history
Backtest on everything you kept
Full‑resolution history rather than a sample, so a walk‑forward test runs over what actually happened at every bar instead of a downsample someone chose years ago.
A forecast you cannot grade six months later is an opinion, not research. Oliver holds the prediction, the numbers behind it and what actually happened in one engine — so skill is something you measure rather than something you assert.
Every forecast is written down the moment it is made. Time passes, the market does what it does, and each call resolves against what was actually observed — not against a backfilled story about it.
Full‑resolution history rather than a sample, so a walk‑forward test runs over what actually happened at every bar instead of a downsample someone chose years ago.
Sweeping every symbol and every horizon is a single read rather than one query per name — which is the difference between a nightly job and an idea you can try before lunch.
What you said and what happened live side by side, so settling a call is a join. No export, no second system, no reconciling two versions of the past.
Horizons in trading time, outcomes taken from observed highs and lows, and a year of chart costing about what a day does.
Store what the model actually said alongside whatever calibration you applied on top. When a day looks wrong, you can tell a broken thermostat from a broken model — and keep the one that was right.
Every key carries its own scope, so an analyst or an agent can explore the record without being handed the keys to all of it.
Five different time‑series forecasting engines have been swapped into the same harness without changing what sits underneath them. The interesting question is never which model you ran — it is whether you can still tell, a quarter later, which one was actually better.
Every symbol, every limit, every horizon — fused into one pass over the data rather than one query at a time.
Small models query the engine directly and return grounded findings a stronger one judges — competing hypotheses, on the record.