Zero-Token SHA-256 Hash Gating¶

AI code indexing tools often waste thousands of tokens re-analyzing files that haven't changed.

TLDRGraph solves this with a zero-token SHA-256 hash gating system backed by a local SQLite cache.

How Hash Gating Works¶

flowchart TD
    A[Source File] --> B[Compute SHA-256 Hash]
    B --> C{Match SQLite Cache?}
    C -->|Hash Identical| D[Skip AST & Re-enrichment]
    C -->|Hash Changed or New| E[Mark Node Dirty]
    E --> F[Re-parse AST]
    F --> G[Update SQLite Cache]

1. Content Hashing: Every source file is fingerprinted with a SHA-256 content signature before any parsing occurs.
2. Local Cache (.tldrgraph/tldrgraph.db): Stores file paths, last modified timestamps, AST symbol signatures, and enrichment summaries in an SQLite database.
3. Dirty Detection:
   • If a file's hash matches the stored signature, its existing AST nodes, intents, and embeddings are retained without modification.
   • If the hash changed, only the symbols within that specific file are marked dirty and scheduled for incremental update.

Benefits¶

• Token Cost Savings: Re-running tldrgraph init on a 50,000-line codebase where only one file was modified spends 0 tokens on the 49,900 unchanged lines.
• Instant CI Scans: Pre-commit hooks and CI runs finish in milliseconds when checking cached snapshots.
• Persisted Summaries: Human-approved LLM enrichments are never accidentally overwritten or lost during routine scans.