Decision systems often make a destructive shortcut: something failed once, therefore stop considering it.
Possibility Reserve Architecture is built around rejecting that shortcut when the failure depended on conditions that can change.
Its protected spine contains four mechanisms.
Conditional Failure Memory
Failure attaches to the conditions under which it occurred. A branch that failed because money, access, timing, information, hardware, or another dependency was missing should preserve that context. The record becomes failed under condition-set C, not globally impossible.
Graded Dormancy
An option can become less active without being erased. ACTIVE → UNCERTAIN → DORMANT is a resource-allocation sequence, not a moral judgment on the idea. Dormancy preserves why the branch was paused and what evidence would justify waking it again.
Causal Reactivation
When the world changes, the system checks which dormant possibilities were suppressed by the changed condition. A new resource, capability, contact, law, tool, or piece of evidence can therefore wake a specific branch without forcing the entire archive back into active attention.
Path-Expansion Search
A branch can deserve exploration because of the future space it opens, even when it is not the highest-probability immediate answer. Information gain, option value, capability gain, and future branching can justify some compute.
The architecture is not an argument for never abandoning bad ideas. It is an argument for making elimination proportional to evidence.
Failure should teach the system what failed, where, and under what state. It should not silently delete futures the evidence did not actually rule out.