Build 9Balance Update

Matched-strength comparison

These are complete required-combat routes, 100 trials/cell. C / D is completion / deathless percent. Deaths and visits are means across all attempts; failures can end early.

Level 9 and all ordinary-reference cells are in the linked tables/CSV; the level-10 result is not being extrapolated to them. The main matrix includes every tier for all five strength fixtures, all three compositions, and all three policies.

Preparation and composition at Master+

These use the independent 200-seed range, maximum legal stat ranks, matching elements, two relevant armor wards and actual paid town visits to change weapons. They are maximum-preparation fixtures, not a claim that a first-arrival Company can afford all this.

The final row is a warning against treating “more control and healing” as automatically better. It does not establish that Juniper should be nerfed or that human optimized play has a 44.5% ceiling. Policy 1 with the same prepared recruit party wins 89%. Prepared Master was 100% complete and deathless for all three parties under both policies.

Individual fresh fights explain the differences. Prepared founders/informed Master+: Spider 100% wins/95.5% deathless; Warden 100/99.5; Veyr 100/100; Owlbear Den 99.5/85; Mites 100/95.5; Index Warden 100/100; Gaoler 100/92.5; Thul 94/63.5. Before matching elements/wards, the same maximum-stat informed founders' fresh Mites result was 0/0 and Thul 41/20. Fresh wins do not prove a route is safe: persistent damage, resource timing and optional fights matter.

The optional Woods spider branch made the neutral maximum recruit route substantially worse (28% completion versus 99% on the goblin branch). The isolated neutral recruit spider fight won 76%, illustrating why a fresh-fight result cannot replace resource-carryover testing. Fire preparation made the fresh founder spider test 100% wins. This is an optional risk/reward and discoverability problem as well as a creature-stat question.

Real campaign progression and economy

Informed policy, all optional content on the selected route, legal training/purchases and recruitment. Levels below are per-hero averages, unlike raw levelStart sums.

The first four Woods completions award 350 Reputation, not enough for numerical combat/economy rank perks. Thus the initial overleveling problem is primarily XP/gear carryover, not Company Rank. On the Woods-to-Master path, the first Heart completion earns 25 more Reputation and Juniper/Heartstring normally. Reserves earn no ordinary XP; introducing Xov and then Juniper legally reduced mean party entry level and initial deathless rates. Their synthetic equal-level performance must not be confused with immediate recruitment value.

Woods-first Master+ was a severe economic gate with this purchase policy: only 2/100 informed campaigns completed it before visiting Heart. Those two then entered Heart Basic around level 11; that tiny survivor group is not population evidence. Alternating first improved the informed Woods Master+ campaign completion to 71/100 in the main seed range, while generic preparation still failed Heart Master+.

The separate aggressive path alternated both books through Master, then selected 20 consecutive legal Woods Master+ runs, then Heart Master+. Lower tiers were never replayed and no rewards were duplicated within a run. In its fresh seeds, 45/100 tactical and 48/100 informed Companies survived the early gates to the later farming stretch. By the later runs they had level 12, weapon 6, armor 4 and resource training 5. At the end of 20 Master+ runs they had Rank X/5,700 Reputation and mean cash 15,431g / 16,942g respectively. Subsequent generic-loadout Heart Master+ still mostly failed: wealth does not automatically make the preparation decision for the player. A stopped campaign can try a different book/roster; these fixed schedules did not exhaust every recovery option.

Wishing-well jackpots were allowed and are separately recorded. A mean +49.99g in a 100-run row represents one +4,999g event, not ordinary income. Do not build normal upgrade budgets around it. Its expectation is about 12.5g per eligible return, but a jackpot is rare and highly disruptive to individual progression. Ten eligible returns give about a 2.47% chance of at least one opportunity; forty give about 9.53% (independent-attempt approximation, with availability promptly consumed).

Exact authored values and placements are in the generated catalog. The following reviews explain their balance implications rather than restating every JSON property.

Heroes, abilities, spells and synergy

All spell options: Ember and Radiant Word are 7.5 automatic single-target damage/1 MP; Briar Lash is 5.5 Earth/1 MP; Void Lance is 9.5 Void/2 MP; Winter Circle is 5.5 Frost per enemy/2 MP. Void Lance depends heavily on affinity because its neutral damage per MP is worse than Ember. Winter Circle is efficient against enough surviving targets but releases any Goat it damages and can defeat the cub before the mother. Gale (-3 attack) and Stone Mantle (+3 AC plus Defended) mainly prevent rolled attacks, not automatic AoE; their duration-2 counters expire at recipients' second upcoming turn start. Barkskin's +2 for 1 MP is useful without a stronger ward, but does not stack and has no Defended benefit. Quiet Focus supplies unlimited eventual Mage mana at an action cost; it does not refill Cleric/Druid mana. Mending, Dawn, Recall, Balm and Goat behave as described above. All 15 current spells were audited; this does not assert all were chosen frequently by the bots.

Dreadbell's player artifact is once per battle, untyped d6+2 per enemy, no stamina, and does not inherit a weapon element. It is strong early against groups and weak per-action against one high-HP target. Heartstring gives exactly one missing MP on fresh combat creation, not a refill and not a reload reward. Its value rises with many small encounters and is lost when entering at full MP. Neither artifact duplicates across wearers. Cosmetic attire is not a combat upgrade.

Status/action economy audit

• Fixed initiative creates predictable early removals; it is not a random initiative advantage every round. Equal initiatives lack an authored hero-first guarantee.
• Natural 1/20 bound ordinary accuracy to roughly 5–95%. At level 12 plus five attack training, hero attack is +19 to +21. Against enemy AC 15–20, much of that training is beyond the useful ceiling. Armor remains valuable only against rolled attacks, with high-tier enemy accuracy eroding it. Guaranteed spells and Backstab bypass the attack roll.
• Personal Defend adds +3 and halves Dreadbell, not general spell damage. Vanish replaces personal Defend. Guard/Shield Wall use the larger AC bonus; Defended duration is independent. A wall or Goat does not turn off the living Dawn Sentinel's passive protection.
• Freeze, Locked and Goat skips overlap rather than queue. They block the whole multiattack action; poison happens first. Frozen flight remains airborne. Goat returns after its second skipped action; damage releases it after applying the original target's defenses. Caster death alone does not remove another actor's Goat/Balm.
• Poison persists into subsequent battles but not walking ticks; healing/rest cures it. Balm cannot rescue lethal poison that resolves first. Sapped is -5 AC/-2 attack, cleared by positive damage after it has affected accuracy, or by HP healing even at full HP. Resource restoration does not clear it.
• Entangled/Off Balance persist through the next completed turn. Critical misses have five penalty outcomes and a larger harmless-miss weight; resource-empty actors cannot roll Stumble. Self-injury is nonlethal. Do not increase fumble frequency to manufacture the desired death rate.
• Support healing, timed party attacks, two-strike actors, flight takeoff/swoops, maternal rage and retreat all run through the real engine. Ordinary buffs lasting “two turns” are not equivalent to two complete rounds of protection; initiative matters.

Every creature and encounter

Economy, challenges and recovery

Leveling adds +3 HP and +1 attack per level, not resource capacity, AC, initiative or spell damage. Training adds +1 attack, +3 HP or +1 capacity; the five purchases cost 15/30/75/150/255g. Weapon ranks cost 20/40/100/200/340/520g; armor costs 30/60/150/300g. Maximum stat/weapon/armor ranks cost 3,335g per hero, before elements and spells. Four heroes therefore require 13,340g, plus wards/spells/attunements. Fully maxed fixtures are a long-run envelope, not a typical two-book first completion.

XP purchases cost 60 × current level. The level-1 auto-promotion at 60 XP does not subtract XP. With ordinary rewarded auto-promotion, cumulative earned XP thresholds are: level 3 120; 4 300; 5 540; 6 840; 7 1,200; 8 1,620; 9 2,100; 10 2,640; 11 3,240; 12 3,900. Manual level 1 purchase before ordinary auto-promotion would differ. Do not silently “fix” the documented existing rule or overwrite old XP.

Gold and XP share multipliers 1/1.8/2.8/4/6, but XP goes to each surviving active hero, not one pool divided four ways. Gold is pooled. Death therefore has both a resurrection cost and a progression cost. The separate XP and gold demand curves are already diverging: level 12 can arrive while substantial upgrades remain unaffordable.

Healing potions cost 6g, average 8.5 HP and consume the user's action; capacity 20. At maximum training a hero has roughly 58–70 HP. Coffee costs 12g for one resource and one action; its action timing is as important as its price. Smoke costs 10g for +3 Guard without Defended and competes with reusable defenses. Provisions cost 15g and cap at 99; free camps do not spend them. Routine authored provision awards remain zero/one, unscaled. Noncombat chest bonus chance is 1/6, uniformly among four items; no combat-chest bonus, no repeat receipt exploitation. Full-stack conversion is scaled 15g; One-Eyed Jack opts out.

Town rest refills all living active HP/resources for free. Return costs one provision or ceil(10% gold), capped at 50g; it can be free at zero gold. Thus purchasing 15g provisions can be irrational below about 140–150g, while spending down the treasury before later returns lowers the fee. This is an incentive, not an infinite-gold exploit. Flee requires a provision plus floor(5% gold); wipe costs floor(10% gold); neither receives a second normal return charge. Resurrection is a flat 30g regardless of tier, full HP but no MP until rest. At 6× reward income that is a smaller burden. Reserve survival suppresses emergency mercy; replacing fallen active members remains a legal escape the fixed-party bots do not exploit.

All current challenge definitions/approaches were inspected. Their DCs do not scale with tier or hero level; party IDs, requirements and modifiers matter. The bot uses the best available modifier and correct authored puzzle answer. This overestimates uninformed first-play performance. Failure is generally nonlethal damage, resource loss or reduced reward, not a route lock. Rewards scale up while DCs stay fixed, making solved content a reliable future income source. Its one-shot receipts work; only a fresh legal expedition resets them. There is no evidence here of a duplicate-reward bug.

Company effects are bounded: rank VI's discount rounds a potion back to 6g; VII increases mean uncapped potion healing to 9.875; VIII discounts stat training, not gear/XP/resurrection; IX changes chest odds to 1/3; X adds 20% ordinary gold, with documented exclusions. In the neutral maximum recruit Woods Master+ ablation, rank I→X changed completion 99%→99%, deathless 87%→90%, and net gold approximately 1,533→1,817g. Those few deathless points are not conclusive; the economic increase is real. All corresponding Heart generic-loadout cases still failed. Rank rewards accelerate saturation more than they directly trivialize combat.

The system has bounded combat stats but unbounded opportunities and currency. Master+ itself does not compound. That avoids runaway enemy scaling, but repeated/future adventures saturate XP, upgrades, learned spells, stocked consumables and rank benefits. A single future book cannot assume “its Basic is first” or infer party strength from that book's tier.

Illustrative projections, explicitly assuming new books resemble the current books and average 250–320 XP per surviving active hero on Basic, all rewards collected, no deaths, no bench time, no wish, and unchanged tier multipliers:

These are arithmetic scenarios, not simulations of nonexistent Adventure 3 content. Actual rewards, recruitment, deaths and entry order change them. The real two-book alternating campaign already confirms the underlying saturation mechanism.

Each book's five first completions give 600 Reputation. Without any repeat grind, two comparable completed books yield 1,200 (VI), three 1,800 (VII), four 2,400 (VIII), six 3,600 (IX), nine 5,400 (X). Existing rank benefits stop adding combat power beyond their configured caps, but more ordinary gold and discounted supplies continue feeding the treasury.

There is a design incompatibility to resolve explicitly: preserving all earned power, keeping every encounter fixed, and requiring the same tactical difficulty from both level-3 and capped Companies cannot all hold simultaneously. Rewards alone cannot correct existing veteran saves. Merely raising the level cap restarts the problem and further flattens d20 accuracy. More enemy HP alone lengthens attacks-only battles without making them tactically richer.

These are proposed acceptance bands for a complete mandatory route at its supported power band, with one camp and normal legal recovery. Optional boss/chest content has a separate risk budget. They describe encounter outcomes, not a quota of unavoidable random deaths. Evaluate at least three compositions, both first-entry and veteran Companies, and include human testing.

Additional gates: a prepared individual mandatory fight should ordinarily win at least 90%; avoid mandatory routes whose aggregate success is dominated by one surprise mechanic. Bosses should generally take 4–8 meaningful rounds, with occasional 9–10 round endgame fights, not repeated 15–20 round neutral-damage slogs. No single untelegraphed enemy cycle should normally remove over half of each healthy properly prepared hero's HP. A full-party wipe from full health before reasonable defensive response should be exceptional. Track mean and tail deaths separately from completion; repeat-resurrection loops are not evidence of fair challenge.

For resource pressure, require Expert+ no-item/no-ability ablations to lose at least 15 percentage points of route completion or materially increase deaths, and show that expenditure is funded by ordinary route income. Do not require a potion count when the party legitimately substitutes spell healing/control. Put an upper bound on menu/travel time through human playtests; six or seven resupply/weapon-change returns in a ten-location book is already a warning. Never count the same convenient preparation/rest visit twice.

Immediate Build 9: the smallest useful tuning pass

1. Adopt the test contract and baseline first. Keep Basic's new-campaign and Heart level-3 accessibility. Use the real-API driver and prepared benchmark; retire generic Mage Skill samples as evidence of player spell balance. Add two healer-free parties and at least one double-Mage control party before accepting a release tune. No gameplay change is required for this gate.

2. Smooth party-wide damage at its source. Candidate: give timed abilities an optional cap on the tier-added damage, default uncapped; set Ward Mite Faultline Pulse's cap to 6. Formula: base die + base bonus + ceil(min(tier.damage, configuredCap) * tierDamageMultiplier). Existing lower tiers 0–6 remain unchanged; Master+ pulse becomes d2+6, or 21–24 combined damage per hero for three mites before resistance, instead of 39–42. Retain Earth defense/Void offense as advantages rather than near-binary requirements. Do not change all automatic attacks or remove elemental identities. A simpler 0.5 multiplier would also weaken Advanced–Master, which is why the cap is the preferred candidate. Re-test the entire Heart route; this may require moving some endgame pressure to readable boss decisions rather than restoring the same burst elsewhere.

3. Make Veyr's signature scale and remain answerable. Candidate: enemy Dreadbell adds ceil(0.5 × tier.damage), preserving the player's artifact and scripted prologue. Basic stays d6+2; Master+ becomes d6+8 before Void resistance/Defended. Keep the explicit warning and once-per-fight action. This corrects a clear scaling discontinuity, but by itself may not reduce prepared deathless completion enough. If it does not, test one additional telegraphed Master+ toll at round 6 using generic authored timing, rather than adding random unavoidable damage or simply doubling HP again. The extra toll is a second-stage option, not a bundled recommendation to ship blindly.

4. Test small higher-tier candidates after those mechanics are separated. Initial sweep values, not approved replacements:

   Tier	HP multiplier candidates	Attack bonus candidates	Defense bonus candidates	Ordinary damage bonus candidates
   Advanced	1.5 / 1.8	2 / 3	1 / 2	2 / 3
   Expert	2.2 / 2.6	4 / 5	2 / 3	4 / 5
   Master	3.0 / 3.5	6 / 7	3 / 4	6 / 7
   Master+	Keep 6 initially	Keep 9 initially	Keep 4 initially	Keep 12 initially

   Select the lowest changes meeting the bands, starting with survival/action opportunity, not simultaneous maximums. Cross every candidate with automatic/two-strike/support/control encounters. Do not label any candidate successful before running it. Current simulations did not mutate tier values.

5. Reduce preparation friction without removing costs that create decisions. Candidate: allow equipping already-owned weapon attunements at an unused or used camp, out of combat, with no purchase/refund there. Keep new purchases, wards, stat upgrades, resurrection and learning in Town. This is a small extension of ownership/equipment, not a new inventory system. Measure actual reduction in town visits; camps are not positioned to eliminate every preparation trip, so do not promise a particular count. Avoid a new rest tax, provision inflation or resurrection-price jump in the same pass.

Affected implementation owners, if later authorized: GlobalContent.json; timed-ability definitions/validation and combat construction in JournalData.cs, JournalContent.Validation.cs, JournalGame.Combat.cs, VintageEncounter.cs; attunement eligibility and presentation in JournalGame.Progression.cs/the existing UI; relevant pure/Unity verification; both handbooks, Combat Abilities, Company/economy guidance where affected, and the Build 9 change record. Preserve shared creature identity, artifact behavior, defaults and old snapshots. New timed data must be generic, not a location-name branch.

These immediate changes do not solve cross-library overleveling. They improve today's outliers and make a later progression decision measurable. Do not claim the full intended difficulty contract while veteran power remains unrestricted in lower tiers.

Library growth: compare the structural options

Recommended long-term contract: explicit power bands, implemented once in shared stat derivation and exposed on adventure/tier selection. Initial ceilings for experiments:

Use effective=min(owned, ceiling); never raise an underleveled hero to the ceiling. Preserve learned choices, recruits, cosmetic ownership and Company history. These ceilings intentionally leave some overlevel advantage on early tiers and full growth in Master+. They require re-tuning the underlying encounters against the resulting envelope; no results in this report measure these proposed ceilings.

A small shared powerBand definition can be referenced by modules/tiers; avoid adventure-name conditionals. Decide openly whether Basic should remain an unrestricted victory lap instead. If so, exempt Basic and say that its “challenging moments” promise applies to new Companies, not veterans. Advanced and above still need an enforceable envelope to meet the stated universal tactical requirement.

The hard part is compatibility, not calculating min: changing the effective HP/resource maximum must never heal, resurrect, refill, erase earned ranks, or corrupt an interrupted combat when switching books, swapping heroes, training or loading. Before implementing, specify a single authoritative representation for wounds/resources across bands, with rounding that cannot be farmed. Persist an additive effective-combat context for in-progress fights so an update cannot reinterpret an old snapshot's actor maxima. Do not patch this with renderer-only stats or repeatedly rescale integer HP at each redraw. Include old +1 legacy items in the ceiling budget instead of deleting them.

Do not bundle the proposed XP cuts with the first band implementation. First measure whether meaningful ability/gear growth and training time remain satisfying. Reward controls are secondary: retain normal gold so players can fund consumables and learned options, and protect recruit catch-up. Consider reserve catch-up independently, e.g. nonretroactive 25% reserve XP below the active median, capped at that median, only if player tests confirm recruitment friction; it is not an implemented rule or a necessary part of this checkpoint's recommendation.

Acceptance tests and stop conditions for a future implementation

1. Run unchanged-seed baseline/candidate comparisons and at least 1,000 new seeds for key release cells. Report confidence intervals, conditional campaign attrition, deaths after resurrection and resource/town tails. Keep policy 0 as a meaningful negative control. Do not hide a weaker policy behind the word optimized.
2. Validate every new ability/scaling field at minimum, maximum and default/legacy values. Verify shared creatures across both books, element rounding, all resistances, player artifact/prologue isolation, multiattack and timed windows. Test capped and uncapped loadouts at each tier.
3. Pass the full legal paths: Woods through Master then Heart; immediate alternating; both recruits at actual recruitment levels; two-books-then-final-tier farming; early wipe with living reserves; empty gold/provisions fallback; no jackpot assumptions. Include a high-wealth existing Company and legacy equipment.
4. If power bands are adopted, repeatedly switch books/tiers, swap low/high-level heroes, save/restore mid-turn, change equipment, hit a camp, and return from town. Prove no HP/MP/XP/currency duplication, no death reversal, no loss of ownership, no altered rewards on old receipts, and stable interrupted combat. Test domain-reload-disabled sessions and real Unity JSON/disk in isolated Practice.
5. Review both canonical handbooks against every changed rule/data entry; update them and subsystem documentation in the same implementation change set. Record first-entry expectations, supported ceilings, actual reward formulas and hidden enemy scaling in the appropriate handbook.
6. Human-test Basic accessibility, whether players discover elemental needs fairly, whether Master+ deaths feel avoidable, and whether recovery/attunement trips are tedious. Physical controller/touch and presentation checks remain required for new equipment/band UI. Stop if the numerical target is met only by unexplained burst, repeated revival deaths, excessive town menus, or one mandatory party composition.

The active Combat Abilities guide retains an older sentence saying Shield Wall does not reduce Dreadbell; current IsDefended and the Player's Handbook say it does. The persistent-death guide still points tier resurrection configuration at ForsakenWoods.json and mentions a six-person roster; current shared configuration is GlobalContent.json and the roster can include eight. Company-Reputation.md retains historical pending-art prose despite all eight Gilded resources existing. Both handbooks' introductory audit labels still say Build 8 even though their maintained appendices include tonight's Build 9 rules. These are documentation findings, not permission to revert implementation. No historical record was rewritten during this audit.

The checkpoint ends here. All proposed values, power bands, reward controls, ability changes and camp equipment access remain unexecuted.