Assumptions & Invariants Ledger

Assumptions are things the project relies on that could be false. If violated, specific parts break. ### A-INS-001: DELL provides z-trajectory **Assumption**: The DELL latent engine streams z(t) embeddings at consistent intervals. **What breaks if false**: Claim detectors have no input, inscriptions cannot be generated. **Detection**: Monitor DELL stream health, alert on gaps > 10s. **Mitigation**: Buffer last-known z(t) and emit degraded-confidence claims. ### A-INS-002: Graph Kernel enforces admissibility **Assumption**: All evidence retrieval for Echo claims goes through the Graph Kernel. **What breaks if false**: Echo claims could reference non-admissible evidence, provenance is broken. **Detection**: Audit Echo claims for valid slice_id references. **Mitigation**: Reject Echo claims without verified admissibility tokens. ### A-INS-003: Basins are rare relative to samples **Assumption**: The number of graduated basins is << number of z-trajectory samples. **What breaks if false**: Lexicon becomes unmanageably large, phrase compression fails. **Detection**: Monitor basin count, alert if > 100 basins per user per month. **Mitigation**: Increase graduation threshold, retire low-dwell basins. ### A-INS-004: N'Ko sigils are Unicode-stable **Assumption**: The 10 N'Ko characters (ߛ, ߜ, ߕ, ߙ, ߡ, ߚ, ߞ, ߣ, ߠ, ߥ) have stable Unicode codepoints. **What breaks if false**: Rendering breaks across systems, corpus becomes unreadable. **Detection**: N/A (Unicode codepoints are standardized). **Mitigation**: Store codepoints (e.g., U+07DB) alongside rendered characters.