---

name: norvidize description: Extract and audit claims for norvid tracking system

Norvidize Skill

<<<<<<< HEAD Extract claims from documentation, code, and tests into a claim tracking system based on multicomputational irreducibility -- claims that cannot be reduced to simpler forms without losing verification guarantees.

Extract claims from documentation, code, and tests into the norvid claim tracking system.

origin/main

When to Use

• After completing a significant feature
• When auditing documentation accuracy
• Before releases to ensure claims match reality
• When docs/tests/specs diverge from implementation

<<<<<<< HEAD

Claim Registry Location

crates/basin-norvid/src/claims.rs

origin/main

Assurance Levels (lowest to highest)

Level	Meaning	Evidence Required
Mentioned	Just referenced somewhere	Any doc
<<<<<<< HEAD
Designed	Has design doc	Design markdown
Specified	Has formal spec	Spec file
Implemented	Code exists	Source file + symbol
Tested	Has passing tests	Test file + test name
Proven	Has formal proof	Proof file + theorem
=======
Designed	Has design doc	docs/design/*.md
Specified	Has sigil spec	*.sigil file
Implemented	Code exists	Rust file + symbol
Tested	Has passing tests	Test file + test name
Proven	Has Lean proof	Lean file + theorem

origin/main

Claim Categories

• Feature: User-visible functionality
• Property: System guarantee (linearizability, durability)
• Guarantee: Promise to users
• Optimization: Performance improvement
• Invariant: Internal system invariant
• Decision: Architectural choice with rationale
• Omission: Deliberate non-feature

<<<<<<< HEAD

Covariant Claim Structure

Claims form a presheaf over the assurance poset: higher assurance levels pull back evidence from lower levels. A Tested claim covariantly transports its Designed evidence forward.

Mentioned -> Designed -> Specified -> Implemented -> Tested -> Proven

Each arrow preserves evidence (covariant functor from assurance levels to evidence sets).

Evidence Types

// Documentation
Evidence::doc("docs/CLAIMS.md")

// Rust implementation
Evidence::rust("crates/foo/src/bar.rs", "SymbolName")

// Test file
Evidence::test("crates/foo/src/bar.rs", "test_prefix")

// Sigil specification
Evidence::sigil("sigil/specs/foo.sigil")

// External (benchmark, paper)
Evidence::external("benchmarks/harnesses/redis/src/main.rs")

Adding Claims

registry.register(
    Claim::new("CLM_UNIQUE_ID", "Short description")
        .with_category(Category::Feature)
        .with_component(Component::Slate)
        .with_assurance(Assurance::Implemented)
        .with_evidence(Assurance::Designed, Evidence::doc("docs/design/foo.md"))
        .with_evidence(Assurance::Implemented, Evidence::rust("crates/slate/src/lib.rs", "FooStruct"))
);

origin/main

Extraction Process

1. Find Candidate Claims

<<<<<<< HEAD Scan sources for claim candidates:

grep -r "CLAIM\|GUARANTEE\|INVARIANT" docs/
grep -r "#\[test\]" -A 2 src/ | grep "fn test_"
grep -r "ops/s\|latency\|throughput" docs/

2. Categorize by Irreducibility

• Irreducible claims: Cannot be verified by simpler means (need full test execution)
• Reducible claims: Can be statically verified (type checking, linting)
• Compositional claims: Verified by composing sub-claim verifications

3. Verify Evidence Paths

Before adding at Tested/Implemented level, verify paths exist.

4. Downgrade Strategy

If evidence path is uncertain, downgrade assurance level. Upgrade later when evidence is confirmed.

Anti-Patterns

• Don't add Tested claims without verifying paths
• Don't skip Designed level -- even Tested claims should have design evidence
• Don't guess component names ======= Scan these sources for claim candidates:

# Explicit claims in docs
grep -r "CLM-\|CLAIM\|GUARANTEE\|INVARIANT" docs/

# Test names suggest guarantees
grep -r "#\[test\]" -A 2 crates/ | grep "fn test_"

# Performance claims
grep -r "ops/s\|latency\|throughput" docs/

# Sigil specs
find sigil/ -name "*.sigil" -exec grep -l "claim\|invariant\|property" {} \;

2. Categorize and Deduplicate

Group into tiers:

• Tier 1 (SPEC_*): Claims with sigil specs - highest value
• Tier 2 (TEST_*): Claims with test evidence
• Tier 3: Design-level claims (docs only)

3. Verify Evidence Paths

Before adding at Tested/Implemented level, verify paths exist:

# Check file exists
test -f "crates/basin-redis/src/crdt.rs" && echo "OK"

# Check symbol exists
grep -l "EntityStore" crates/basin-meta/src/indexes/*.rs

4. Add to Registry

Add claims to claims.rs in appropriate section:

• SPEC_* claims near line 1720
• TEST_* claims after specs
• PERF_* claims in performance section
• Product claims (CLM_STASH, etc.) with products

CI Verification

The CI test filters claims at Tested/Proven level and verifies:

1. Evidence file exists
2. Symbol/test found in file

# Run verification
graft test -p basin-norvid test_all_claims_verified

Common Path Mappings

Old/Wrong Path	Correct Path
crates/stash/	crates/basin-redis/
crates/shelf/	crates/basin-mesh/ or crates/spool-core/
crates/basin-jepsen/	tests/basin-jepsen/
products/nfs/tests/	crates/nfs/tests/

Downgrade Strategy

If evidence path is uncertain, downgrade to Designed level:

// Was Tested but can't verify path
.with_assurance(Assurance::Designed)
.with_evidence(Assurance::Designed, Evidence::doc("docs/CLAIMS.md"))

Upgrade later when evidence is confirmed.

Parallel Audit Pattern

For bulk verification, spawn sonnet agents:

Task: Audit PERF_* claims
- Check benchmark files exist
- Verify they actually measure what claim says
- Report which claims have valid evidence

Task: Audit product claims (CLM_STASH, etc.)
- Find actual test locations
- Verify symbol names
- Report path corrections needed

Anti-Patterns

• Don't add Tested claims without verifying paths - CI will fail
• Don't embed large values in claim IDs - Keep IDs short like CLM_STASH
• Don't skip Designed level - Even Tested claims should have design evidence
• Don't guess component names - Check Component enum in identity.rs

Quick Reference

# Find existing claims
grep -n "Claim::new" crates/basin-norvid/src/claims.rs | head -20

# Check CI status
graft test -p basin-norvid test_all_claims_verified 2>&1 | grep -E "FAILED|passed"

# Find test files for a crate
find crates/basin-meta -name "*.rs" -exec grep -l "#\[test\]" {} \;

# Count claims by category
grep -E "Category::" crates/basin-norvid/src/claims.rs | sort | uniq -c

origin/main