How to read the numbers

Weights, conditional shares, stage gates and expected counts before fit: what each number on the Terminal State loot pages means.

This page covers loot. For enemy spawn counts, family shares and room modifiers, see How to read enemy numbers.

From table row to item in the world

Every loot holder (a desk top, a cabinet, a toilet stall) resolves to one table in its level and room, then passes these stages once, when its contents are first rolled. The level pages show each stage for each holder context.

  1. Empty? The holder is empty with its empty chance. P(not empty) = 1 − empty. exact
  2. How many lots. A lot count n is drawn uniformly from the lot range, so E[n] = (lo + hi) / 2. exact
  3. Which item per lot. Each lot picks one row of the table with probability weight / Σ weights: the conditional share. Plain rooms first pick a category, then an item. conditional
  4. Stack size. Ammo boxes and handfuls get their round counts from the box and handful ranges.
  5. Scarcity. Ammo is kept exactly as rolled. Every other find survives with the theme's keep chance and may be re-themed to an on-theme or off-theme category, limited by what the holder may hold. conditional
  6. Limb treatments. A share of healing finds become limb treatments.
  7. Fit. Each item must fit on a real surface of the holder. An item that fits nowhere is dropped. not derivable
  8. Finite and persistent. Contents are rolled once, remembered, and never refilled or re-rolled.

The “expected units” and “chance of at least one” figures on this site are exact for stages 1–6 under the game's own numbers. They are measured before fit, so they are upper bounds on what is actually placed.

Weight is not a chance

A row's weight only matters relative to the other rows of the SAME table. Changing one weight shifts draws between that table's rows; it says nothing about how often the table is drawn at all.

Share = conditional selection chance

share = weight / Σ weights of that table (the denominator is shown). It is the chance that ONE lot drawn from that table, in that context, is that item — given that a lot is drawn. It is not the chance a holder contains the item and not a chance per level.

Stage gates of a holder

Each holder rolls once: empty? (empty chance) → lot count n ~ U{lo..hi} (E[n] = (lo + hi) / 2) → per lot an item by share (+ stack size) → scarcity: ammo kept exactly; any other entry survives with the theme's keep chance, then is on-theme (the theme's chance) or off-theme and may be re-themed within the holder's allow list → limb post-pass (some bandages / first-aid kits become limb treatments) → fit / occupancy on the real surfaces (items that fit nowhere are dropped). Contents are finite, rolled once and remembered; nothing refills.

Expected units before fit (exact)

Expected lots / units per holder are computed analytically from the same values the roll functions read (exact under ideal uniform draws) and checked against the real roll functions in the test suite. They are BEFORE fit / placement, so they are upper bounds of what ends up placed.

P(>= 1) per holder

The chance that ONE holder of that kind in that context rolls at least one lot of the item (before fit). Lots are independent given their count, so P = (1 - empty) x (1 - E_n[(1 - y)^n]), y = the per-lot chance after every stage.

Not derivable here

Fit / occupancy (which items fit which surfaces), placement rejection of floor finds and how many holders of a kind a level generates are generator outcomes: they are labelled "not directly derivable" and never folded into a chance. No per-level "chance to find X" is shown, except for the level-wide rolls (rare gear, masks) whose single roll per level is exact before placement.

Sample census

Which holder contexts occur in a level comes from a small fixed-seed census of the real level generators ("observed in N of M sample levels"). It is a sample, not a guarantee; declared contexts that no sample placed are listed as "declared".

Tuning knobs

Every value lists where it lives (file + export) and which per-level override reaches it: itemWeight / categoryWeight (row and category weights), keep (scarcity survival), slot.empty / slot.lots (per slot category), floor.chance / floor.count (loose finds), gear.mean / gear.cap / gear.weight (rare gear), rewards (replace a reward table), tables (give the profile its own rate tables), none (edit the source).

Source labels

How each loot source is labelled
LabelMeaning
GuaranteedAlways yields its stated contents (the starter kit, the roof cache picks). Fit can still matter where noted.
ConditionalTable-driven and gated by chances: empty chance, lot count, per-lot pick, scarcity, fit. Most holders and floor finds.
Special ruleAn explicit rule outside the holder tables, such as rare gear or weapon finds with their paired ammo.
Non-lootRolls nothing new. Fixed fixture contents, or rules that only move or release existing finite lots (pry, spill, detach).
Decoration onlyLooks like stock but is not an item you can take (for example shelf filler).

Override knobs

Each value on this site names the source file and export where it lives, so feedback can point at the exact knob. A level's profile can also scale values without editing the shared tables, through these override knobs:

Per-level override knobs (LootOverrides in the level profile)
KnobWhat it changes
itemWeightper-item weight multiplier
categoryWeightper-category weight multiplier
keepscarcity keep multiplier
slot.emptyslot empty-chance override
slot.lotsslot lot-range override
floor.chancefloor find chance multiplier
floor.countfloor find count multiplier
gear.meanrare gear mean
gear.caprare gear cap
gear.weightrare gear weights
rewardsreplace the reward tables
tablesgive the profile its own rate tables
noneno override reaches it: edit the source value

Lowering a row's weight moves draws to the other rows of the same table. Lowering a keep chance removes finds without moving them anywhere. Lowering an empty chance or raising a lot range adds finds to every holder of that kind.

What this site never states, and why