Released entries are append-only. Model changes do not rewrite historical signal versions or outcomes.
TESBOT-STACK/v3.5.0
Released: 24 August 2026
Affected modules: input safety, publication policy, signal lifecycle and public validation.
Summary: Stale, missing, conflicting or insufficient inputs now force WAIT. Every unique paper analysis, including WAIT, is retained with its evidence and lifecycle.
Validation impact: Directional counts may fall. Confidence remains a checklist pass rate, never win probability.
Migration: Forward-only provenance, lifecycle and finalized-snapshot fields; existing paper rows and outcomes retained.
TESBOT-STACK/v3.4.0
Released: 8 August 2026
Affected modules: signal persistence, model registry and evidence ledger.
Summary: Each new paper analysis receives a model stamp and a hash-chained immutable publication record.
Validation impact: No numerical signal change. Pre-release rows remain version-unknown where the original version cannot be proven.
Migration: Forward-only ledger and model-version columns; existing paper rows retained.
TESBOT-STACK/v3.3.0
Released: 8 August 2026
Affected modules: SARVAM-369 final decision layer.
Summary: Added a master evidence blend that may veto an existing direction to WAIT but cannot invent a new direction.
Validation impact: Directional output can only decrease relative to v3.2; comparisons across versions must be separated.
TESBOT-STACK/v3.2.0
Released: 22 July 2026
Affected modules: core directional checklist.
Summary: Volume and swing checks became direction-specific, removing points that could previously align with both LONG and SHORT.
Validation impact: Checklist confidence is generally lower than earlier versions; old and new scores are not directly comparable.
