Select replenishing hunts that can fund recovery

[adrunkhuman]

The supply-aware selector can choose a cheap but non-replenishing two-spawn pocket. A poor bot needs a repeatable hunt that earns gold, not an adjacent-point patrol that prevents normal respawns.

Observed failure

After #181 (branch revision 7e2eb49, merged as 3fe99eb), the normal Carlin-start soak selected a two-Troll pocket:

Evidence, 2026-09-09 UTC	Value
Selection at 21:43:15	lowest_potion_consumption_then_xp
Atlas site/variant	14906939626827278936
Center	(32459,32276,8)
Patrol endpoints	(32458,32274,8) and (32459,32276,8)
Supplies / money at departure	2 potions; 100 carried gold + 32 bank gold
Budget	2 reserved, 0 routine; estimated consumption 1 potion; does not fit
Patrol after initial combat	291 waypoint arrivals from 21:44:47 to 21:49:12, with no new target selections during that loop

The adapter creates approach points near spawns and removes only exact duplicates. Its recurring-XP estimate uses spawn intervals without accounting for the bot blocking replenishment. Spawn::findPlayer / spawnMonster suppress ordinary respawns near players unless the monster ignores spawn blocking. The trace establishes unproductive pacing; it does not contain per-spawn rejection telemetry.

Local evidence: /tmp/playerbot-stuck-check.log. This file is not a portable artifact; the essential observations are recorded above. The soak is not a deterministic reproduction or a completed #148 benchmark.

Agreed direction

• Keep isolated 2–3-spawn pockets as atlas building blocks, but do not offer them as standalone sustained hunts. They may contribute to a larger viable circuit. Do not add one-time-clear goals as the solution.
• Require respawn-compatible movement for sustained hunts. Spawn count alone is not enough if the bot remains in blocking range of every spawn.
• Start with gross coin gold per minute, estimated from static loot probabilities and expected sustained kills. Record actual collected coins separately. Exclude resale valuation and cost subtraction; never inspect unopened corpse contents.
• When supplies are low and resupply is unaffordable, prefer gold income over XP among otherwise eligible hunts. Low potions alone must not trigger money farming when resupply is affordable.
• Keep hunt facts separate from selection preferences so Add bounded progression preferences #173 can later adjust objectives above deterministic execution and hard safety gates. Do not implement personality in this issue.
• Preserve Budget hunt supplies, prioritize Exura, and avoid transit fights #181's safety gates, emergency/route reserves, Exura priority, and transit behavior. Do not change potion thresholds or seed funds to hide this failure.
• Keep selection town-independent, bounded, map-derived, and explainable. Do not hardcode rats, wolves, caves, or NPC names.

Next work and acceptance

• Reject standalone 2–3-spawn pockets while retaining their membership in larger candidates. Define additional bounded geometry/timing criteria from actual atlas candidates; spawn count alone must not establish viability.
• Test replenishment-compatible circuits and invalidation when patrol changes. Recreating the exact Carlin failure is not required.
• Estimate sustained gross coin gold per minute and emit income, cash pressure, and viability facts in candidate telemetry; record verified coin acquisition and hunt income outcomes.
• Add focused comparisons where a safe gold-producing hunt outranks an XP-focused alternative when recovery money is scarce. Preserve safety, supply-budget ordering, emergency/route reserves, and affordable-resupply behavior.
• Run focused regressions and a normal soak; report productive combat, acquired coins, and any empty-patrol or no-candidate limitations. Do not claim sustainable 8–20 progression from fixture success.

Net profit, item resale, travel/service cost conversion, exact incident reconstruction, and the personality implementation are outside this iteration.

Start in server/src/playerbothuntregionadapter.cpp (candidate extraction/throughput), playerbothuntregions.h and playerbotsupplypolicy.h (ranking/budget), playerbotcombat.cpp (route reconciliation/patrol), and spawn.cpp (replenishment rules). Inspect existing loot valuation and sale planning before adding a parallel value model.

Related: #51 owns hunt strategy; #148 owns the controlled three-town 8–20 benchmark; #35 covers loot valuation/disposition; #46 covers recovery supplies. #181 is the implementation baseline, not completion of these contracts.