# Vorax Surgical Instrument tiers

The score measures how much final Ultimate Activity a newly equipped instrument adds to a fixed board on average. Use it to compare effects when their conditions can be met. The experiment gives every instrument the same boards and ignores its offer frequency.

## Reproduce the table

From the repository root, run `npx --no-install vite-node scripts/vorax/instrumentTiers.ts`. It writes `website/src/guides/data/vorax-instrument-tiers.json`. Inputs are the committed Vorax WASM engine, its default live odds, instrument definitions and built-in value policy. Every continuation runs to the end of round 13. The generator fails if a WASM operation falls back to TypeScript. No logs from players, external rankings, network calls or time-based stopping criteria enter the calculation.

## Board set and comparison

1. Play 24 synthetic games using the default settings and policy. The RNG seed is `mix(501, i)` for game indices 0 through 23. Take the first ordinary potion hand in rounds 3, 6 and 9, and the idle phase in round 11, after resolving any instrument offer. This samples early growth, mid-game, the last potion rounds and passive triggers. The four snapshots within one game are related.
2. Keep the board, hand and remaining refreshes from each snapshot, including stats accumulated by the original build. Clear its instruments and pending picks. Mark both thresholds passed so future instrument offers cannot affect the comparison. Equip the candidate immediately with its activation clock starting at the snapshot round. The baseline equips nothing.
3. For each snapshot and loadout, let the value policy choose the first action with seed `mix(gameSeed, round)`. Run the WASM engine's paired playouts with that same seed, future indices 0 through 11. This aligns the round and effect random streams across all candidates and the baseline. The policy may choose different potions as the board changes.
4. Subtract the empty-loadout final UA for the same snapshot and future. Average all 1,152 paired differences with equal weight. The JSON rounds scores to whole UA. The tier uses the unrounded mean. S starts at +100,000 UA, A at +50,000, B at +15,000, and C includes everything below +15,000. These are editorial cutoffs on absolute gain.

## Combinations and confidence

Take the four highest-scoring basic and seven highest-scoring advanced instruments, breaking single-score ties by lower ID. Test all 49 distinct pairs with at most one basic, and all 84 triples with one basic and two advanced. The advanced-only pairs measure their contribution in isolation. The shortlist excludes duplicate stacks and can miss strong interactions between individually weak instruments.

Synergy equals the combination's mean gain over empty minus the sum of its members' individual mean gains over empty. Use the unrounded values for this subtraction. Keep up to four pairs and four triples with positive gain and positive synergy, ordered within each size by synergy. Triple synergy includes its pair interactions. Each whyKey identifies a plausible mechanical link from the engine rules. For example, Cortex can create the Rare Agito required by Pituitary. Fusion preserves existing stat gains, including buffs previously applied by Sprout. Fusion itself does not trigger Sprout.

Each row's positiveBoards counts snapshots where its average gain over empty was strictly positive across the 12 futures. Read it out of 96 as a consistency note. For combinations it describes the total gain, not how often synergy was positive. It is neither a confidence interval nor a player win rate. A rare high-score outcome can move a mean or a tier. Real instruments arrive at different rounds, periodic timers may already be running, and the best choice can depend on the visible board, offered cards and reward target. These tiers apply to the fixed sample and policy.

Engine references: [instruments](../../src/core/vorax/instruments.ts), [board events](../../src/core/vorax/board.ts), [value model](../../src/core/vorax/value.ts), [paired futures](../../src/core/vorax/engines.ts) and [model assumptions](../vorax-sources.md).
