QUESTION: Which treatment raises the synthetic response after blocking?
KERNEL THEOREMS: balance of all shifts; counts; coverage; additive integer identity.
ASSUMPTIONS: additive effects; iid Gaussian noise; fixed nuisance levels.
PROVIDER: deterministic-residual-rule/v1 (no AI inference)

ROUND 1 · NUMERICAL CALCULATION FROM 16 SYNTHETIC OBSERVATIONS
Residual RMS: 1.172399; residual df: 6.
Additive assumption: challenged.
T1 − T0: 1.019578; conditional 95% CI [-2.292977, 4.332134].
T2 − T0: 1.959773; conditional 95% CI [-1.352782, 5.272328].
T3 − T0: 4.851390; conditional 95% CI [1.538835, 8.163945].
conditional on additive model and iid Gaussian noise; model challenged: intervals are diagnostic only

REVISED QUESTION · UNVERIFIED PROPOSAL
Is the treatment-3 contrast row-dependent under complete column coverage?
Reason: Observed residual RMS 1.172 exceeds 0.900.
Validated follow-up shifts: 1, 2, 3

FOLLOW-UP · 48 NEW SYNTHETIC OBSERVATIONS
Combined N: 64; residual RMS: 1.445516.
T3 − T0 by row, with all four columns represented in each mean:
  Row 0: 3.208810
  Row 1: 3.286240
  Row 2: 2.910443
  Row 3: 10.768962

NEXT QUESTION · UNVERIFIED PROPOSAL
Which row conditions modify the treatment-3 effect, and does that pattern replicate with independent blocks? This is an unverified proposal.

SYNTHETIC EVIDENCE ONLY: neither efficacy nor model assumptions are proved.
PROVENANCE: see manifest.json, observations.json, summary.json and proof-audit.txt.
Observation SHA-256: 832c18e1e5e8f976b6dee72537b102d28ff7895d22dd73a70b2aadb06e790a83