# Cinder Vault design and simulation log Status: selected for a first paper playtest on 2026-08-16. This log records design iteration, reproducible pacing-model output, and two complete abstract walk-throughs. The model is not a Magic rules engine and its percentages are not empirical win rates. Paper games remain the authority. ## Inputs and boundary The four immutable inputs are the selected 60-card `planned.txt` lists for Mathise, Delina, Vrondiss, and Wyll with Inspiring Leader. The preserved pod review gives them identical 23-land structural bands, ordinary turns-seven-to- nine finishes, 14–16 explicit d20 cards, seven or more early development pieces, and zero current Game Changers or intentional deterministic combos. Exact-list probability and sequencing review established these useful bounds: - Raw two-through-four-land opening hands occur about 77% of the time before the free multiplayer mulligan. - By turn two, 70–77% have seen an early development card and 80–89% have seen at least one interaction or protection card. - Across the party, realistic live interaction is roughly three to six cards by turn three and four to eight by turn five. - Gross attack potential moves from approximately 6–20 in round four to 30–80 in round seven. Ordinary blockers reduce those bands substantially. - A target near 110–130 combined Vitality and blocking pressure is appropriate for a round-eight median attempt, provided excess damage cannot skip stages. The local model samples those envelopes, development stumbles and surges, Encounter pressure metadata, public Reserve cadence, collective blocks, interaction budgets, vitality gates, Doom, and Room pressure. It does not simulate individual cards, mana colors, stack ordering, or exact combat. Reproduce the selected rules-closed comparison from the repository root. The earlier loop percentages below are historical outputs from superseded timing models and are retained to show why the design changed. ```powershell python .\sites\d20-adventurers-pod\scripts\simulate-dungeon.py ` --runs 10000 --seed 1729 ` --revision latest --fail-on-target ``` ## Loop 1 — hostile Archenemy skeleton Initial contract: - 50 shared life. - Collapse at Doom 8. - Every Dungeon turn included a full creature combat. - Adventurers could not block for one another. - Room setup added immediate pressure. The first small batch was effectively unwinnable: the party commonly died in round three before any five-mana commander or substantial combat engine could matter. This was an action-economy failure, not a reason to shave one damage from individual cards. The structural correction retained two Dungeon casting windows per round but made only the second a full assault. That gives Mathise and other interactive decks two stack decisions without doubling every Monster's damage. The party also gained collective blocks, 60 shared life, and the Doom-10 clock. The historical hostile-loop comparison produced only a **1.3%** clear rate and **8.2%** boss reach, with a round-five median ending. It is rejected. ## Loop 2 — cooperative party raid Second contract: - 60 shared life and Doom 10. - Two Dungeon turns but one Dungeon combat per round. - Any adventurer creature may block the assault on Focus. - 115 gated Dungeon Vitality: 20 Gate, 25 Hall, 30 Furnace, and two 20-point boss phases. - A real 30-card Encounter library and face-up four-card Reserve. - Monsters, Hazards, and Tactics use normal Magic zones and interactions; Rooms and gates remain scenario objects. This solved the premature-kill problem and preserved Wand of Wonder, Aetherize, Wild Magic Surge, Wizard's Spellbook, theft, sweepers, counterspells, and Ancient Dragon combat triggers. The initial global shuffle, however, allowed late Furnace or boss threats on turn one and created high-variance early collapses. The next pass added `minimum_stage` to every Encounter. A global shuffle then overcorrected: several future threats could clog Reserve and produce empty Dungeon turns. A pre-stacked chapter attempt avoided the clog but buried the boss chapter beneath twenty earlier cards. Loop 4 retains separate face-down chapters and places each new chapter on top only when its stage opens. Room-generated Minions also received population checks. The Gate stops making Cinderlings at three Dungeon creatures, the Hall stops making Reflections at four, and the boss converts a full board's spawn into a temporary combat bonus. This prevents two lair rolls per round from becoming an unbounded army. With one setup-damage packet still enabled, the historical 10,000-run loop-2 model produced: | Metric | Result | | --- | ---: | | Clear rate | 60.4% | | Boss reached | 72.2% | | Clear before round 6 | 0.0% | | Median ending round | 9 | | Median life in successful runs | 15.5 | | Median Dungeon Vitality in failed runs | 70 | This is playable, but setup damage is narratively flat and punishes a new room before the party can read it. ## Loop 3 — provisional first-playtest contract This pass added no-damage setup, one assault per round, vitality gates, a targetable Heartbound boss, public Reserve, and deterministic target and block orders. Its original 10,000-run output appeared to pass at 73.2% clears and 80.5% boss reach. It was rejected after exact card and timing review. Unopened boss cards were buried below twenty earlier cards; temporary combat Tactics could be spent in the no-combat window; newly cast and tapped creatures attacked too early in the model; Heartbound reset every turn; three mandatory-target spells could not legally reach their fallbacks; and Wyll's land catch-up package was blank against a landless Dungeon. The passing percentage was therefore not a valid paper-rules baseline. ## Loop 4 — selected rules-closed first-playtest contract The fourth pass closes those gaps: - Stage 1 alone starts as the library. Each later chapter is shuffled onto the top when its stage opens, so the Furnace and boss actually use their cards. - Mirror the Vanguard, Claim the Vanguard, and No Safe Ground are castable by the automa only in the second Dungeon turn. Creatures obey tapped status, haste, summoning timing, and attack tapping in the model. - Mandatory-target fallbacks are real printed modes. Encounter cards have generic mana costs and color characteristics, so player-cast effects work. - Heartbound removes 8 Vitality only once per expedition round and the boss phase exchange has explicit new-object, attachment, and attack timing. - Four starting Depth land tokens and one new Depth per round keep Wyll's white catch-up package live without virtual opponents or usable Dungeon mana. - The Dungeon remains one opponent. Myriad, melee, and Veteran Soldier are an explicit first-paper-test watch rather than a global rules patch. The rules-closed 10,000-run comparison, seed 1729: | Metric | First-playtest target | Result | | --- | ---: | ---: | | Clear rate | 55–75% | 73.4% | | Boss reached | at least 80% | 81.1% | | Clear before round 6 | at most 5% | 0.0% | | Median ending round | 7–9 | 8 | | Median life in successful runs | at least 8 | 16.9 | | Average Mathise answers | observation only | 2.89 | | Average Delina answers | observation only | 2.19 | | Average Vrondiss answers | observation only | 2.02 | | Average Wyll answers | observation only | 2.30 | | Gated Vitality share: Mathise | 15–35% watch band | 19.2% | | Gated Vitality share: Delina | 15–35% watch band | 29.0% | | Gated Vitality share: Vrondiss | 15–35% watch band | 21.1% | | Gated Vitality share: Wyll | 15–35% watch band | 30.6% | This is the selected first paper baseline. Both clear and boss-reach rates sit near their acceptance boundaries, which is useful: the first real game can move either result meaningfully. The model's 1.29 pressure calibration converts coarse Encounter values into shared-life loss; it is not a hidden multiplier players apply at the table. ### Local simulation engine v3 audit The selected result above was rerun after strengthening the local simulator without adding hands, mana, or a Magic rules engine. Version 3 now records the actual Focus adventurer, applies delayed disruption to that profile, returns destroyed Encounter permanents to the Dungeon graveyard, recycles that graveyard into an empty library, and exposes structured event traces, per-Encounter activity, gate distributions, loss causes, table-load metrics, confidence intervals, paired-seed parameter sweeps, and per-run JSON Lines. Across the same 10,000 seeds, the clear rate is **73.4%** with a **72.6–74.3%** 95% Wilson interval. Bound Warden is reached in **81.1%** of runs and Crown Unbound in **78.7%**. Party-Life defeat accounts for **26.5%** of all runs; only seven runs reach a Doom loss in this abstraction. The median gate rounds remain 5, 6, 7, 8, and 8. The maximum visible Reserve is three in every baseline run because the model does not synthesize adventurer-caused draws, bounces, or unusual copy chains. The maximum modeled non-Depth permanent count averages 6.29 and has a 90th percentile of nine after including setup, lair, boss, and grouped token objects. With the ordinary Depth lands included, total Dungeon permanents peak at 15.08 on average and 17 at the 90th percentile. The Encounter library averages 0.40 graveyard recycles per run. Reserve overflow and cleanup therefore remain explicit paper-play observations rather than invented probabilities. See `simulation-guide.md` for commands and interpretation. A one-variable sensitivity pass used 5,000 common seeds per value. Each cell below is clear rate / boss reach; these are diagnostics, not proposed rule changes. | Layer | Lower | Baseline | Upper | | --- | ---: | ---: | ---: | | Dungeon pressure 1.22 / 1.29 / 1.36 | 78.4% / 84.7% | 73.4% / 80.9% | 67.2% / 77.0% | | Party attack 0.95 / 1.00 / 1.05 | 71.6% / 79.9% | 73.4% / 80.9% | 74.1% / 81.4% | | Party defense 0.90 / 1.00 / 1.10 | 69.9% / 78.8% | 73.4% / 80.9% | 76.3% / 83.1% | | Answer rate 0.90 / 1.00 / 1.10 | 66.0% / 75.3% | 73.4% / 80.9% | 79.3% / 85.1% | Answer availability and collective defense move the result much more than a five-percent change in raw party attack. The first paper logs should therefore record which threats were answered, which potential answers were unavailable, and how much collective blocking prevented before treating end-of-game damage as a tuning diagnosis. ### Opening tempo and starting-Treasure probe The separate opening helper sampled 10,000 paired openings per deck with seed 1729. No-Treasure, one-Treasure, and two-Treasure versions of each seed use the same shuffled library, London-mulligan decision, opening hand, normal draws, and named die-result streams. Extra roll modifiers can still change which die is kept. The helper uses the exact selected lists and reviewed early mana and signature abstractions; it is still a characterization model, not an empirical game result or a complete Magic engine. The most useful comparison is the chance that each commander's characteristic engine has happened by the end of turn three / turn four: | Deck | No Treasure | One Treasure | Two Treasures | Bodies on T3 / T4 with two | | --- | ---: | ---: | ---: | ---: | | Mathise | 14.2% / 79.6% | 57.3% / 86.0% | 62.4% / 87.0% | 1.38 / 1.64 | | Delina | 1.2% / 14.5% | 5.3% / 41.8% | 15.4% / 64.2% | 1.73 / 2.34 | | Vrondiss | 2.1% / 16.4% | 7.0% / 39.1% | 13.4% / 59.8% | 1.41 / 2.67 | | Wyll | 50.4% / 72.0% | 62.1% / 80.3% | 66.1% / 83.7% | 2.08 / 2.74 | Two Treasures also put Delina and Vrondiss onto the battlefield by turn three in 77.4% and 82.6% of openings, compared with 19.5% and 16.0% without them. Wyll is cast on turn one in every modeled two-Treasure opening; Mathise is cast on turn one in 81.9%. The decks spend an average of 1.83–1.94 of the two starting Treasures by turn five, so the second token is not merely insurance. This rejects two Treasures as a neutral default for the first paper test. They materially replace the pod's early development game, create substantially more turn-three and turn-four board presence, and would require a new whole-Dungeon pressure calibration. One Treasure is a useful later fast-start candidate, but it remains asymmetric: Mathise and Wyll convert it into especially early commander turns, while Delina and Vrondiss receive the larger turn-four signature improvement. The selected first-playtest baseline therefore remains zero starting Treasures; record real opening hands before revisiting the variant. ### Selected opening fiction — the ambush The first paper test keeps both **zero starting Treasures** and the existing round-one assault. That means the adventurers may have only a land or mana piece when the second Dungeon turn sends its ready creatures at Focus. This is intentional rather than an overlooked grace period: the party entered the vault underprepared, did not expect this level of danger, and is surprised by enemies before it can establish a formation. The shared 60 life is part of the party's opening resource; taking damage before presenting blockers establishes the Dungeon as the aggressor. Do not soften the opening merely because the first attack connects. During the paper test, record the round-one combat and noncombat damage separately, which Encounter and lair results contributed, what one-mana responses were actually available, and whether each player describes the surprise as exciting, tense, or frustrating. Revisit a round-one grace rule only if the ambush repeatedly removes meaningful choices rather than creating urgency. ## Complete abstract walk-through A — controlled clear Reproduce with `--trace-seed 2`. This uses disclosed role envelopes rather than a card-by-card Magic engine. - **Rounds 1–3:** Cinderlings and the Gate consume early answers. The single assault each round cuts Party Life to 27.5 while the party chips only 4.8 Vitality, showing that the opening still demands real development. - **Round 4:** Wyll, Mathise, Delina, and Vrondiss remove setup bodies and push the Dungeon to 96.3. The party prevents all combat damage this round. - **Round 5:** Mathise reaches the 95 gate. The Hall opens immediately; Vrondiss and Wyll push through 14 more before persistent Chains deal the only damage. - **Round 6:** Delina and Vrondiss reach 70. The newly opened Furnace defender is tapped and cannot erase their window. Wyll and Mathise reduce it to 54.4. - **Round 7:** Vrondiss and Wyll break the Furnace. Mathise and Delina push the first boss phase to the 20 gate. Feed the Heart and the Heart's pulse leave the party at 14.6 life after Crown Unbound enters. - **Round 8:** Wyll crosses the final gate. The party wins at **14.6 life**, Doom 7, with gate rounds 5, 6, 7, 7, and 8. ## Complete abstract walk-through B — boss collapse Reproduce with `--trace-seed 14`. - **Rounds 1–4:** Gatekeepers, Cinderlings, and two large assaults reduce Party Life to 7.3. The party has removed several defenders but the Gate remains at 108.3 Vitality. - **Round 5:** All four adventurers contribute damage and Wyll reaches the 95 gate. The Hall opens with only 6 Party Life after persistent pressure. - **Round 6:** Delina and Vrondiss advance the Hall; Wyll reaches 70. Furnace Heat and a surviving persistent effect leave 2.1 life. - **Round 7:** Vrondiss and Wyll break the Furnace. Mathise and Delina push Bound Warden to 23.2, including the round's single 8-point Heartbound loss. Scorch the Ranks then deals the lethal packet before the assault. The party loses at **0 life**, Doom 8, with Vhal-Kez visible at 23.2 Vitality. Gate rounds are 5, 6, and 7. This is a useful failure: every Room is seen, the boss is wounded, and one prevention or lifegain spell would change the line. ## Scripted actual-card compatibility walk-through This branch is deliberately constructed to exercise rules collisions; it is not included in the model percentages. 1. Mathise counters an Encounter Tactic with Arcane Denial. The Tactic goes to the Dungeon graveyard, Mathise gets the delayed draw, and the Dungeon's two delayed draws enter Reserve face up rather than resolving immediately. 2. Delina uses Firbolg Flutist to borrow a normal Monster, then targets that controlled nonlegendary Monster with Delina and attacks with both during the same turn. Heartbound prevents using this line on Vhal-Kez. 3. Vrondiss resolves a printed d20 instruction. The Dungeon roll does not trigger Vrondiss, but Vrondiss's own roll does; its self-damage produces one Dragon Spirit. The Spirit may block collectively for another focused hero. 4. Wyll points Wyll's Reversal at Stonefall. Targets are announced before the roll. A successful redirect can point Stonefall at the Dungeon; a copied Stonefall is also controlled by Wyll's pilot and uses normal targets. 5. The first Dungeon turn creates a tapped Depth without playing it. Deep Gnome Terramancer and Archaeomancer's Map may trigger normally; Knight of the White Orchid and Battle Angels of Tyr count the Depth tokens already in play. 6. Mathise casts Aetherize during the one Dungeon assault. Attacking Encounter Monsters return to Reserve face up, then the oldest cards beyond four are discarded during cleanup. 7. Delina activates Wand of Wonder against the chaptered Encounter library. It finds Tactics because they are real instants or sorceries. Unopened chapters are not yet in the library; after a chapter opens, the adventurer may cast its exposed Tactics normally. 8. Vrondiss uses Beast Within on a Hazard and Wyll casts Slash the Ranks. Encounter permanents are affected normally; the Room, gates, and trackers are untouched. The resulting Beast remains a normal Dungeon-controlled token. 9. Wyll uses Swords to Plowshares on Vhal-Kez. Heartbound replaces only the zone change, taps the boss, and removes 8 gated Vitality. Swords continues resolving, so the Dungeon gains life equal to Vhal-Kez's power up to the current stage ceiling. Other spells likewise resolve nonconditional text such as creating a token, manifesting, rolling, or revealing cards. 10. Mathise's Bar the Gate counters a Monster spell and ventures into an official Magic dungeon. It never advances a Living Dungeon Room. 11. The Deck of Many Things returns a Dungeon Monster. If its delayed text would make the Dungeon's owner lose, the scenario replaces that loss with 10 gated Vitality damage. No collision above requires an emergency object-type exception. ## First paper test acceptance sheet The first playtest succeeds as a rules prototype if all of these are true: - Setup takes no more than 12 minutes. - Each Dungeon turn takes no more than three minutes. - Round-one combat and noncombat damage are recorded separately, and every player characterizes the opening ambush as exciting, tense, or frustrating. - No target, block, mode, sacrifice, Reserve, gate, or Heartbound decision requires an improvised rule. - The game reaches the boss by rounds 6–8 and ends by round 10. - Every deck makes at least one contribution beyond raw combat damage by the end of round 5. - Mathise cannot answer every important threat, but at least two Encounter spells or abilities are meaningful targets. - Delina can borrow or copy a normal Monster without copying the boss. - Vrondiss creates and uses at least two Dragon Spirits without triggering from Dungeon rolls. - Wyll's one-opponent scaling feels at least 70% as useful as the other decks' signature contribution. - Clarity, agency, and character identity average at least 4/5; administrative burden averages no more than 2/5, where lower is better. Record Party Life, Vitality, Doom, each gate round, every dead Reserve card, every ambiguous automa choice, schemes answered by each hero, the first meaningful printed d20 turn, and a one-sentence high point for every pilot. ## Tuning order after paper play Change only one layer between games: 1. If the party dies before the Hall, remove one stage-1 Gatekeeper copy from the Encounter deck and add one Mirror Shade to stage 2. 2. If the party repeatedly reaches Doom 8 before the Furnace, reduce the Gate from 20 to 17 Vitality and the Hall from 25 to 22. 3. If the boss is reached with more than 35 Party Life, increase Furnace Heat's cap from 4 to 5 before adding Monster power. 4. If Vhal-Kez dies in one adventurer turn per phase, raise each boss phase from 20 to 25 while preserving both gates. 5. If Mathise suppresses more than 70% of Encounter casts, reduce starting Reserve answers or add one “cannot be countered” line to a single boss-stage Tactic—never to the Room itself. 6. If Wyll is materially behind, first test the narrow raid rule “myriad and Veteran Soldier use raid count 3.” Do not make the Dungeon three opponents for every `each opponent` effect. Do not change the four deck lists during the first scenario test. Mode balance and deck balance are separate questions.