Horadric Cube Affix Optimizer v3

Hybrid exact solver with feasibility checks, decomposition-plus-ILP routing, and finite-horizon budget-DP escalation.

Current Item

Enchanted affixes are immune to cube operations. Any cube-touched affix is forced to non-GA.

    Target Item

      Optimizer Result

      Recommended Action -
      P(Success, GA Preserved) -
      Expected Cube Steps - Counts steps until success or GA break

      Use this after each in-game result so the optimizer can continue from the new state. Just Apply Outcome records the result and keeps the same mode/prism/result for rapid data entry without re-optimizing.

      Debug & History
      Std Dev (Success Paths) -
      Strategy -
      Feasibility -
      Decomposition / ILP -
      Residual Status -
      Solver Iterations -

              How To Use

              1. Fill in your item's current affixes. Check GA for any Greater Affixes and Enchanted for the affix modified by the Enchantress (at most one).
              2. Fill in up to 4 target affixes — the affixes you want on the finished item. Check Needs Improvement for affixes that are already present in your current item but at an unsatisfactory roll value — this tells the optimizer to treat the affix as missing and plan to re-roll it.
              3. GA preservation is automatic: any GA on a current affix that aligns with a target affix is protected throughout optimization. The optimizer will never recommend an action that risks losing it.
              4. Check Legendary if your item cannot have affixes removed (Legendary items skip the remove step and go straight to cube-reroll). The optimizer tracks whether the Enchantress is still available from the Enchanted checkbox in the current panel — if no affix is marked Enchanted, the one-time enchant is still available.
              5. Set Gear Slot to your item type to restrict suggestions to slot-legal affixes and reduce compute time. Leave it at Any to allow all affixes.
              6. Set Class to your character class to narrow the Adept skill pool to that class's skills (plus class-agnostic generals). Leave it at Any to keep every class's skills in the pool — useful for browsing but not realistic for actual crafting.
              7. Click Optimize. The recommended cube action appears with success probability and expected steps to a terminal outcome (success or GA loss).
              8. Perform the action in-game, then record it with Apply Last Outcome: pick the modification type (Added / Removed / Changed / Enchanted), the prism used (pre-filled from the recommendation), the source affix that changed, and the new affix that appeared. Click Apply Outcome and Optimize to update the item state and immediately re-optimize. Applying a result automatically clears Needs Improvement from any target affix that was just obtained.
              9. Click Reset to clear all state and start over.

              Settings — Slot Legality Overrides
              If an affix you want to target isn't appearing in the autocomplete (because the built-in slot legality table doesn't list it for your gear slot), click Settings in the toolbar to add a custom override. Specify the Slot, an optional Class restriction, and the affix name. The override is saved permanently and takes effect immediately — no page refresh required. Overrides that are already covered by the built-in table are rejected automatically. To remove an override, open Settings and click the × next to it.

              Current Limitations
              • Affix pools are based on prism categories. Slot-accurate roll weights and per-affix frequencies are not modeled, so some suggestions may be theoretically possible but practically rare.
              • Affix values and thresholds are not modeled — the optimizer targets affix identity only, not roll quality (e.g. it cannot require at least 1500 Maximum Life).
              • Gold and crafting material costs are not included in the objective. The optimizer minimizes expected cube steps only.
              • Expected Cube Steps counts steps until a terminal outcome — either success (all targets met with GAs preserved) or failure (a protected GA is lost). A lower step count alongside a lower success probability is not a contradiction: most runs end quickly via failure.
              • Residual cases that exceed the abstract-state or iteration budget return an approximate best-so-far result with explicit diagnostics. Increase Thinking Time to expand the budget; 0 uses the largest configured budget.

              Settings

              When on, each optimization run collects per-rollout outcome data and shows a Simulation Graph button below "Just Apply Outcome". The graph displays the distribution of cube steps across simulated playthroughs. Overrides the Tighten Steps Estimate setting.
              Developer
              Chooses which engine produces the recommendation. Rust / WASM and JavaScript run the exact optimizer (decomposition + LAO*); Rust loads on first use and falls back to JS if the WASM module fails. Rules-based runs the experimental heuristic rule list (no search) and uses its first-firing rule's action as the headline recommendation — enable "Tighten Steps Estimate" for Monte Carlo step and success statistics.
              When on, every applied Add / Focused / Chaotic / Remove / Enchant result is logged (with the item state before the reroll) to learn affix roll weights. Data is stored locally, persists across reloads, and is discarded automatically if the model version changes. 0 outcomes tracked.

              Add overrides to make an affix legal on a slot that it wouldn't normally appear on. Overrides are additive (allow-only) and survive a main-state Reset.

              Slot Class Affix

              Copy Config

              This JSON captures the full optimizer state — gear slot, class, legendary flag, current and target affixes (with GA / Enchanted / Needs-Improvement flags), thinking time, skill filter, slot-legality overrides, verification level, and the JS/Rust engine choice. Copy it to save or share a setup.

              Load Config

              Paste a config JSON (produced by Copy Config) below and click Load. This replaces all current optimizer state, including the JS/Rust engine choice and slot-legality overrides.

              Export Tracking Data

              Logged reroll outcomes plus the live one-shot weight estimate. Feed this JSON to scripts/learn-weights-from-tracking.js to update the baseline weights. This is a separate system from Copy Config and is never included in it.

              Simulation Graph

              Green bars = successful runs. Red bars = GA-break endings. Gold curve = Gamma distribution fit to successful runs (omitted when data is insufficient).