ĀML™ is being tested on real websites.
These are project-authored field experiments with preserved pre-change checkpoints, explicit UI targets, provenance, declared policy inputs, and before/after evidence as each experiment is completed.
Timberline Greens mobile shortcuts
The live Timberline Greens mobile shop bar now uses the pinned ĀML browser compiler to suppress a redundant current-page Samples or Build kit shortcut. Phone ordering remains available. If compilation fails, all links appear. Four route scenarios and two failure paths passed deterministic checks.
This is project-authored engineering evidence using declared design scores. No visitor attention, accessibility improvement or sales lift was measured. The report includes source provenance, before/after checkpoints, decisions and limits.
ĀRU site compiler integration
The /aml policy lab now compiles an ĀML transmission with the browser engine taken from aml-core, rather than evaluating a separate React copy of the rule. The table below executes the same adapter used by the live lab. Values are declared scenario inputs.
| Declared scenario | Attention | Restoration | Compiler decision |
|---|---|---|---|
| Deep article | 3.2 | 9.1 | ALLOW |
| Focus reminder | 2.8 | 7.4 | ALLOW |
| Infinite scroll | 8.5 | 3.0 | SUPPRESS |
| Rage bait | 7.5 | 2.0 | SUPPRESS |
| Equal boundary | 4.0 | 4.0 | ALLOW |
Provenance: aml-core/docs/aml-browser.js blob e4e64a00d20718c1b74d8e5dd9f9dd08b1b84865. The policy is restoration_value ≥ attention_cost. This integration verifies compiler behavior for five fixtures; it does not measure visitor experience, sales, health, or ethical outcomes. Read the report for build checks and limitations.
Security website field pilot
The current study examines repeated, high-salience interface elements across navigation, hero areas, page sections and fixed mobile controls. Contact information remains site-specific and is not normalized.
Pre-pilot checkpoints
| Property | Checkpoint | Source commit |
|---|---|---|
| Security First Alarm Portland | 6aa5e639d7fa72652bd096b7 | 6b903756b6c8fb44ef9fa7845b68235b62193d1b |
| PDX Fire Alarms | 6aa5e63ff2d1143d0dd5e05b | 671c0b40c9736fd909d80f2872686a2ca4204bc2 |
| PDX Surveillance | 6aa5e6468e445d8df8a55c80 | cdf1a44d889190b62c6656b77f776d4f172ecd39 |
| Portland Home Security | 6aa5e64b3affece0c34c969e | c3612b1b735871d192ecf0dc246b56a0a9039f91 |
| Vancouver Security | 6aa5e652f1d3871b3d0b749c | 26b4fb00d7d011f4dbe242358addc585709e1b22 |
| Security First Alarm | 6aa5e658bf81b5ee1d7731dc | c7e28764878186a5780de23c73438e2dde39d60d |
| Security First Alarm Sales | 6aa5e65d726855933055877f | 7c2775d50ca20964ba5ce9d508e6490f0b538ac2 |
| Security First Alarm Service | 6aa5e664d0e21c24894c2acf | 4e3e84a26a4819bd88b769e7852cf3babb429908 |
Evidence boundary
These pilots are project-authored field evidence. They are not independent reproduction, third-party certification, scientific validation, or proof of commercial lift. Attention and restoration values are declared or modeled inputs, not objective measurements.
render_allowed = restoration_value >= attention_cost