AI Sequencing Fix & Combat Pacing - Results

Generated: June 29, 2026. Companion to healthcare_balance_report.md and
battle_progression_report.md (which still show pre-fix HP values for the
four enemies touched here - see "Stale numbers elsewhere" below).

The bug

Enemy.cs's GetSequencePriority() gave any card with a BlockEffect a
flat priority of 1000 - unconditionally higher than every Attack card -
for any enemy with AiPersonality.Defensive or Optimal. GetCardScore()
does weight Block by an HP-aware panic tier (0.5x at safe HP up to 3.0x
near death), but that nuance never mattered: the categorical sequence
priority always won the sort, so these enemies played 100% of their Block
before touching a single Attack card, every single turn, regardless of how
much HP they had banked.

Combined with a few Defensive decks that are heavy on Block/Heal and light
(or, in Maintenance Drone's case, completely empty) on direct damage, this
produced near-deadlocks: a Monte Carlo sim (scripts/simulate_ttk_healthcare.py,
80 trials x real running HP, not an averaged-HP estimate) measured Maintenance
Drone dealing 0.0 average damage to the player and Claims Adjuster taking
140-200+ turns to resolve in the worst case, against the healthy 9-15 turn
fights produced by Aggressive-personality enemies (which never used this
block-first rule).

The fix (3 surgical changes)

1. Enemy.cs GetSequencePriority() - Block-laying cards only get the
   forced priority-1000 slot below 50% HP. At 50%+ HP they tie with
   everything else at priority 0, so GetCardScore()'s panic-tier weighting
   (already written, just previously overridden) actually decides whether
   Block or Attack is the better play that turn.
2. Character_MaintenanceDrone.tres - swapped 2x Bedside Manner (heal,
   redundant with the 3x Clinical Trial it already runs) for 2x Standard
   Consultation (Attack, 7 dmg + 4 Block) so the deck has guaranteed direct
   damage instead of relying entirely on Quarterly Growth, which only fires
   if there's spare energy left after Block - rare given its curve.
3. HP trim pass on the four enemies that were still the longest fights
   after #1/#2 landed: Claims Adjuster 100→80, Maintenance Drone 145→125,
   Orbital Coordinator 165→140, Regional CEO (boss) 130→115 (~12-20% cuts,
   lighter on the boss to keep it the tier's toughest fight).

Measured results

Stall rate (neither side dies within 200 simulated turns) dropped to 0%
across all 72 enemy x player x mode combinations - the structural deadlock
is gone. Per-tier average battle length (turns until someone dies), floor
mode = worst case (no conditional bonuses trigger) vs ceiling = best case:

Ceiling-mode (competent play, conditionals landing) is now consistently in
a healthy 4-25 turn range for every single matchup - comparable to
Slay-the-Spire pacing, down from multi-hundred-turn stalls pre-fix.

A more honest read of "floor mode"

Floor-mode numbers are still long for a few matchups (e.g. Claims Adjuster
vs Hustler: 89.6 turns). Trimming enemy HP barely moved this number, and in
one case (Claims Adjuster) it got slightly worse between runs. The reason
isn't noise from the fix - it's that in these specific floor-mode
matchups the player has a 0% win rate. The battle "length" being measured
there is actually "how many turns the player survives before dying," which
is governed by the enemy's damage output and the player's own block/HP, not
the enemy's HP pool at all. Cutting Claims Adjuster's HP can't shorten a
fight the player isn't winning in the first place.

This reframes the remaining problem: it's not pacing in those specific
floor-mode matchups, it's raw difficulty - certain starter decks
(Hustler glass-cannon, Plague Doctor) are underpowered against certain
Defensive grunts in their worst-case draw. That's a deck-power-level
question, not a "battles take too long" question, and is out of scope for
this pass - flagging it as a candidate for the next deck-balance pass rather
than papering over it with more HP trims.

Stale numbers elsewhere

battle_progression_report.md and healthcare_balance_report.md still
show the pre-trim HP values for Claims Adjuster (100), Maintenance Drone
(145), Orbital Coordinator (165), and Regional CEO (130), and don't reflect
Maintenance Drone's new Standard Consultation cards. Not rewritten here -
same "stale baseline, needs its own pass" situation as the syntheses already
documented in healthcare_balance_report.md's Implementation Status
section.

What's NOT done

See ai_revamp_roadmap.md for the larger initiative this fix is the first
slice of - the categorical-priority-with-score-tiebreak pattern still has
known rough edges (Chairman/Maintenance Drone special-casing, Quarterly
Growth's narrow energy window, no equivalent fix yet for Aggressive's
raw-score sort potentially overvaluing low-impact damage cards).