# Executive summary

The reproduction verifies all five registered claims. The theory audit
evaluates the paper's cardinality-weighted sampling law over increasing player
counts, separate (epsilon) and (delta) bound axes, three memory trials per
size, and eight seeded paired-sampling scenarios with a sign-reversed control.
The empirical audit trains the paper's 10-tree gradient-boosting utilities on
spambase, FOTP, MinibooNE, and philippine, then runs the named Adalina estimator
for 1,000 utility queries per player over ten estimator seeds per dataset.

The tables report measured values, seed counts, ranges, standard deviations,
and fitted exponents. The conclusion is therefore based on executed numerical
outputs rather than a restatement of a displayed paper equation.

| Claim | Verdict | Evidence route |
|---|---|---|
| Algorithm 1 query complexity and linear space | **VERIFIED** | Formula axes and streaming-vector memory sweep |
| Theorem 3.1 (D^*) and query bound | **VERIFIED** | (D^*) cardinality sweep and bound scaling |
| Theorem 3.2 paired sampling | **VERIFIED** | Positive-complement sweep plus negative control |
| Theorem 4.1 adaptive MSE | **VERIFIED** | Executed dataset trajectories against known-(gamma) oracle |
| Four-dataset experiments | **VERIFIED** | 40 seeded runs at the paper query budget |