This document is the current public evidence boundary for TNView.
E001 is preregistered but has not been canonically run on a real Quimb corpus. The repository currently supports local fixture smoke checks, run-log validation, deterministic diagnostics, comparison commands, and optional duck-typed adapters.
- Current diagnostics are deterministic threshold rules, not a validated scientific failure classifier.
- Fixture logs demonstrate CLI behavior but do not provide precision, recall, false-stop rate, or compute-saved evidence.
- Capacity-limited labels for E001 must come from paired higher-accuracy runs, not subjective inspection.
- A future learned predictor should not be added unless transparent thresholds fail under the preregistered controls.
- Quimb and TeNPy are optional dependencies.
- Adapter tests use fake library-shaped objects unless an optional integration environment installs the real libraries.
- No Quimb or TeNPy source is vendored here.
- TNView records telemetry summaries and does not serialize full tensor-network objects.
- Logger-overhead numbers from
./scripts/reproduce_e001.sh smokeare smoke measurements for the local environment, not canonical benchmark results. - Parser throughput, memory scaling, and logger overhead still need canonical corpus measurements.
- Terminal rendering is optimized for operator inspection, not for preserving every raw event field in the visual output.