The feedback controller was not developed from a single long experiment. Each module was first studied through a focused experiment that answered one specific question. The table below is the public research map; it describes the purpose and evidence without releasing the full laboratory recipe.
| Experiment | Question | Measured outcome | Design consequence |
|---|---|---|---|
| Glass-slip optical perturbation | Does a local optical-path disturbance appear clearly in the interferometric channel? | Abrupt OPL response and altered interference condition | Treat discontinuities as quality/recalibration events, not ordinary environmental drift. |
| OPL-environment co-variation | Which environmental quantities track OPL change? | Temperature, pressure and humidity trends can co-vary with OPL | Use environmental telemetry as model predictors and diagnostic context. |
| Four-region PT104 mapping | Is enclosure temperature spatially uniform? | Different channels can follow different thermal histories | Warm-up and model validation must consider gradients, not one sensor only. |
| TEC setpoint/step response | How fast does the thermal plant affect measured optical state? | Delayed, distributed response rather than instantaneous centroid correction | TEC is the coarse actuator; cadence and settling time matter. |
| AO direction and return tests | How does AO command translate into detector motion? | Directional response, cross-coupling and finite travel must be calibrated | Use bounded matrix-based fine correction and unloading logic. |
| MaxIm DL centroid validation | Are image-plane errors repeatable and sign-correct? | Reference-relative $dX$, $dY$ and radial error can be measured frame by frame | Validate reference and sign before allowing centroid data to drive AO. |
| Long-run hybrid feedback | Can TEC and AO recover from disturbed operation together? | Recovery can occur, but residence near zero depends on disturbance history, reference quality and AO range | Evaluate threshold residence, bias, recovery time and range use rather than a single final point. |
Each test reduces a different uncertainty:
\[\text{disturbance} \rightarrow \text{measurement validity} \rightarrow \text{model validity} \rightarrow \text{safe correction}.\]For example, it would be unsafe to tune AO from a centroid signal before validating its direction, or to tune the TEC from a single sensor before checking whether the enclosure has a thermal gradient. The component studies establish the assumptions needed by the hybrid controller.
The project does not publish exact experimental geometry, physical component locations, calibration constants, control thresholds, hardware communications, long-duration raw data, or full automation scripts. The public record communicates the scientific reasoning and validation sequence while retaining the operational implementation as private research material.