NEXT-STATE TRAJECTORY
SIMULATION / SYNTHETIC
Synthetic illustration
Interaction 0
Slider: ← / →, Home / End. Elsewhere: J / L to step, Space to play or pause, Esc to pause. One pass, 600 ms per interaction.
OBSERVATION → ACTION → NEXT STATE
Current state and predicted consequence
| Record | Position | Velocity | Model version | Uncertainty¹ |
|---|
Predictions precede realization under the export contract. This viewer shows exported timing declarations and references; it cannot verify execution ordering independently. ¹ Uncertainty is an exported model quantity, not a calibrated probability.
Action and prediction provenance at this interaction
EXPORTED GLOBAL EVALUATION
Old-regime probes
Inspect absolute error alongside the reported degradation fraction. A tiny baseline can make a fractional change very large. These global probes are separate from the local validation recorded in update receipts. Missing probes are shown as Not exported.
| Method | Probe before MSE | Active after MSE | Degradation fraction |
|---|
ERROR OVER INTERACTIONS
Acceleration squared error
MODEL HYPOTHESES
Competing explanations
Confidence values are exported estimates. They do not establish a physical cause.
SEMANTIC EVENTS
Events at this interaction
Event labels are recorded annotations. A surprise event alone does not identify a regime or cause.
MODEL UPDATE RECEIPTS
Updates recorded through this interaction
Receipts describe what the export records. This page cannot verify training, promotion, parameter change, or digest authenticity.
OBSERVED CONTEXT / STORED ADAPTERS
Cached adapters at this interaction
Context IDs identify stored records. They do not name a physical cause or environment. The selected version is resolved at this forecast boundary, using its historical availability.
Historical cache catalogue
Availability starts at the recorded prediction boundary and ends before the retirement boundary. Final memory figures describe tensor or logical payload storage, rather than measured process RAM.
SELECTION HISTORY / NO PARAMETER PROMOTION
Cached adapter selections
Cached adapter selected · parameters unchanged. Stored cached weights stay unchanged during a switch. Selecting another stored adapter can change the active parameter values; it creates no training or promotion receipt.
PROVENANCE
What this file declares
Exported per-method summary
Aggregates describe the exported scope. This viewer does not calculate cross-seed confidence intervals or certify benchmark conclusions.
READING THE EVIDENCE
Keep the distinctions visible
- Simulation is the only supported run kind. It is not physical-world evidence.
- Synthetic fixture values were authored to illustrate the view; they were not measured.
- Observation is the current exported state. Realized next is the exported consequence after the action.
- Prediction is a method’s next-state estimate. Hypothesis is an uncertain explanation, when explicitly exported.
- Derived error is arithmetic on the same exported transition. Exported error has its own target definition.
- Update receipts and provenance are declarations in the file. They are not an independent audit.
Evaluator artifact boundary Excluded from this inspector
Hidden regime labels, change indices, and true parameters belong to separate evaluator artifacts. This inspector accepts the reviewed rex-run/0.1 and rex-recurrence-view/0.1 simulation formats. Evaluator fields are rejected and are never used to create hypotheses, events, predictions, or model inputs.
The old-regime table contains exported evaluation metrics. The recurrence sample exports no retention probes.
Import limits and normalization notices
Unknown fields are discarded. Imported strings are rendered as text, never markup. No file path, URL, command, or code in a run is executed.