Diagnostics for the 12-month walk-forward redeployment simulations
(events_desk/backtest/walkforward.py). A walk-forward replays the past year week by
week with an honest bankroll: cash is finite, positions lock capital until they resolve, and
every model re-evaluation must be justified by a trigger. Treatments are separate runs of the
same harness with different estimate engines.
Weekly marks. Bars count markets evaluated (a trigger paid for a fresh model estimate) and wagers taken (we hold to resolution — nothing is sold early). Hover a week for its evaluations and wagers — click a market name in the card to open it in the Market Microscope below. Dashed baseline = starting bankroll; the strip below is drawdown from peak. Stub-engine runs are plumbing tests, not results. Brier note: a diagnostic only, never a success gate.
Price is the market's daily YES probability; the stepped line is our estimate, which only moves when a trigger paid for a re-evaluation (tick marks along the bottom, colored by trigger). The shaded band is the period we held a position. News-volume bars will appear here once the scout data layer (uncapped GKG article counts) ships.