The adapter mapped available fields into 6 tests. Results flow into each debate’s sub-thesis verdicts, the Power map, and investment synthesis; blocked questions become specific experiment requirements.
97% of selected cases link decision, action, outcome, and resolved grade.
Interpretation: The selected dataset is sufficiently complete to analyze the operating feedback loop.
Boundary: Coverage establishes that experience is captured; it does not independently validate grading quality or prove that the system learns from it.
Next: Independently regrade a stratified sample and audit missing links.
Validation selected the base-rate model; richer historical models produced no validated out-of-sample gain.
Interpretation: The selected cases do not clear the pre-set 0.005 Brier materiality threshold versus the simple base-rate benchmark.
Boundary: This is a time-safe predictive test, not proof that deployed learning caused better actions or economic outcomes.
Next: Run a versioned or randomized action-policy comparison on a new batch.
No material pooling advantage Local minus pooled Brier: 0.003.
Interpretation: 28/59 patterns cross segments; 0/5 evaluated segments show material negative transfer. The positive pooled-learning sub-thesis fails the pre-set 0.005 Brier materiality threshold in this dataset.
Boundary: Groups are customer segments, not identified customers. This tests transfer of grade prediction; improving chosen actions requires an action-policy experiment.
Next: Repeat with verified customer identities and compare pooled versus local action policies on held-out customers.
Expanding from the earliest cohort to full history changed Brier by -0.000; the curve reverses direction.
Interpretation: The positive marginal-value sub-thesis fails: the gain is below 0.005 Brier or reverses as history grows.
Boundary: The result is estimator- and time-window-specific and does not by itself establish causal policy improvement.
Next: Repeat on a fresh time window and isolate new patterns from changing case mix.
25% of development history reaches within 0.005 Brier of the full-history estimator.
Interpretation: A relatively small historical subset reproduces nearly all measured predictive performance, which weakens a raw-history durability claim.
Boundary: This is an internal history-ablation proxy, not a competitor-access, public-data, or calibration-budget challenger test.
Next: Run a budgeted challenger using only public features, compressed rules, synthetic cases, and limited calibration.
291/300 selected cases include finite economic outcome values.
Interpretation: The dataset can describe stakes and grade-associated outcomes, but cannot yet identify incremental value caused by a better policy.
Boundary: Units must be consistent, and observed association is not causal incremental value.
Next: Define value units and compare matched or randomized policy outcomes net of operating cost.
Case rows cannot establish contractual retention, training, or cross-customer reuse rights.
Interpretation: This question correctly routes to contracts, governance, and operational-access evidence.
Boundary: Inferring legal permission from technical availability would be invalid.
Next: Review contracts and data governance against the intended learning uses.
Selected cases alone cannot establish switching costs, scale economies, brand, or counter-positioning.
Interpretation: Case data may quantify product performance, while durable barriers require customer, cost, workflow, and competitive evidence.
Boundary: Do not infer a Helmer Power solely from case volume or predictive accuracy.
Next: Join workflow adoption, replacement, pricing, cost, and competitor evidence to the analysis.