Data Quality Beats Scale Past a Point — hypothesis
Supplied working report; original verdicts not validated.
Review status: Needs review. This is an archival source note, not a passed proof.
hypothesis.md
# Data Quality Beats Scale Past a Point — hypothesis **Source:** Incoprea Theory Volume VIII, Theory 067 / Information & Computation. **Claim (verbatim):** "Beyond a threshold in dataset size, further improvement comes from data quality rather than quantity, so a smaller curated dataset should outperform a larger noisy one at equal compute." **Predictions (verbatim):** - P1 — Curated subsets should match larger noisy sets. - P2 — Gains from volume should saturate. - P3 — Quality metrics should predict performance past the threshold. **Refutation (verbatim):** Continued gains from volume at controlled quality would leave the scaling law intact. **Pre-registered test.** Generate a synthetic classification task with a known signal, then train a logistic classifier on (a) a small clean set and (b) a large set with a swept label-noise rate. Compare held-out accuracy, sweep the noise rate to find where the small clean set matches or beats the large noisy one, and separately sweep volume at a fixed noise rate to show saturation. **Fixed before running:** data-generating model and test set fixed; classifier and training procedure fixed.