Failure policy

Declarative dispatch from type[EntorinError] to a recovery PolicyOutcome. The control plane names the recovery; the orchestrator acts on it.

from entorin.errors import TransientError, ParseError
from entorin.policy import PolicyTable, PolicyOutcome, RetryPolicy

table = PolicyTable(bus=bus, default=RetryPolicy(max_attempts=1))
table.register(TransientError, RetryPolicy(max_attempts=3))
table.register(ParseError,     RetryPolicy(max_attempts=1))

outcome = await table.handle(error, ctx, attempt=current_attempt)
match outcome:
    case PolicyOutcome.retry:           ...
    case PolicyOutcome.replan:          bus.publish(replan_event); ...
    case PolicyOutcome.fallback_human:  await checkpoint.request(ctx, ...)
    case _: raise

Outcomes

Five fixed values: retry, replan, skip, fallback_human, terminate. Reference policies (RetryPolicy, ReplanPolicy, SkipPolicy, FallbackHumanPolicy) ship; user-supplied dispatch logic plugs into the same shape.

Lookup

MRO-walking — registering EntorinError catches everything not otherwise handled. Six typed categories all inherit EntorinError: TransientError, PolicyError, BudgetError, UserAbortError, ToolError, ParseError.