Temporal telemetry — one worked example, four components

A short note on temporal_telemetry.py: a single proxy/truth stream carried from decoupling detection through an early warning to a governed action taken before the failure is visible.

---

The pipeline

Step	Component	What it produces
1	decoupling_monitor	a proxy (reported metric) and an independent truth signal stream in; it flags when they decouple and reports the alert step and the lead time before the truth crosses a failure line
2	forecast (heuristic)	the decoupling becomes a forecast probability that the truth will breach — projected from the truth's downward slope using only data up to now; an estimate, never a fact
3	temporal_governor	the recorded stream = past (VERIFIABLE); the breach = future (FORECAST); now = present (ACT_BOUNDARY)
4	governed_decision	fires at the boundary — footing from the recorded past, belief from the forecast, timing from the cost asymmetry — and acts on a reversible, human-authorized mitigation

What the run shows

On the shipped stream: the monitor raises a decoupling alert at step 16, and the truth crosses the failure line at step 25 — a 9-step early warning. As the truth keeps falling the forecast probability rises; the governed decision crosses its act threshold and fires AUTHORIZED_ACT at step 19 (forecast ≈ 0.66), leaving 6 steps of runway before the visible failure. The action is a reversible throttle-to-safe-mode, human-authorized, one logged step.

Two honest properties fall out of the composition, not from any single tool:

• The decision waits at the first alert and acts later. At step 16 the failure is still far off, so the honest forecast is low and the decision is GATHER_MORE — waiting is correct there. It acts only once the forecast has risen enough to clear the cost-weighted threshold, which is still before the failure. Early warning and premature action are kept distinct.
• It acts on a forecast, never a fact. Because the breach is governed as a FORECAST, the decision is driven by a probability that rises with the evidence — and a future asserted as certain would have been refused upstream. The recorded past supplies verifiable footing; the future supplies belief; the present is merely when.

Honest limits

• The forecast is a transparent heuristic, not a calibrated model: it projects the truth's recent slope to the failure line. A better forecaster drops in at that seam; the composition and its guarantees do not change.
• Lead time depends on the monitor's sensitivity (window, thresholds), which are declared config, not universal constants — a more sensitive monitor warns earlier but false-alarms more, which is exactly the cost asymmetry the decision layer weighs.
• It governs the decision process, not the outcome. Acting early on a reversible mitigation is the right shape of response to an early warning; whether this mitigation fixes this failure is a domain question the example does not settle. Non-safety-critical scope; deterministic; self-testing; reuses decoupling_monitor, temporal_governor, temporal_decision_seam (hence governed_decision + optimal_timing) unchanged.