The temporal ↔ decision seam — note

A short note on temporal_decision_seam.py: a worked example wiring temporal_governor to governed_decision at their natural join — the present.

---

The seam

A decision is always made now, but it draws on the other two tenses, and the composition makes each one's role explicit and honest:

• Past → evidentiary footing. A VERIFIABLE (recorded) track record is a mature basis to decide on; an UNRECORDED past is not. The recorded-past status selects the evidence the decision stands on, so a claim you cannot verify cannot silently become the ground you act from.
• Future → belief. A FORECAST's probability becomes the decision's posterior. Only a forecast qualifies: a future asserted as certain is refused upstream, so the decision is driven by an explicit probability, never by a pretended fact.
• Present → when. The ACT_BOUNDARY is where the decision fires — you observe or decide now, and cannot re-check it later.

What it enforces (verbatim from the runnable)

Past	Future	Decision
recorded (VERIFIABLE)	forecast 0.92 + human signed	AUTHORIZED_ACT
recorded (VERIFIABLE)	forecast 0.20	GATHER_MORE (below the act threshold)
unrecorded (UNRECORDED)	forecast 0.92	WITHHOLD (no verifiable footing)
recorded (VERIFIABLE)	asserted certain	refused at the seam — no decision fires on a pretended fact

The composition thus makes one discipline structural: act now, on a footing you can verify (the recorded past) and a belief that is explicitly a probability (the forecast) — never on an unrecorded past treated as known, nor a forecast treated as fact. The two failure modes the temporal governor names (an unrecorded past asserted as known; a future asserted as certain) become, at the decision, a WITHHOLD and a refusal-to-fire respectively.

Honest limits (inherited)

Both components' limits carry through unchanged: has_record and the forecast probability are declared inputs, not things the tool measures; a VERIFIABLE past can still be false against its record and a FORECAST can be badly calibrated; and the case-to-belief mapping is a modeling choice that demonstrates the composition's guarantees — verifiable footing required, probability-not-fact required, human authority required — not the correctness of any particular number. Non-safety-critical scope, deterministic, self-testing; reuses temporal_governor and governed_decision unchanged.