Pattern Proposal: Deterministic Context Grounding via Git-Delta Fact Ingestion

Summary

Transforms raw Git changes into validated PromptFact records before presenting context to LLM agents.

Candidate Pattern

• ID: git-delta-prompt-fact-projection
• Classification: intelligence
• Maturity: strategic_direction
• Type: emerging

Observed Signals

No specific signals listed.

Evidence

Supporting Evidence

• src/cas_runner/steering/models.py (GitChangeType, RawGitDelta, PromptFact, SteeringRuntimeIndex, ArchitecturalDelta) - Typed structures modeling code deltas as prompt facts.

Scope Assessment

• Evidence Strength: high
• Confidence: high
• Applies to: ideas-to-life-cas

Rationale

• Intent: Eliminate LLM hallucinations regarding code changes by strictly binding prompt context to version-controlled deltas.
• Context: Continuous architecture analysis runs during CI/CD workflows to inspect commits and merge requests.
• Force: Raw Git diffs can be noisy and exceed LLM context windows
• Force: Agents need typed metadata about change intent rather than raw unstructured strings

Traceability

Artifact source: proposed-patterns