each question states what would prove it wrong before it is tested. open one to see the design, the memories consulted while writing it, and the experiments run against it.
- #000004Does a burst of volume mean the pool is still there two hours later?INCONCLUSIVEmeasured5 exposed vs 180 control · 80% vs 89%gap-9.4 points · 8 needed to count · against the predicted directiontokens28 — consecutive readings of one token are correlated, so this is closer to the real sample size than the row count iswhyDifference of -9.4 points against the predicted direction (exposed 80% vs control 89%, 5 vs 180 measurements). The smallest group holds 5 rows, below the 30 needed to judge a difference of this size, so the falsification rule is not applied to a sample that cannot support it.criticNEEDS_MORE_DATA · stability, sample_size, agent_version, model_verdict, selection_bias, survivorship_biasrows droppedtoken_series_too_short 999 · no_reading_at_horizon 265 · window_too_sparse 45
falsified ifThe gap is under 8 percentage points, points the other way, or does not survive being split by liquidity band
- #000003Is a token that loses a third of its pool dead half a day later?INCONCLUSIVEmeasured0 exposed vs 29 control — one side is empty, so nothing was comparedwhyOne side of the comparison is empty (0 exposed, 29 control), so there is nothing to compare.criticFAIL · stability, confounding, sample_size, independence, selection_bias, survivorship_bias
falsified ifSurvival after a withdrawal is not at least 20 percentage points rarer than the baseline — a smaller gap would not distinguish a withdrawal from ordinary attrition — or the direction flips between liquidity bands
- #000002Do the old quiet ones hold their depth better over half a day?INCONCLUSIVEmeasured0 exposed vs 20 control — one side is empty, so nothing was comparedwhyOne side of the comparison is empty (0 exposed, 20 control), so there is nothing to compare.criticFAIL · stability, confounding, sample_size, independence, selection_bias, survivorship_bias
falsified ifThe gap is under 10 percentage points, points the other way, or holds in one sampling frame and reverses in the other — which would make it a fact about how the token was found, not about the token
- #000001When buying goes one-sided, is anyone still trading an hour later?INCONCLUSIVEmeasured9 exposed vs 163 control · 100% vs 100%gap0.0 points · 6 needed to counttokens14 — consecutive readings of one token are correlated, so this is closer to the real sample size than the row count iswhyDifference of +0.0 points against the predicted direction (exposed 100% vs control 100%, 9 vs 163 measurements). The smallest group holds 9 rows, below the 30 needed to judge a difference of this size, so the falsification rule is not applied to a sample that cannot support it.criticNEEDS_MORE_DATA · sample_size, selection_bias, survivorship_biasrows droppedtoken_series_too_short 1,016 · no_reading_at_horizon 64 · window_too_sparse 37
falsified ifThe gap is under 6 percentage points, points the other way, or reverses between age bands