GUIDE 03 · COLLECTION INTELLIGENCE
Rarity starts with honest math
Weights describe relative preference, not a guaranteed occurrence count. Rules and available combinations matter just as much, so use Collection Doctor and realized summaries to understand what your project can actually produce.
Relative weights, not promises
If a group contains a trait with weight 9 and another with weight 1, the canonical model treats them in a 90:10 relative proportion before constraints. A finite generated run can be near that proportion, but it will not guarantee exactly 90 and 10 items.
Equal weights are the default. Labels such as Common, Rare, and Epic are UI conveniences; they are not mathematical guarantees. A “1/1” label must not be turned into a magical weight preset or a promise of one occurrence.
Rules change the available space
- Requires permits a choice only when its required partner is present.
- Excludes blocks an incompatible pair.
- Only With limits a trait to an explicit set of partners.
- Optional None is a first-class choice and can affect both valid capacity and projected rarity.
Because constraints can remove combinations, multiply-the-groups math is only raw combinations. It is not the exact valid capacity.
Target, projected, simulated, realized
Target expresses the weighted expectation in the unconstrained canonical model. Projected accounts for the project's valid combinations and rules when the analysis is exact. Simulated shows the result of a bounded deterministic sample. Realized is what a generated collection actually contains.
When a project is too large for exact analysis, the tool reports bounded or unknown results instead of inventing precision. A requested supply greater than unique valid capacity needs duplicate protection disabled, or it must be rejected.
A practical workflow
- Import and enable the traits you intend to generate.
- Set relative numeric weights and decide which groups allow None.
- Add compatibility rules only when the design requires them.
- Run Collection Doctor and read capacity, unreachable traits, and rule impact.
- Use Rarity Lab for projected and bounded simulation results.
- Generate a deterministic collection and compare its realized summary with your intent.
Put the model to work
The Generator keeps selection, rendering, and verification local so you can inspect the result before exporting.