03 / IN PRACTICE
What does the evidence actually support?
Explore two separate explanations: a fictional comparison you can inspect, and a guided account of a narrow connected local pilot. Neither is a live system.
01 / SECTION
Compare before deciding.
Synthetic teaching example. No real participants, model calls, customer data, or deployment. This browser walkthrough is not the Python pilot.
| Measure | Baseline | Fast candidate | Evidence-first candidate | Illustrative criterion |
|---|---|---|---|---|
| Draft quality | 80% | 90% | 90% | At least 85%, and at least 5 percentage points above baseline |
| Reviewer defect detection | 85% | 70% | 90% | At least 80%, with no decline from baseline |
| Review time per case | 5 minutes | 3 minutes | 4.5 minutes | At most 6 minutes, with no increase from baseline |
The fixture declares 20 cases per measurement. It has no sampled replies, participants, statistical uncertainty calculation, or real study. These teaching values are not recommended targets. Read source ↗
Choose a comparison
EVALUATION / BASELINE
The reference point
01
Draft quality
Baseline reference · 80%
02
Reviewer defect detection
Baseline reference · 85%
03
Review time per case
Baseline reference · 5 minutes
The baseline is a reference, not an improved candidate.
02 / SECTION
A connected family, with separate boundaries.
Browser explanation of a local fixture. This is not a live integration or the repository’s original read-only report. The source report uses pinned Design components as disabled historical views.
STEP 01 / 07
Evidence prepared
Evolution supplies supported synthetic evidence from the fixture. This does not establish real-world benefit.
Management decision ≠ publication evidence ≠ local activation ≠ real-world benefit
Verified manifest publication does not deploy a model or verify activation. Coordinated family recovery is outside the demonstrated scope. Read source ↗