Validation

Methodology and holdback evidence

Enterprise trust page for Ether Data Lift and Woodman's holdback results. This page summarizes the metrics used in the homepage proof block.

Ether Data Lift

M0→M15 adds +0.1525 R² above calendar baseline.

Study context: NYC 911 patrol calls, n=492,316, balanced primary analysis.

M0 baseline R²

$R^2_{M0}=0.025$

Calendar baseline

M1 increment

$\Delta R^2_{M1}=+0.1223$

Largest single lift component

Total model R²

$R^2_{M15}=0.1775$

M0→M15 full stack

Ether Data Lift

$\Delta R^2=+0.1525$

Above baseline

Woodman's Holdback

$R^2_{raw}=+0.80,\ R^2_{log}=+0.92$

Structural priors predict hourly restaurant volume across 126 operating days, without transaction history in training.

Privacy stance: zero individual-level data, zero device IDs, administrative and structural inputs only.

Study Setup

Dependent variable: NYC 911 patrol calls.
Sample size: n=492,316 observations.
Comparison path: M0 calendar baseline → M15 full stack.
Evaluation: balanced primary analysis.
Output target: lift above baseline, not vanity fit.

Woodman's Holdout Detail

Horizon

R²(raw)

R²(log)

Status

3-day holdout

+0.80

+0.92

Stable

7-day holdout

+0.80

+0.92

Primary reference

14-day holdout

+0.80

+0.92

Stable

Privacy and Data Discipline

All validation uses administrative counts and structural signals. No individual-level data, no MAIDs, no transaction identities. Inputs are government-sourced or aggregate behavioral context suitable for structural inference.

Methodology and holdback evidence

M0→M15 adds +0.1525 R² above calendar baseline.

Rraw2=+0.80, Rlog2=+0.92R^2_{raw}=+0.80,\ R^2_{log}=+0.92Rraw2​=+0.80, Rlog2​=+0.92

$R^2_{raw}=+0.80,\ R^2_{log}=+0.92$