says Y."
Options:
- "Create evidence for both, note contradiction"
- "Prioritize more recent source"
- "Prioritize more authoritative source"
- "Research further for resolution"

Evidence gate failing

Question: "Only [N] evidence objects for [pillar]. Need [5-N] more to pass gate."
Options:
- "Continue researching this pillar"
- "Accept partial evidence (will affect synthesis quality)"
- "Deprioritize this pillar for MVP"
- "Help me identify additional research areas"

Output

After evidence collection:

## Evidence Collection Complete: [pillar]

**Evidence Objects Created:** [N]
**Gate Status:** [PASSED/FAILED]

### Evidence Summary
| ID | Claim Summary | Confidence |
|----|---------------|------------|
| EV-market-tam-b2b-saas | TAM is $X billion | 0.80 |
| EV-market-pricing-smb-wtp | SMB WTP peaks at $29/mo | 0.75 |
| ... | ... | ... |

### Key Findings
- [Top 3 findings from this pillar]

### Contradictions Noted
- [Any conflicting evidence]

### Gaps Remaining
- [Research questions not fully answered]

References

references/evidence-object-schema.md - YAML schema
references/research-protocols.md - Pillar-specific research guidance
references/id-generation-rules.md - Semantic ID creation