AmberContinuumAI collaborative research lab

Experimental Programs

Let the method push back.

A study earns trust by saying what was frozen, what was measured, and what happened when a prediction missed. The record stays open to correction.

Studies with public records

Under test and on record.

PDOT

Registered · pilots documented

Can changes in a human–AI conversation be measured when the conditions shift?

What I watch
Operator anchoring (λ: how human intent carries into AI behavior), coherence (κ: whether shared context stays consistent), procedural autonomy (θ: whether human choice remains), drift (ε: movement away from intent), recovery after a change, and CCS = R − τ (resonance minus a registered threshold).
Evidence and limits
If these measures cannot be scored reliably or do not behave as registered when conditions change, the proposed interpretation needs revision. The registered plan governs that decision; no confirmatory outcome is claimed here.
Public record
OSF registration · 17 May 2026
Open the record ↗Methods and study status ↗

QRT Bell Test

Recorded outcomes

Do filed quantum hardware predictions survive the tests written for them?

What I watch
Bell fidelity, then corrected Ramsey and replication endpoints on IBM hardware.
Evidence and limits
The first two Bell fidelity tests were negative under their filed rules. A later verdict was withdrawn; later results are kept as separate findings.
Public record
IBM Quantum job IDs · Test #1: d93t71fu62ks7395a110 · Test #2: d93tv0mvtlqs73ftkh80. Preregistered protocols, outcomes, and corrections remain public.
Open the record ↗Test #1 outcome and job ID ↗Test #2 hardware results and job ID ↗

A result is a record, not a sticker

Read the miss beside the claim.

PDOT · registered

The method is public; the outcome is not claimed.

The OSF registration freezes the question and analysis plan. Pilots are documented in the public study status. Confirmatory progress and results are left unstated here until there is an appropriate public record.

Open the registration ↗
QRT · recorded negative

Bell fidelity did not hold.

Filed tests #1 and #2 returned negative decisions. The repository also keeps a withdrawn verdict, corrections, later measurements, and a second-device replication as distinct entries.

Follow the correction ↗

An applied contribution

From method to dataset.

The crystalline polymer work turns a research question into engineered features and a public dataset for others to inspect and use.

Crystalline Polymer Dataset

NeurIPS 2025 competition entry

Can descriptors derived from polymer structures help examine five material property targets?

What I watch
SMILES-derived features including crystalline propensity, dipole mobility, radial mass, and entanglement proxies. This method produced a public dataset others can inspect and test.
Evidence and limits
The contribution is the published dataset and its feature design. Any claim of predictive improvement needs a reproducible comparison on the target properties.
Public record
Entry for NeurIPS – Open Polymer Prediction 2025. Published on Kaggle as “Crystalline Polymers and Quantum Stability” for five polymer property targets. The features are structure-derived, physics-inspired proxies.
Open the record ↗Competition ↗

An interactive instrument

Notice the prompt’s effect.

Operator Assessment is a live exercise in noticing how a question applies pressure before answering it.

Operator Assessment v.1

Interactive assessment · exploratory

Can a person notice how a question shapes their response before they answer it?

What I watch
A live, 12-question pressure test with a post-assessment debrief focused on noticing the prompt’s effect.
Evidence and limits
This is an exploratory instrument, not a validated diagnosis. Its score should not be treated as a clinical or empirical finding.
Public record
Public interactive assessment and debrief
Open the record ↗

Exploratory models

Follow each transformation.

These models expose their inputs and transformations. Interpretation remains separate from physical evidence.

Interactive model

Model Observatory

Change an input, inspect the equation, and see where a mathematical comparison turns into a symbolic reading.

Enter the Observatory →
Symbolic engine

Zodiac Engine

Follow astronomical input, degree assignments, and interpretive rules into a narrative. The reading is an exploration, not a physical finding.

Open the computational layer →
Visual model

Particle Symphony

Explore a visual account of how a person’s input may change the direction of AI inference.

Open the model ↗

Sources in order

Follow the citation ladder.

  1. PDOT preregistration ↗Question and analysis plan frozen before outcomes
  2. QRT hardware record ↗Negative results, withdrawal, corrections, and later tests
  3. CSA specification ↗The governing method
  4. Human Closure paper ↗Agency and recovery framework
  5. Exploratory theory preprint ↗A proposal, separate from empirical results