The bounded reading becomes a one-sided equivalence test at the SESOI¶

# research(2026-10). A cell read bounded when its upper bound fell below the effect its simulation says it detects at its Holm level: a margin set by the design's size, not by what matters. Equivalence bounds are set by the smallest effect size of interest (Lakens, Social Psychological and Personality Science 8(4), 2017; Lakens, Scheel and Isager, AMPPS 1(2), 2018, which also describes bounds at the smallest effect a study has the power to detect), and this study has a justified SESOI, δ (2026-09-30 and 2026-10-09 entries). So a cell now reads bounded when its upper bound at its Holm level lies below δ (decomposition.cell_verdict, margin): one-sided, since both hypotheses are directional, so a reversed effect still reads bounded. A cell's simulation stays required before a test read, as its power, not its margin; a cell without one is still never read bounded.

Power of the new reading under a true null, from the registered simulations: the share of null studies whose interval lies inside ±δ (null_reads_absent; every such study reads bounded or, if its lower bound clears zero, supported, so the bounded share is at least this share less the false-positive rate) is 1.0 at both Holm levels and every spread target, for OpenStack at K = 10 (partition-sensitivity-openstack-rp1.0-logprob_per_token.json) and Wikimedia at K = 20 (partition-sensitivity-wikimedia-rp1.0-logprob_per_token-k20.json).

Decided after the development pilots were read, and disclosed as such in the Stage 1 design history. On the 22-run likelihood pilots (partition-pilot-<org>-likelihood.json), OpenStack's +0.0231 [+0.0081, +0.0402] at 97.5% is supported under either margin. Wikimedia's +0.0058 [-0.0004, +0.0125] at 97.5% read inconclusive against its registered detectable effect, 0.0071, and would read bounded against δ, 0.0183; at 95%, the level H1 alone is read at, it is supported ([+0.0003, +0.0117]) under either margin. That 97.5% cell is the only development verdict the change moves. The change makes a bounded reading reachable wherever the interval excludes δ, which is its purpose; the development window never stands for the test window. It supersedes the second sentence of item 6 of the H2 plan above: a cell whose detectable effect is δ or more now reads bounded when its interval excludes δ, which an equivalence test permits whatever its power. The Stage 1 design history states the change and its timing (papers PR #88, org-house-style/sections/30-rq.tex).