Validation Status and Reproduction Requirements
The component and zone thresholds in this beta skill are documented heuristics. This repository does not currently include a pinned historical dataset or a walk-forward harness capable of reproducing quantitative return, win-rate, or drawdown claims. Accordingly, this skill makes no such claims and its zone labels must not be treated as allocation or performance rules.
What is verified in-repository
- Pure component calculations and boundary behavior are covered by unit tests.
- Offline bull, bear, degraded-data, and sparse-data snapshots exercise the
same
run_analysis()entry point used by live mode. - Missing inputs fail closed: a composite classification requires at least four components representing at least 65% of original model weight.
- Report and package drift are checked by repository validation gates.
These tests establish deterministic software behavior. They do not establish predictive power or investment performance.
Required artifacts for historical validation
A future validation change must include all of the following in the repository:
- A runnable harness that imports the shipped calculation code.
- A data manifest with source URLs, retrieval timestamps, hashes, symbol map, delisting policy, and point-in-time universe construction.
- A pinned or deterministically retrievable input snapshot that permits an offline rerun without silently substituting current constituents.
- Explicit no-lookahead rules, missing-data treatment, transaction-cost assumptions, and non-overlapping or dependence-aware statistical summaries.
- Machine-checked expected aggregates and documented commands to reproduce them.
Until those artifacts exist, RISK_ON, NEUTRAL, and RISK_OFF are descriptive
heuristic bands only. Any historical experiment performed outside this
repository is exploratory and must not be presented as verified evidence here.