Epistemic Context Grounding

LOW CONTEXT SENSITIVITY:
- Task is bounded, clear outcome
- Domain well-known (basic CRUD, standard patterns)
- Example: "Add a button to the UI"

HIGH CONTEXT SENSITIVITY:
- Task involves parsing, data formats, integrations
- Domain has canonical specs/docs
- Example: "Fix data pipeline" -> needs format knowledge

SEARCH ORDER (adapt to your context OS structure):
1. specs/canonical/*.md       - Blessed architecture docs
2. specs/{project}/*.md       - Project-specific specs
3. **/CLAUDE.md               - Navigation guides
4. **/README.md               - Component documentation
5. knowledge_base/**/*.md     - Knowledge graph nodes

## Assumptions Audit

| # | Assumption | Can Verify? | Source |
|---|------------|-------------|--------|
| 1 | DB contains the field I need | YES | Query DB schema |
| 2 | Current data source is sufficient | MAYBE | Check if richer data exists |
| 3 | No canonical spec exists | NO | MUST SEARCH FIRST |

STRONG FALSIFIABILITY (preferred):
- Solution grounded in specific spec citation
- Claims verifiable against canonical source
- Example: "API uses OAuth 2.0" [GROUNDED: provider docs, auth section]

WEAK FALSIFIABILITY (warning):
- Solution based on assumptions about data
- Claims not verifiable without checking
- Example: "This field should be enough" [UNVERIFIABLE]

[GROUNDED: spec:line]    - Solution based on canonical spec
[INFERRED: from X + Y]   - Deduced from multiple sources
[UNGROUNDED]             - Assumption, needs verification

## Context Sensitivity Check

**Task:** [What user asked for]

**Classification:** [LOW | HIGH]

**Reasoning:**
- Does task involve data formats? [Y/N]
- Does task involve parsing/serialization? [Y/N]
- Does task involve external integrations? [Y/N]
- Is domain knowledge critical to solution? [Y/N]

If 2+ YES -> HIGH context sensitivity -> Continue to Step 2
If 0-1 YES -> LOW context sensitivity -> Skip to context-gap-analysis

Task received
    |
    +-- Data formats involved? --YES--+
    +-- Parsing/serialization? --YES--+
    +-- External integrations? --YES--+-- HIGH -> Continue
    +-- Domain knowledge critical? ---+
    |
    +-- None of above -----------------> LOW -> Skip to context-gap-analysis

# 1. Find canonical specs (adapt paths to your system)
Glob: "specs/canonical/*.md"
Glob: "**/*DATA_MODEL*.md"
Glob: "**/*SPEC*.md"

# 2. Search for domain terms
Grep: pattern="[domain-term]" path="specs/"
Grep: pattern="[key-concept]" path="knowledge_base/"

# 3. Check navigation guides
Read: relevant CLAUDE.md files
Read: relevant README.md files

# 4. Knowledge graph
Glob: "knowledge_base/**/*.md"
Grep: pattern="[domain-term]" path="knowledge_base/"

## Domain Knowledge Found

| Document | Relevance | Key Information |
|----------|-----------|-----------------|
| [[data-model.md]] | HIGH | Documents record types and schemas |
| [[CLAUDE.md]] | MEDIUM | Navigation to specs |
| None found | - | Need to explore further |

## Assumptions Audit

### Data Assumptions
1. [ ] I assume [X] is the data source
2. [ ] I assume [Y] field exists
3. [ ] I assume [Z] contains what I need

### Domain Assumptions
4. [ ] I assume no spec exists for this
5. [ ] I assume current approach is best

### Solution Assumptions
6. [ ] I assume [approach] will work
7. [ ] I assume no simpler solution exists

**Verification Plan:**
- Assumption 1: Check by [specific action]
- Assumption 4: Search specs/ first

## Falsifiability Assessment

| Assumption | Grade | Evidence |
|------------|-------|----------|
| DB has needed field | STRONG | Can query schema |
| Field data is sufficient | WEAK | Haven't checked what else exists |
| No spec exists | UNVERIFIED | MUST SEARCH BEFORE DESIGNING |

**Verdict:**
- If any UNVERIFIED assumptions -> Search first
- If WEAK assumptions -> Flag in design
- If all STRONG -> Proceed to implementation

## Epistemic Grounding Report

**Task:** [Original request]

**Context Sensitivity:** [LOW | HIGH]

**Domain Knowledge Found:**
- [[spec-file-1]] - [Relevance]
- [[spec-file-2]] - [Relevance]

**Assumptions Verified:**
- [X] Assumption 1 [GROUNDED: source]
- [ ] Assumption 2 [UNGROUNDED - needs checking]

**Falsifiability:**
- Solution grounding: [STRONG | WEAK | UNVERIFIED]
- Recommendation: [Proceed | Read specs first | Verify assumptions]

**Next Step:**
[Specific action - "Read data-model.md" or "Proceed to context-gap-analysis"]