What has Michael Darius Eastwood measured?

Unit 83 (DOI 10.17605/OSF.IO/MN73W; results DOI 10.17605/OSF.IO/98HKR): the per-round lead widened by 0.0100 (95.5 per cent interval 0.0092 to 0.0108), above the registered tolerance of 0.0075. One of its two engines cleared the tolerance on its own, the mechanism axis was not evaluable, and one call was retried under a disclosed recovery amendment. Unit 06 (DOI 10.17605/OSF.IO/RV2B5) ran on 6 October 2026; one of its two runs could not start, so it is reported as administratively incomplete, with no verdict.

The early single-model estimate of about 2.24 was retracted when a blinded replication on five frontier models did not reproduce it. Paper II's surviving estimate of Law I's improvement exponent is about 0.49 by regression (standard error 0.20) and 0.59 by endpoint estimate, whose bootstrap interval runs from -1.3 to 2.9: inconclusive. Sequential reasoning beat parallel sampling wherever both exponents could be estimated; Sharma and Chopra (arXiv:2511.02309, 4 November 2025) gave the earlier and more robust demonstration of the sequential advantage at matched compute.

Alignment under blinding: Paper IV.d compares an earlier unblinded run with a later one that added a bundle of blinding steps. Gemini's association with reasoning depth turned from significantly positive to significantly negative, and DeepSeek's became null; several things changed at once. An internal code review in October 2026 found defects, so no single cause is claimed. Paper V: a stakeholder-care instruction raised scored stakeholder care significantly in all five analysable model runs. Paper VII: in 19 of 25 empirical domains, the scaling family predicted from each system's composition, classified before fitting, matched under strict AICc; this is a descriptive tally whose analysis was not frozen in advance, and the six misses are published.

Not yet measured on any real system: Law II's correction and drift exponents, beta and k, and Law III's gamma.