Build 10Balance Update

5.2 / Build 10: first implemented balance pass

24 September 2026. Implemented for playtesting, following the owner's authorization to move beyond the earlier analysis-only brief. This does not create a release binary, change upload counters, or declare the complete balance vision achieved.

Executive result

Woods and Heart now share a tier-unlock ladder: complete both on Basic for Advanced, both on Advanced for Expert, and so on. Completed tiers can be replayed for ordinary XP/gold and completion Reputation. Expert+ recommends near level 12, with no level requirement. Earned access and higher completion history are preserved. The traversal-only Lost Dungeons preview remains playable but is excluded from required clears until its actual content is ready.

Basic is unchanged. Advanced is firmer; Expert and Master have much stronger stats, with bounded enemy status-proc increases and specific boss/encounter adjustments. In the final independent 500-seed screen, prepared level-12 founders complete Master+ 52.6% in Woods and 58.2% in Heart, versus 100%/91% in the newly matched pre-pass benchmark. Deathless completion is 19.6%/17%. This is a useful first playtest package, not a universal success-rate prediction.

The growth escape route is real: all 500 modeled Companies reached level 12 before Expert after four additional Advanced circuits through both books, earning XP and paying for training normally. Two extra circuits brought them to about level 10 before Expert. Players can also attempt Expert immediately after its clear requirements; the direct campaign enters it around level 7. No artificial XP catch-up or well windfall is required.

Remaining weaknesses are explicit: better gear still trivializes lower tiers; basic-attack play can limp through some higher content using recovery; Master has a steep equipment threshold; the recruit-heavy party performs poorly on Master+; and frequent town/attunement visits may feel repetitive. Those are the next tester questions, not reasons to hide the current results or apply another blanket multiplier.

Source and preserved work

Source HEAD at task start: bc7f1638bb28d13114d0f720b0b24b174a192fd8, main reported seven commits behind its remote-tracking reference. No pull, branch switch, reset, commit or push was performed. The initial dirty tree already contained the approved elemental-affinity changes, ally-targeted potions, narrated opening, Lost Dungeons traversal/assets, related documentation, artwork/import edits and prior campaign-model tooling. Those changes were preserved and the first two gameplay changes are included in the simulation baseline.

Before editing, 398 relevant files were copied with hashes to ignored Logs/Build10-Balance-2026-09-24/before, alongside initial status and the preexisting diff. The durable evidence directory contains source/driver snapshots, manifests, raw trials, summaries and validation evidence. A working-tree source manifest is essential: this pass was not simulated against clean HEAD or the older September 23 audit source.

Read with AGENTS.md, Adventure Library, Player's Handbook, GM Guide, Master+, and the historical twelve-iteration proposal. The earlier ten-adventure story-only campaign, hard level-12 gate, bounded power bands and speculative hero/spell redesigns remain unimplemented.

Implemented rules and tuning knobs

GlobalContent.json owns progression.requiredAdventures, currently Woods/Heart, and allowCompletedReplays: true. Add a content-ready book to the required list when it should participate. Library validation rejects a required ID whose module is missing or unavailable. No adventure-name branch was added to generic combat.

The authoritative JournalGame.Difficulty.cs checks global eligibility and each book's own previous clear. New saved fields difficultyProgressionRevision=1 and unlockedDifficulty permanently record earned access. A future book affects the next unearned unlock, not old access. Revision-zero saves infer earlier clears from their old linear tiers and retain earned higher-tier access. There is no roster-level gate.

The introduction has a tier selector, completion marks, preparation advice and missing-clear explanations. Replacing an unfinished expedition at another tier requires confirmation. The transaction resets only that book's route, camps and receipts, retaining Company resources, higher clears and other books. A new expedition gets a distinct ledger ID and can earn its rewards once again. Fresh starts restore living active heroes; resurrection rules remain unchanged.

Enemy proc bonuses by tier are +0/5/10/15/20 percentage points for poison, on-hit debuffs and Sapped; +0/3/6/9/12 for Freeze/Lock; +0/1/2/3/4 for explicitly resisted Freeze. Soft effects cap at 60%, turn denial at 50%. Absent effects stay absent and an authored chance already above a cap is not reduced. Grave Chill becomes 33/36/39/42/45%, or 10/11/12/13/14% with Frost resistance. Lockdown becomes 40/43/46/49/50%. Ability schedules and durations are unchanged: increasing a successful check does not make a dead or controlled enemy act. This avoids turning ordinary fights into repeated guaranteed turn loss.

Creature tierAdjustments apply before global scaling. Veyr gains two strikes on Expert+, with base bonuses 0/8/16. Master+ Hollow Warden gains two strikes and base bonus 6. The three Ward Mites' timed pulses use d2+2/+3/+6 on Expert/Master/Master+, without adding global tier damage again. Expert/Master Mother uses base HP 38/34, attack 3 and bonus 0. Master+ Thul uses HP 55 (330 scaled), normal d6+2 before tier bonus, Cleave d4+2 plus half tier damage, and three-strike rage below 25% HP. Exact probabilities and profile semantics are synchronized in the GM Guide.

New fights capture resolved tier/enemy profiles in combatTuning, including adventure-owned snapshots and detached save copies. Loading an interrupted fight preserves those stats, abilities and probabilities after later tuning edits. Old fights without a context use legacyCombatTiers and pre-adjustment profiles. This is scoped protection for this balance pass; current authored encounter relationships still govern protection and retreat. Existing save schema/IDs, XP curve, level/gear caps, prices, rewards and Company benefits remain intact. No player saves were read or altered by verification.

Iteration decisions

The twelve earlier hypotheses are retained in the September 23 report. Five further implementation screens incorporated current affinity/potion behavior:

The policy correction matters: the old generic recovery thresholds sometimes sent excellent gear into Thul depleted while weaker gear triggered a town visit first. A fresh prepared boss check also exposed this. The corrected final comparison gives both baseline and candidate the same deliberate boss recovery. Earlier H10/B11/B12 and final rows are retained as development evidence, not mixed into the final estimate. No post-holdout parameter adjustment was made.

Method and limits

Final final-v2: 500 samples per controlled cell, seeds 1093000–1093499; 105,000 full-route attempts, 19,500 individual prepared encounter trials, and 19,412 legal campaign expedition segments. Matched baseline baseline-v2: 200 samples, seeds 1093000–1093199. Development screens used 1024230 onward (H10 200; B11–B14 120). All use the real rules classes and actual authored route, combat and reward definitions. No damage-only surrogate is used.

Three policies: basic attacks with legal recovery; reasonable tactics prioritizing wounds/skills; informed tactics using threat/control, spells, healing and known encounter mechanics. These are deterministic heuristics with seeded combat, not optimal play. Prepared profiles additionally choose owned attunements from actual weaknesses/resistances and rest before final bosses. The updated potion policy selects injured living allies. The one-use wishing well is disabled in every recovery. Invalid travel or simulation round limits fail the run; no bound or wish grant was reached in final evidence.

Controlled profiles are synthetic entry states, not claims of campaign affordability:

Relevant spells and artifacts are supplied to controlled profiles; campaign paths instead learn/buy/earn them. Parties: founders (Alaric, Markovia, Vestin, Mystere), reserves (Thoradin, Markovia, Vestin, Spargle), recruits (Xov, Markovia, Mystere, Juniper). These are three compositions, not all 70 four-hero combinations. Basic/Advanced controlled screens use founders; Expert+ includes all three. Current abilities, affinities, status immunity, action costs, enemy schedules, healing, armor, criticals, depletion and town costs run in production rules.

Route attempts stop at the first party wipe; individual deaths and resurrection are counted separately. Legal campaigns allow at most two recovery retries per failed location and stop if recovery is unaffordable. Required routes use the short Woods branch and omit the optional Owlbear Den; individual encounter screens include that den and all ordinary encounters. Known puzzle solutions and specialist challenge choices favor informed play. Living reserves do not receive free XP. The third traversal preview and its Reputation rewards are excluded from these campaigns.

Town policy responds to mean HP below 72%, any hero below 42%, deaths, or mean resource below 35%; prepared profiles also pay travel for attunement changes and fully recover before final bosses. Camps are once per run. Town counts therefore combine recovery and preparation. Campaign spending prioritizes spells/upgrades, preserving modest recovery funds, and retains the earlier Void/Frost armor-purchase heuristic rather than dynamically rebuilding armor per adventure. Fixed-stock and campaign results are not interchangeable.

Wilson 95% intervals quantify sampling error only. Paired seed prefixes do not imply identical subsequent dice after decisions diverge. Policy quality, full knowledge, target selection, party composition, optional-route choices and human execution add unquantified uncertainty. Five hundred successes do not establish invulnerability. Combat duration below is total rounds across attempted encounters, not wall-clock time.

Final route results

All rows below use founders and the informed policy, n=500 each. Different rows use the stated intended preparation, so this is not a same-party tier progression. All baseline and final summary cells are provided in the evidence package.

The corresponding matched baseline completion rates were 100% for every listed row except Heart Master+ at 91%. The largest change is late-game threat and deaths, not Basic access. Maximum gear still completes Expert/Master almost effortlessly; expected growth remains strong.

These averages include failures and omit any recovery after a terminal wipe. Synthetic initial inventory is not charged to that attempt; depleted stock therefore matters alongside net gold. Campaign rows include actual restocking and purchases. No reward inflation was justified in this pass: earned-reward campaign paths can support continued preparation, and positive fixed-run proceeds do not imply free resurrection after every wipe. Smoke use was zero in these founder-policy rows; the model often preferred Shield Wall, so this is not evidence that every consumable is useful or balanced.

Legal campaign growth and replay

The main path is Woods Basic → Heart Basic → Woods Advanced → Heart Advanced → both Expert → both Master → both Master+. Recruitment/unlocks/rewards are actual game operations; no clears, gear, money or levels are granted. Repeated-Advanced variants insert two or four additional Woods+Heart circuits before Expert. The original story prerequisites remain in force. Every final candidate launch ran with the ordinary travel/unlock guards enabled.

After the first Basic pair, founders reach roughly level 4; after the first Advanced pair, roughly level 7. Two extra Advanced circuits give approximately level 10 at Expert entry; four give level 12 for all 500 Companies. There is no permanent progression gap between Advanced and Expert. Repeat runs remain available even after attempting Expert, with confirmation before replacing that book's unfinished route.

Direct Expert has 100% Woods / 95.2% Heart conditional completion with bounded recovery retries, but only 50.2% / 32.4% deathless completion; it is substantially less forgiving than the old near-deathless campaign. Players who prepare through extra Advanced runs reduce that risk. Reaching level 12 alone does not supply maximum equipment: direct Master+ remains a poor bet, and gear preparation still matters. Read campaign rows with their surviving sample count; a later 100% is not automatically 100% of the original Companies.

Company Reputation is unchanged and repeats continue to award it. As more books become required, new Companies naturally enter later tiers with more XP, gold and Rank; permanent unlocks avoid making existing Companies redo earned access. The model therefore reduces early sequence variance but does not permanently solve power variance. Revisit required-set size, rewards and equipment costs with each content-ready book. The earlier ten-adventure report remains an accounting proposal, not new shipped XP budgets.

Prioritized remaining issues and tester acceptance

1. Composition gap, especially recruits. Excellent level-12 informed Master+ completion: Woods founders/reserves/recruits 52.6%/32.4%/23.8%; Heart 58.2%/36.6%/9.4%. The policy is not an optimal solver for Juniper/Xov, but this gap is too large to dismiss. Next pass should compare their action/MP economy and policies before globally weakening bosses or declaring recruit viability solved.
2. Master equipment threshold. Typical level-12 gear completes Woods Master only 0.6%, versus 100% with the prepared strong profile. That comparison also adds attunements, wards, funds and deliberate boss recovery, so it does not isolate one armor rank. Explain preparation clearly and collect actual equipment/party/consumable data with failure reports. Avoid using level alone as a power score.
3. Basic attacks still work too often. Ordinary Advanced basic-attack completion is 89.2% Woods / 64.8% Heart, but deathless rates are only 17%/10.4%. Typical level-12 Expert attack-only completion is 82.8%/63.6%, with 12.2%/2.8% deathless. Healing/recovery outside combat remains available to that policy. The vision of ability use is improved but not enforced; do not raise every enemy until these rates collapse at the expense of reasonable parties.
4. Town and boss preparation. Heart Master/Master+ prepared routes average roughly seven visits for five required fights, partly to change elements. Test whether preparation feels rewarding or like menu work. The correction to pre-boss recovery shows how misleading a generic health threshold can be. Consider clearer boss warnings or more convenient owned-attunement selection before adding travel costs.
5. Preview Reputation farming. Lost Dungeons has no XP/combat yet but retains normal completion Reputation. Lower-tier replay now makes that preview loop easier to repeat. It is excluded from the XP/campaign evidence and required clears. Keep this explicit during internal testing; disable or redesign preview completion rewards before treating Rank progression as final production balance.

Suggested playtest targets, conditional on the stated preparation: Basic tactical completion ≥90%, deathless ≥70%; Advanced tactical completion 85–99%, with clearly more deaths/resource use when attacking only; Expert near-cap typical completion 80–95%, materially more ability/consumable use and recovery than Advanced; Master strong-preparation completion 85–98% with occasional deaths; Master+ excellent preparation roughly 45–60% completion and <25% deathless. The current founder Master+ screen is close. Heart Expert completion and Woods Master deathless rates remain above these provisional ranges. A useful next acceptance gate is no representative well-played Master+ composition below 30%; this pass fails that gate for recruits.

Keep changes small for the next round: collect party level/gear/wards, abilities used, boss entry HP/MP, deaths versus wipes, consumables, town visits and reasons for quitting. Retest one changed tier/profile or creature adjustment using the reusable screen, then validate with fresh seeds. Do not use another blanket scaling increase to address a single boss, a fixed-output spell or a party-specific action deficit. Strict level gates, power scaling, reward caps and ability redesigns need their own evidence and migration decisions.

Reproduction and verification

See the runner README and evidence manifests, inputs, raw trials and summaries. The quick screen uses 120 seeds; the final screen uses 500. Overrides live only in the snapshot when supplied with -Config.

& Verification/BalanceCheckpoint/Build10/Run.ps1 -Output Logs/Balance-Repeat -Samples 500 -Seed 1093000
python Verification/BalanceCheckpoint/Build10/Summarize.py Logs/Balance-Repeat
& Verification/RunChecks.ps1
& Verification/CompileUnity.ps1

The 38 pure suites passed, including 44 new difficulty/migration/combat-profile assertions. Existing fixtures that specifically test tier reset, attire or ledger behavior now seed earned eligibility; actual all-book gating is checked separately. The obsolete rule that a mostly-basic-attacking Expert party must win half its boss fights was removed from the old build smoke test; fight termination still runs, and viability is reported through this broader live-rule screen.

Runtime and Editor compilation passed. Isolated Unity Play checks passed 60 introduction, recommendation, replay, cancellation/replacement and actual JsonUtility assertions over 4:3, 16:9, 21:9 and 32:9, along with the existing Practice cycles and simulated touch checks. Unity JSON/disk/backup recovery passed, followed by 770 full campaign storage, migration, isolation and live UI checks. Captures of both introductions, the selector and replacement confirmation were visually reviewed. Initial fixture setup failed because it omitted the Practice travel flag; it was corrected. Saved-profile normalization reuses the catalog's enemy validation so JsonUtility's omitted inline objects cannot introduce phantom abilities. Canonical handbooks, current subsystem guides, root instructions and Build 10 notes are synchronized with the gameplay diff.

No Windows/iOS/tvOS player build, physical controller/touch pass, human playtest, new-adventure combat test or optimal-policy proof is claimed. All live verification used its own project and isolated data; the user's Editor and player saves were preserved.