AcroSketch Statistics

Production intelligence

Quality and learning statistics

Updated Oct 23, 2026, 8:17 AM
Accuracy is measured through production decisions

The primary quality measure is production acceptance—not the model's own confidence. A true geometric precision score will require labeled expected layouts for direct piece-by-piece comparison.

Approved references 39 11 homepage approval records · 2 approved without placement
Needs production review 2 Oldest is 81 days
AI plan acceptance 81.8% 9 approved of 11 reviewed
First-pass interpretation 70.7% 29 approved without interpretation edits

Reference pipeline

From synced ticket to trusted example

14.4% of tickets are approved references
1 270 Synced tickets
2 43 Interpreted
3 2 Needs review
4 39 Retrievable references
Interpretation coverage 15.9%
Verified AcroSketch mapping 82.6%
Average approval time 2.8 days From stored plan to approval

Planning quality

AI-generated plan decisions

11 reviewed generations
9 Approved
2 Needs changes
Production acceptance 81.8%
Passed stock-fit validation 100%

Only reviewed OpenAI-generated layouts are counted. Built-in plans and exact approved references are excluded.

Interpretation quality

Ticket review outcomes

41 approved interpretations
Approved without interpretation correction 70.7%
Required interpretation correction 31%
AI-reported extraction confidence 85.3% Diagnostic only—not a measured accuracy score.

Six-month activity

Approvals and planning decisions

Ticket references AI plan approved Needs changes

Production feedback

Why AI plans need changes

  • Production ready2 decisions
    100%

Model monitoring

Acceptance by model

ModelReviewedAccepted
gpt-5.6-sol 11 81.8%

Metric definitions

How to read this page

AI plan acceptance

Approved AI generations divided by all AI generations reviewed by production.

First-pass interpretation

Approved ticket interpretations with no recorded material, dimension, or note correction.

Approved reference

A retrievable ticket plan with verified positioned pieces. Interpretation alone is insufficient.

Accuracy limitation

Acceptance measures usefulness. Exact precision requires an expected layout and geometric comparison.