Recovery with incomplete knowledge
Question and method change
18 November 2026. The foray asks whether alternative construction methods remain feasible after irreversible work. Essai 2 computed paths in a fully observed finite graph. Here the same physical alternatives become an information-space game: the controller must choose a common first action for worlds it cannot yet distinguish. This extends the enquiry; it does not alter the earlier model or certify a bridge design.
LaValle’s Sensors and Information Spaces provides the move from unknown physical state to an information state. Planning Under Sensing Uncertainty provides planning over that space. The tiny bridge game, durations and exact boundary below are authored extensions. No stochastic probabilities or opponent preferences are assumed.
Complete finite game
At time zero exactly one route is available, L or C, and that hidden regime stays fixed. The cable route also requires a retained conversion interface. There are no later disruptions, damage from refusal, resource conflicts or partial successes. The two methods are substitutes for finishing the same remaining bridge work. Completion exactly at the access deadline counts as in time.
The team can wait for exact routine status at time r, buy an exact direct check taking q ticks, attempt lift, or attempt cable if its interface exists. Successful lift takes a=2 ticks including preflight; successful cable takes b ticks including preflight. An unavailable lift refuses after fL=1 tick; an unavailable cable route refuses after fC=2 ticks, revealing the other world. No irreversible work happens on refusal. A successful attempt finishes the work. Attempts are nonpreemptive; an arriving routine report cannot shorten an attempt. Direct checks have time cost only. There is no monetary optimisation.
After any informative outcome the regime is known, so completing its sole remaining route is optimal. Waiting before choosing cannot help: all durations are constant and only time to closure shrinks. Repeated checks cannot help because the first is exact. Thus the four first-action policies exhaust the relevant policy class. If r=0, the report is available before the first choice.
F* = min(r + max(a,b), q + max(a,b), max(a, fL+b), max(fC+a,b)) Guaranteed recovery iff interface exists and F* ≤ deadline. Each world individually reachable iff interface exists and max(a,b) ≤ deadline.
The interface-free case has no completion in C, so no common guarantee, whatever information is supplied. These formulas are independently checked against the branch solver. The gap between the two conditions is the price of not knowing the world in time. The formula is specific to this complete finite action set; it is not a general closed-form solution to partially observed planning.
Result and counterexamples
With a=2, b=3, r=2, q=1, fL=1, fC=2, the world-specific completion times are 2 and 3 but F*=4. Closing at 3 defeats every common policy. Closing at 4 restores a witness: check for one tick then use the revealed route. A safe lift-first attempt ties it. If b=2 through work completed before this decision, F*=3. If r=0, F*=3. If the interface is absent no information improvement restores the cable route.
Fast status is therefore sometimes a construction option, in the precise sense that it makes a feasible conditional policy executable before the deadline. It does not substitute for missing hardware. The result is a robust finite-horizon reachability statement, not a claim about structural safety or an infinite-horizon viability kernel.
Verification and open direction
The independent test compares all 600 interface/deadline/receipt/check/cable-time combinations exposed by the interface with the closed-form minimum, and checks the opening counterexample. The new page is reached through the existing foray entrance. It is intentionally ephemeral; there is no save, upload, receipt object, file creation or history to reopen.
A stronger next pull is timed games with imperfect information: which observation delays and disturbance sequences still admit a strategy after irreversible transitions? The present exact game gives a baseline for that question, while avoiding an unjustified guarantee over disruptions it does not contain.