# Pre-committed results-narrative branches
Rule: select branch by the SIGN of the intercept (dose-5 reference) coefficient
'const' in the locked v2 delta model on FINAL data:
  const >= -0.05            -> Branch A (Bayes-consistent / over-conformity)
  const <  -0.05            -> Branch B (under-use of context)
Only [RESULTS SENTENCE PLACEHOLDER], the abstract results clause, and Section 7's
opening paragraph take the branch text. Nothing else differs between branches.
Any deviation from this rule must be reported in the deviations section.

## Branch A text (Delta >= -0.05)
RESULTS SENTENCE: "Contrary to laboratory predictions, autonomous reviewers'
weight on the visible consensus was statistically indistinguishable from (or
above) the Bayesian optimum: these reviewers do not herd beyond information -
and the institution's exposure comes from the architecture alone."
ABSTRACT CLAUSE: "the estimated conformity gap is [X]; AI reviewers in situ use
shown peer evidence [as an optimal Bayesian would / more aggressively than
optimal], so the measured risk to aggregation is architectural rather than
behavioral."

## Branch B text (Delta < -0.05)
RESULTS SENTENCE: "Against the laboratory literature's prediction, autonomous
reviewers systematically UNDER-used the visible consensus: effective weight on
peer opinion sat far below the Bayesian optimum, i.e., in production they
behaved nearly independently - leaving calibration gains uncollected while the
institution remains exposed through shared-context architecture alone."
ABSTRACT CLAUSE: "the estimated conformity gap is [X]: AI reviewers in situ
under-weight shown peer evidence relative to any optimal updater, so laboratory
conformity fails to transfer to incentivized production review."

## Deviations section text (to include regardless of branch)
"All analyses followed the registered pipeline. Post-registration additions,
all labelled descriptive/exploratory: (i) median and trimmed conformity-gap
descriptives alongside means, motivated by the non-Gaussian shape of w_eff
under zero-dispersion contexts; (ii) the no-context anchor subsection added at
design time after observing empty contexts occur operationally; (iii) this
branch-selection rule, committed before final data examination."