Odds Lab
Measured odds: Cube Greed, Vorax, TOWER, Tartarus
The game publishes few of these chances. Those in its own tables are shown as the tables give them, next to what we measured and what the advisors play, each with its source and date. The rest are counted from what players share with TLI Toolkit, each with a 90% range and the number our calculators assume. A counted number shows only once enough players and outcomes are in.
— of 10 sets measured so far
Measured odds need enough outcomes and players to meet publication requirements. The bar shows published progress when available.
Loading the measured odds…
Greed chance
Every Greed the game log showed while a shared map was on: won (the rewards doubled) or lost (the Cube broke).
Measured odds replace the assumptions once there are at least 100 Greeds and the player and publication requirements are met.
What the app assumes until then
| Outcome | Assumed |
|---|---|
| Greed succeeds | 50% |
Maps from the app add to this automatically (Settings → Community to turn it off).
Potion cards
Each card of every 3-card potion offer the Vorax advisor saw (refreshed offers too), counted once per offer.
300 cards needed The app plans with the game’s own offer weights.
Instruments in the first offer
In Vorax you pick one of 3 Surgical Instruments. This card counts the first offer of a run, which comes from the basic instruments: how often each one is among the 3, from the runs players share. The advisor uses these odds to judge what the next offers may bring.
100 offers needed The app plans with the game’s own offer weights.
Instruments in later offers
Every offer after the first comes from the advanced instruments. This card counts how often each one is among the 3 offered, from the runs players share. A later offer never repeats an instrument you already have; the numbers here count all later offers together. The advisor uses these odds to judge what the next offers may bring.
100 offers needed The app plans with the game’s own offer weights.
Runs from the app add to this automatically (Settings → Community to turn it off).
Chips offered per round: Basic 2‑2‑2, In-Depth 3‑2‑2‑2 (confirmed in the game).
Chip frequency, Basic
Each chip’s share of all chips offered; the model draws the 7 chips evenly (1 in 7 each).
Measured odds replace the assumptions once there are at least 300 chips and the player and publication requirements are met.
What the app assumes until then
| Outcome | Assumed |
|---|---|
| Chip 1 | 14.3% |
| Chip 2 | 14.3% |
| Chip 3 | 14.3% |
| Chip 4 | 14.3% |
| Chip 5 | 14.3% |
| Chip 6 | 14.3% |
| Chip 7 | 14.3% |
Chip frequency, In-Depth
Each chip’s share of all chips offered; the model draws the 7 chips evenly (1 in 7 each).
Measured odds replace the assumptions once there are at least 300 chips and the player and publication requirements are met.
What the app assumes until then
| Outcome | Assumed |
|---|---|
| Chip 1 | 14.3% |
| Chip 2 | 14.3% |
| Chip 3 | 14.3% |
| Chip 4 | 14.3% |
| Chip 5 | 14.3% |
| Chip 6 | 14.3% |
| Chip 7 | 14.3% |
Runs from the app add to this automatically (Settings → Community to turn it off).
Caches per Riot (log)
From the game log of shared Tartarus runs: every Warden’s Cache the Riot’s loot room opened, by its rarity, and the two special caches besides: the one at the start of the fight, and the Jailer’s, which drops only when you meet and kill the Jailer during the fight. No advisor journal needed.
30 Riots needed
Caches per Riot
From the advisor’s journals of runs that went to the Riot: the average number of caches of each rarity it opened.
30 Riots needed
FE per cache
What one more cache of a rarity adds to a Riot’s FE after tax, fitted over shared Tartarus runs that carry their caches. The advisor still uses value ratios you can set; these are FE.
60 Riots needed
FE by compass
Net FE after tax per Tartarus run, grouped by the compasses spent opening it (the Shackles compasses add drop quantity).
20 Riots needed
Runs from the app add to this automatically (Settings → Community to turn it off).
Vorax: the game’s tables against our measurements
From the game’s own config tables (read from the game files) next to what we counted in recorded games. Where they differ, both are shown. Last updated 2026-11-11.
| What | The game’s table | Measured | The advisor plays |
|---|---|---|---|
| Monster rarity (Normal / Magic / Rare / Boss) | 50 / 30 / 15 / 5% | 50 / 34 / 12 / 4% (recorded real games) | Measured |
| Potion card rarity (Normal / Magic / Rare / Supreme) | 40 / 44 / 15 / 1% | 40.4 / 44.3 / 15.3% among potions; Supreme (red) 1.2% a card, from about 1.48 million cards | Measured (red 1.2%) |
| Potion Kits among the cards | 4.5% | 4.1% | Measured |
| Rounds, rerolls, hand, vials | 13 rounds (11–13 deal no potions), 3 rerolls, a hand of 3, 6 vials | Same | The table |
| Monsters you start with | 3 | A fourth in 10% of runs | Measured (the table has no fourth) |
| Activity / Quantity by rarity (Normal, Magic, Rare, Boss) | 1 / 36, 5 / 24, 15 / 12, 300 / 1 | Same | The table |
The game’s own tables disagree with each other (the config says 50 / 30 / 15 / 5, the monster rows 46 / 40 / 10 / 4) and with the recorded games (50 / 34 / 12 / 4). The advisor plays the measured chances: the game’s server deals the cards, and a client table may not be what it uses. If a count of many real boards follows the table instead, the advisor changes with it. The red-card gap (1.2% against 1%) barely matters: with 0.74% red and no Xenopanacea the advisor reached 840k as often (79.0% against 79.5%, 600 games).
The Holiest Water on a Boss Agito gives +120 Activity: seen and confirmed in the game, and played that way.
Vorax: how often the advisor reaches 840k
Our engine, simulated games from the same start at the app’s settings (WebAssembly engine, early stop). This is the model’s chance, not real runs: those are under “Following the advisor” below. Last updated 2026-11-11.
| Setting | Reached 840k | Average final score | Games |
|---|---|---|---|
| The app (20,000 playouts), fixed games | 79.5% | 1.22M | 600 |
| The app, fresh games (new seeds) | 77.7% | 1.29M | 300 |
| 2,000 playouts | 73.0% | 1.17M | 600 |
| 6,000 playouts | 76.7% | 1.22M | 600 |
| 60,000 playouts | 79.3% | 1.25M | 600 |
More search pays up to the app’s 20,000 playouts and not beyond: three times as many gain nothing at 840k. The shown chance is calibrated to the advisor (fitted on 590 games and 10,375 decisions, scored on games it was not fitted on): the raw playout number ran 17.5 points off, the shown one 0.4.
Tartarus Prison: the game’s tables
From the season’s two Tartarus config tables, read in full from the game files. Every number the advisor plays that the tables give matches; the few it plays that they don’t are measured. Last updated 2026-11-11.
| What | The game’s table | The advisor plays |
|---|---|---|
| Board and run | 16 cells; the Interrogation Room is cell 0; a standard run is 20 days (30 with Iron Law) | Same |
| Destination cards a day | 3, on cells 1–6 ahead (a die) | Same |
| Rooms open at the start | 4 | 4 (one fewer in 5% of runs: measured, not in the table) |
| Door rarity (Normal / Magic / Rare / Legendary / Ultimate) | 45 / 35 / 10 / 8 / 2%, the same for every Code | Same; the first board measured 44.5 / 35 / 10.5 / 8 / 2% |
| A won fight (Consume) | +1 rarity tier | Same |
| Vault (Chromatic Inmates held) | Up to 10 | Same |
| A special theme (Tartarus Rumor) | 12% of runs | Same |
| Instruments offered at the start; Objectives | 3 to choose from; 1 Objective | Same |
| A Replace’s new Inmate (Normal … Ultimate) | not in the table | 42 / 34 / 11.5 / 9.5 / 2.5%, measured; never the replaced Code |
Tartarus Prison: the rules, read from the game’s texts
The incidents, Objectives, themes, Torture Instruments and treasures have no config table; what each does is in the game’s own card texts. All 76 were read (English, with Chinese, Korean and Russian checked for the same numbers) and set against the advisor’s rules: every number is the same. Last updated 2026-11-11.
| What | In the game’s texts |
|---|---|
| 17 Prison Incidents | Taste of Freedom opens 3 cells; Extra Portion raises up to 3 Inmates by 1 tier; Ravenous Appetite makes 5 tries at 50% to raise the lowest-rarity Inmate; Crow Whispers shows the next 4 cells; Usurpation I to IV escalate from +1 tier for one Inmate to +2 for all |
| 5 Objectives | An Inmate reaches Ultimate; Consume 4 times; open 10 cells; trigger 2 incidents; move 18 tiles. The reward is one more Torture Instrument pick (and, in the Riot, an Ultimate or Chromatic cache) |
| 4 themes (Rumors) | Tyranny: Instigate and Bestow 3 more times at the Room; Opulence: 2 more treasure choices; Iron Law: a 30-day run; Cage Fight: Inmates eat lower-rarity countered Inmates and keep their Code |
| 31 Torture Instruments | Chances run from 25% (Rusty Saw: +1 tier after a Consume) to 80% (Lost Ring of Keys: opens both neighbours after opening; Barbed Collar: +1 tier on a chained Consume; Warden’s Scales: shows the rarest Inmate after a Replace); 40% for Burning Stake and the Long-Chained Shackles (moves of 4+), 50% for Forged Key and the Short-Chained Shackles (moves of 3 or fewer) |
| 4 Warden’s Treasures | Three-Headed Thumb Ring opens up to 3 cells; Brass Horn of Strife has every Inmate with a target eat once; Crimson Vintage raises every Inmate below Legendary by 1 tier; Brand of Original Sin re-Codes up to 4 Inmates so they have a target |
One rule was wrong and is fixed: the Brass Horn’s bites also reach an Inmate behind Scratch Marks (the card’s own reading). It matters only in a run holding the Filthy Claws, where playing it right is worth about +0.17 utility a run (400 paired runs on each of two suites); in ordinary runs nothing measurable changes. The texts give no weights for incidents or instrument offers: those come from the client tables above and earlier readings of them.
Map modifiers: the Drop Quantity of each
From the game’s own map modifier table (4,368 rows, read from the game files). A modifier added at the map device carries a Drop Quantity of its own that its text doesn’t show. Last updated 2026-11-11.
| Modifier pool (ids) | Drop Quantity each | Rows |
|---|---|---|
| 5111xxx, 5112xxx, 5121xxx, 5122xxx, 5212xxx, 5311xxx, 5422xxx, 5521xxx; 3500205–3500214; 8002101–8002114 | +15% | 164 |
| 5634xxx, 5633xxx, 5690xxx; 3500100–3500123; 3500300–3500304 | +25% | 65 |
| 3500200–3500216 (6 rows, lowering ones), 3500400–3500404 | −30% | 11 |
| 5700000–5700015 (a pool of its own) | +42.5% and +7.5% additional | 8 |
| The rest (3500000–3500317, 8002000–8002010, 5710000) | none | 62 |
On Deep Space maps nine modifiers are listed, and their Drop Quantity added up to exactly +100% on 248 of the 249 map openings we counted (the other listed none), all with six modifiers added. Adding them costs Netherrealm Resonance, doubling each time (1, 2, 4, 8, 16, 32: six cost 63). Not settled: whether the server grants what the client’s table says, and what the +42.5% pool adds on Deep Space maps.
How to read it
- “Measured from N outcomes by M players, 90% range”: the true chance is inside that range nine times in ten. A wide range means few outcomes so far, not a different chance.
- “Assumed” is what the app’s calculators and advisors use today: a community figure or our model. When the measured range doesn’t include it, the assumption is probably off.
- Nothing is shown below 5 players. Each set also needs its own minimum of outcomes (listed with it); until then it says “not enough data yet”.
Following the advisor: how runs ended
From the Vorax and Tartarus advisor journals players share. A run counts as “followed” when at least 9 in 10 of its known moves were the advisor’s top move, with 10 or more moves known. Runs with fewer known moves, and journals that stop before the end of the run, aren’t counted; practice runs are never sent.
- Goal: 840k Ultimate Activity on the last board the advisor saw (round 10, before its card and the last three rounds), so the final number is at least that.
- Goal: 6 or more Chromatic Inmates contained at the end.
- Rows require at least 3 players and 10 runs; outcome numbers require at least 5 players. Publication also depends on privacy checks.
This is correlation, not a controlled test: players who follow the advice may play differently in other ways too. The advisor’s own chances are simulated; these are real runs as the journals saw them.
Vorax
Tartarus Prison
Memory Revival
Not measured: the game log doesn’t record what a revival rolled, so there is nothing to count. The calculator uses the chances from the patch notes (Wax and Wane 20%, Rising Moon 20%, targeted Lunar Phase 50%).
How to add your runs
In the app: Settings → Community. “Share my runs” sends your maps (the Cube Greeds come with them); “Share minigame runs” sends the Vorax and Tartarus advisor journals and the TOWER rounds the calculator read. Both are on by default from version 0.1.0-beta.30 (off until you turn them on in older versions), one click there turns either off, and “See exactly what’s sent” shows the next upload. No names, no account ids.
