hypothesis-tracking

name	hypothesis-tracking
version	2.0.0
description	Confidence level management and hypothesis lifecycle
metadata	{"openclaw":{"requires":{"bins":["primr-mcp"],"env":["GEMINI_API_KEY"]}}}
mcp_server	primr
tools	["get_hypotheses","save_hypothesis"]
resources	["primr://memory/{company}"]

Hypothesis Tracking Skill (v2.0)

You are an expert at managing research hypotheses, tracking confidence levels, and maintaining epistemic rigor.

Confidence Level Framework

Primr uses a four-level confidence system:

Level	Meaning	Evidence Required
UNTESTED	Claim extracted, not yet verified	None (initial state)
VALIDATED	Supporting evidence found	At least one corroborating source
INVALIDATED	Contradicting evidence found	At least one contradicting source
CONFIRMED	High confidence, multiple sources	Multiple independent sources

Hypothesis Lifecycle

                    ┌─────────────┐
                    │  UNTESTED   │
                    └──────┬──────┘
                           │
           ┌───────────────┼───────────────┐
           │               │               │
           ▼               ▼               ▼
    ┌─────────────┐ ┌─────────────┐ ┌─────────────┐
    │  VALIDATED  │ │ INVALIDATED │ │  (remains)  │
    └──────┬──────┘ └─────────────┘ └─────────────┘
           │
           ▼
    ┌─────────────┐
    │  CONFIRMED  │
    └─────────────┘

Hypothesis Structure

hypothesis:
  id: "h_001"                    # Unique identifier
  claim: "Company uses AWS"      # The testable claim
  confidence: validated          # Current confidence level
  evidence:                      # Supporting/contradicting evidence
    - "Job posting mentions AWS certifications"
    - "CTO blog post discusses AWS migration"
  topic: "technology"            # Category for filtering
  created_at: "2026-02-01T10:00:00"
  updated_at: "2026-02-03T14:30:00"
  expires_at: "2026-05-01T10:00:00"  # Optional expiration

Operational Capabilities

1. Retrieve Hypotheses

Tool: get_hypotheses

# Get all hypotheses for a company
get_hypotheses(company="Acme Corp")

# Filter by confidence level
get_hypotheses(company="Acme Corp", confidence="validated")

# Filter by topic
get_hypotheses(company="Acme Corp", topic="technology")

# Include expired hypotheses
get_hypotheses(company="Acme Corp", include_expired=True)

2. Update Hypothesis Confidence

Tool: save_hypothesis

# Validate a hypothesis
save_hypothesis(
    company="Acme Corp",
    hypothesis_id="h_001",
    confidence="validated",
    evidence="Found AWS case study on company blog"
)

# Invalidate a hypothesis
save_hypothesis(
    company="Acme Corp",
    hypothesis_id="h_002",
    confidence="invalidated",
    evidence="CEO interview states they use Azure exclusively"
)

# Confirm with high confidence
save_hypothesis(
    company="Acme Corp",
    hypothesis_id="h_001",
    confidence="confirmed",
    evidence="Multiple sources confirm: job posts, blog, press release"
)

3. Create New Hypothesis

Tool: save_hypothesis

save_hypothesis(
    company="Acme Corp",
    hypothesis_id="h_new_001",  # New ID
    claim="Company is expanding into healthcare",
    confidence="untested",
    topic="strategy"
)

Evidence Quality Guidelines

Strong Evidence

• Direct quotes from company executives
• Official press releases
• SEC filings (for public companies)
• Published case studies
• Job postings with specific requirements

Moderate Evidence

• Industry analyst reports
• News articles with named sources
• Partner announcements
• Conference presentations

Weak Evidence

• Anonymous sources
• Speculation in articles
• Social media posts
• Outdated information (>1 year)

Hypothesis Generation Patterns

From Scrape Results

Analyst Subagent extracts claims:
- Technology stack mentions → technology hypotheses
- Leadership quotes → strategy hypotheses
- Financial mentions → financial hypotheses
- Market references → market hypotheses

From Deep Research

External sources provide:
- Competitive positioning claims
- Market share estimates
- Growth trajectory claims
- Industry trend alignment

Expiration Management

When to Set Expiration

• Financial claims: 90 days (quarterly updates)
• Technology claims: 180 days (tech changes)
• Strategy claims: 365 days (annual planning)
• Leadership claims: No expiration (until change)

Handling Expired Hypotheses

1. Query with include_expired=True
2. Review expired hypotheses
3. Either:
   - Re-validate with fresh evidence
   - Mark as invalidated if no longer true
   - Extend expiration if still relevant

Example Workflow

User: "What do we know about Acme Corp's technology?"

Agent:
1. get_hypotheses("Acme Corp", topic="technology")
   
   Results:
   - [CONFIRMED] Uses microservices architecture
     Evidence: CTO interview, job posts, tech blog
   - [VALIDATED] AWS as primary cloud
     Evidence: Job posting mentions AWS
   - [UNTESTED] Considering Kubernetes migration
     Evidence: None yet

2. Present to user:
   "Here's what we know about Acme Corp's technology:
   
   ✓ CONFIRMED: Uses microservices architecture
     - Multiple sources confirm this
   
   ✓ VALIDATED: AWS as primary cloud
     - One supporting source (job posting)
   
   ? UNTESTED: Considering Kubernetes migration
     - No evidence yet, extracted from blog speculation
   
   Would you like me to research any of these further?"

3. If user provides new evidence:
   save_hypothesis(
       company="Acme Corp",
       hypothesis_id="h_003",
       confidence="validated",
       evidence="User confirmed: spoke with Acme engineer at conference"
   )

Constraints

• Unique IDs: Each hypothesis must have a unique ID per company
• Evidence Trail: Always record evidence when changing confidence
• Topic Consistency: Use consistent topic names for filtering
• Expiration Awareness: Check expiration before presenting claims