cording.ai / TightLoop
Notes on ideas, implementation, and validation.
A compact notebook for cording.ai work around TightLoop: residual-error control, neuromorphic correction, forecast-to-action systems, implementation notes, experiments, and validation records.
Residual error
What remains after normal controllers, filters, and forecasting models have already done their work.
Existing systems first
TightLoop is framed as an adaptive layer beside current controllers and models, not as a replacement for them.
Forecast to action
The practical question is not only what the model predicts, but what an operator should do with uncertainty, drift, and shortfall risk.
Writing log
Public notes
This site will collect selected writing derived from cording.ai pages, TightLoop research notes, implementation work, experiments, and validation results.
Google’s Negative Free Cash Flow: The War Tax of America’s AI Religious Civil War
Alphabet’s first-ever quarter of negative free cash flow was not merely an engineering bill. It was a war tax imposed by America’s AI religious civil war.
Premature Ejaculation Is Not Always a Fixed Trait
A hypothesis for how sleep, pain, stress, exercise, and recovery may influence day-to-day variability in ejaculatory control through autonomic regulatory reserve.
A wind gearbox warning is valuable if it creates a service window
Wind gearbox failures are not only prediction problems. They are scheduling problems. A warning is useful if it creates enough time to plan service, parts, access, and downtime.
VR horizon stabilization as a replay problem
VR horizon drift is one of the easiest residual errors to see. The baseline camera may track the target, but the horizon still wobbles enough to create discomfort or make the op...
A turret sightline replay under shock and platform movement
The turret sightline replay is a digital-twin-style visualization of a common control problem: the platform moves, the controller compensates, but the line of sight still carrie...
A NASA turbofan warning benchmark for Sentinel
NASA C-MAPSS turbofan data is a useful benchmark because it separates numerical life prediction from operational warning policy.
A surgical-style tremor correction demo
The surgical-style demo is about a narrow problem: the tool still moves more than the operator intended. That can come from tremor, delay, overcorrection, or command jitter.
Sterile filling early warning as a maintenance-window problem
In a vial filling line, a weak problem can become expensive quickly. At high line speed, minutes matter because thousands of units may be exposed before an operator sees the iss...
A solid-state battery dendrite-risk digital twin
The solid-state battery dendrite demo is different from the motion-control demos, but the structure is similar. A fixed formation control policy leaves residual risk signals. Ti...
SLNN foundation: online residual learning
SLNN is the early control idea behind much of the TightLoop work: a spiking-liquid-style neuromorphic system built around timing, online adaptation, and residual error.
A ROS2 robotic gripper digital twin for slip and overgrip
The ROS2 gripper digital twin exists to show a tradeoff that is easy to miss in a single metric. Reducing slip is useful only if the controller does not solve it by crushing the...
Residual-error assist for surgical-style teleoperation
Robotic surgery and teleoperation are not good places to promise a replacement controller. The safety envelope, validation burden, and regulatory path are too serious. A more re...
Robot grip control is a slip versus damage tradeoff
Contact control is hard because the obvious solution to slip is often more force, and more force can damage the object. A useful gripper controller has to reduce slip without si...
RIFT and feedback time
RIFT is a feedback-time theory rather than a product module. It tries to reason about systems where event timing, attention weight, relation-driven meaning, and prediction diffi...
A neuromorphic warning layer tested on turbofans, bearings, and wind gearboxes
Predictive maintenance is often presented as a prediction problem. Lower RMSE, better RUL curves, cleaner anomaly scores. Those numbers matter, but they are not the whole produc...
A precision-stage replay for wafer-alignment-style residuals
Precision-stage errors are small, but they are not abstract. They can become alignment loss, measurement noise, or process-window violations.
Precision optics as a residual-error problem
Inspection, metrology, precision stages, and optical stabilization systems turn small motion into visible value loss. A few micrometers or a tiny angular residual can affect ali...
nmFUSION and fusion-edge instability delay
nmFUSION explored a neuromorphic control idea inside a BOUT++ fusion-edge instability setting. The reported result was not a full reactor-control product. It was an experiment w...
nmCRYPT and software-defined neuromorphic entropy
nmCRYPT is one of the stranger branches of the cording.ai archive: a software-defined neuromorphic entropy experiment.
NdFractal as memory geometry for residual regimes
NdFractal is a memory-geometry concept. Instead of treating history as only a longer buffer, it compresses past inputs into a state-function landscape intended to separate simil...
ICU monitoring and the cost of false alarms
ICU monitoring is a difficult area for product claims because safety requirements are high and false alarms already burden clinical teams.
HRR as a rescue layer for recurring failure sectors
HRR in TightLoop is used as a structured residual-correction idea: bind context and recurring error patterns into sectors that can be recognized again.
Gimbal inspection footage as a stabilization test
Industrial inspection footage often fails in practical ways: vibration, acceleration events, rolling platform motion, and periods where the sensor view becomes less useful even ...
Stable sensor heads in motion
Drones, payload gimbals, mobile inspection rigs, OIS modules, and robot sensor heads all share the same visible failure: the view still moves after the stabilizer has done its j...
An EUV precision-stage digital twin for process-window residuals
EUV and precision-stage systems are useful examples because residual error has an obvious business meaning: process-window violations, alignment loss, and measurement instabilit...
Longer line-of-sight hold under platform motion
EO/IR turrets, pan-tilt heads, and long-range inspection systems often start with serious control stacks: filtering, feed-forward, friction and shock compensation, gain scheduli...
Turning Chronos-Bolt forecasts into bounded actions on a Jetson Orin
Time-series foundation models are useful because they can produce probabilistic forecasts without training a new model for every site. But operations teams usually do not buy a ...
Drone payload stabilization as residual-error assist
Drone payload stabilization is usually judged visually: how much of the footage remains usable, whether the target stays centered, and whether the horizon or sightline is stable...
Cooling-loop fouling and earlier operational warning
Data-center cooling failures are operational problems, not just sensor anomalies. The cost appears as overheating risk, capacity derating, emergency maintenance, and sometimes u...
Adaptive camouflage as residual contrast control
The camouflage-control experiment treated visible and thermal contrast as something a controller can reduce. Instead of stabilizing a camera or tool, the system adjusts tile tar...
A browser demo for adaptive camera stabilization
The VR adaptive camera demo exists to make the residual-error idea visible. A camera controller can be good and still leave enough motion to bother the operator. The demo shows ...
Automotive perception stabilization under road shock
Automotive perception stacks already have tracking, filtering, sensor fusion, and model-level robustness. But road shock and sensor vibration can still degrade lane-line stabili...
An adaptive camera SDK for residual motion
Game and VR cameras are already controlled by strong baselines: smoothing, spring arms, damping, predictive tracking, and per-title heuristics. The remaining problem is not a la...