This is the full journal of our forecasts — each one published BEFORE the market showed the result, and none deleted. Green means it was right, red means it was not. You can download the whole history and check any row yourself.
Outcome = bot call (LO/SH/FL/OFF). Fact 3h / Fact 24h = what the market did. Fact color = grade: green matches Outcome, red misses, muted when OFF (no directional call). OFF despite agreeing models = weak evidence (long needs avg ≥+5).
Accuracy = (correct + 0.5×partial) ÷ graded. 24h: {ok}+{part}+{wrong} = {graded} of {total} (no fact: {pending}). 3h: {ok3}+{part3}+{wrong3} = {graded3} (no fact: {pending3}).