calculated fallbacks - the chain learns which doors open
This commit is contained in:
@@ -700,3 +700,27 @@ and the receipts on the surviving copies would still prove the seal
|
||||
was due.
|
||||
|
||||
Inevitability by construction. Down to the second, eventually.
|
||||
|
||||
### Calculated fallbacks — not a fixed ladder
|
||||
|
||||
Upgrade once the substrate lives: fallback ORDER is computed per
|
||||
attempt from the opinion space itself. Every path is an entity with
|
||||
its own readings:
|
||||
|
||||
trust(path) # has this route verified before?
|
||||
freshness(path) # when did it last work?
|
||||
cost(path) # energy/time to try now
|
||||
stability(path) # how often does it change?
|
||||
|
||||
next_move = argmax over open paths of
|
||||
trust_eff · freshness − cost
|
||||
clamped by §9 tolerance
|
||||
|
||||
Every attempt reinforces or wounds the path's own record — the chain
|
||||
LEARNS which doors open easiest in which weather. And because paths
|
||||
are entities, a failing route gets negative reinforcement exactly like
|
||||
a lying source: the system's confidence in it sinks until repair.
|
||||
|
||||
The fixed order above is version 1 — enough to ship. The calculated
|
||||
version is where it wants to end up: a seal that doesn't just keep
|
||||
trying doors, but knows which door to knock on first.
|
||||
|
||||
Reference in New Issue
Block a user