Add storylet and eligibility-hub specialist search #110

Product outcome

Test mechanically driven storylet hubs for rare eligibility states, starvation, deadlocks, cooldown/timeout behavior, and newly enabled content without enumerating every ordering of already-seen storylets.

Parent: #97. Related: #53 shared frontier research, #60 variable inspection, #63 checkpoints, #105 marginal value.

Trigger signals

Repeated return to a hub plus changing variables, changing offered-choice sets, conditional gathers, cooldown/visit counters, or many content knots reachable from one dispatcher. Signals are factual and may be uncertain.

Specialist behavior

• prioritize novel eligibility/choice-set transitions and newly enabled authored content;
• retain diverse hub checkpoints;
• detect observed starvation/deadlock witnesses;
• discount permutations that return to an equivalent effective hub state;
• reserve broad probes for rare later eligibility changes.

Acceptance criteria

• Hub and offered-choice-set identities are deterministic.
• Eligibility novelty is separate from raw terminal/state multiplicity.
• Reports show which storylets became eligible, remained unseen, or were observed starved; bounded unseen is not unreachable.
• Fixtures include quality ranges, cooldowns, mutually exclusive storylets, timeout endings, priority starvation, and a rare late unlock.
• The specialist finds the predeclared late unlock earlier than the fixed portfolio on at least one fixture without losing critical evidence across the family.
• Checkpoint/frontier memory remains bounded under Bound checkpoint frontier memory and make deep runs stop cleanly #98.
• Fixed-seed decisions and witnesses replay exactly.
• Promotion requires Build the search-strategy promotion benchmark harness #56 public/synthetic family evidence, not one favorable hub.