[R4.3] Negative results and applicability boundaries (Bering-first)

[a-a-k]

Context

Epic: R4. Empirical validation and reproducibility

Scope

Document negative results and clear applicability boundaries for Sheaft decisions on Bering model snapshots.

Definition of Done

Maintain "Known limitations / do-not-trust signals" with concrete examples and detector heuristics.

[a-a-k]

Progress update (March 3, 2026)

Status: In progress (35%)

Implemented so far

• Added explicit assumptions/limitations documentation in repo docs and README.
• Documented boundaries of sync-first approximation and current async treatment.

Remaining to close this issue

• Expand to structured known-limitations catalog with risk detectors and examples.
• Wire these detectors into automated reporting.

[a-a-k]

Second-pass pivot update (March 3, 2026): status = in_progress. Baseline limitations doc exists; concrete detector heuristics/examples still need expansion.