Description
Uniformly choose a unique item ID excluding the last K caller history IDs; empty feasible support rejects instead of falling back.
A random construction is defined by its sampling support: positions, permutations, complete tilings, proposal sequences or another explicit set. Uniform choices at one stage do not necessarily make final visible outcomes uniform. A replay seed gives a deterministic sequence; equal values can hide different sampled positions. These tools are for reproducible synthetic examples rather than cryptographic entropy.
Uniformly choose a unique item ID excluding the last K caller history IDs; empty feasible support rejects instead of falling back.
When to use this tool
Use this tool to uniformly choose a unique item ID excluding the last K caller history IDs; empty feasible support rejects instead of falling back.
How to use the tool and how it works
- Enter items, history, cooldown, seed using the definitions below.
- Run the operation. Resolve input-format errors before interpreting an output.
- Read result.seed, result.item, result.eligible, using the stated boundary, ordering, sign and absence conventions.
Take the last cooldown history entries, or no history when cooldown is zero. Exclude those exact IDs from the distinct item bank, choose one remaining item uniformly and return the eligible bank.1
| Input field | Meaning and units | Accepted input |
|---|---|---|
| items | Distinct exact string item IDs, retaining their supplied order where specified. | Dense list of at most 128 literal strings, each at most 8,192 UTF-16 units. |
| history | Literal prior item IDs in chronological order, with recent entries at the end. | Dense list of at most 128 literal strings, each at most 8,192 UTF-16 units. |
| cooldown | Number of trailing history entries to exclude, integer 0–128. | Finite safe integer satisfying the stated bounds. |
| seed | Nonblank replay seed, at most 256 UTF-16 code units; an injected source overrides seeded draws. | Text, at most 8,192 UTF-16 code units; narrower grammar/range rules also apply. |
| Output field | Interpretation |
|---|---|
| result | Object containing the fields below. |
| result.seed | Original supplied seed string; injected-source replay additionally requires that source’s state. |
| result.item | Uniformly selected eligible item ID. |
| result.eligible | Eligible exact IDs in original bank order. |
{
"input": {
"items": [
"a",
"b",
"c"
],
"history": [
"a",
"b"
],
"cooldown": 1,
"seed": "batch56"
},
"output": {
"result": {
"seed": "batch56",
"item": "c",
"eligible": [
"a",
"c"
]
}
}
}Limitations and assumptions
- cooldown is 0–128. Unknown history strings do not add candidates, and an empty eligible bank rejects rather than relaxing the cooldown.
- A nonblank seed of at most 256 UTF-16 code units is required. With ordinary use the seed initializes one deterministic advancing source; an application-injected random source takes precedence and the seed alone then cannot replay the result. Uniformity statements assume independent uniform source draws in [0,1). Source calls are bounded and degenerate injected sources can fail a rejection budget. This is not cryptographically secure.
- Every declared outer field is required and unknown outer fields are rejected. Strings and JSON text have size limits, lists are bounded to 128 entries, and JSON nesting/node counts are bounded. Tighter operation-specific bounds stated above take precedence.
- Ordinary numeric outputs must be finite and within safe-integer magnitude. Exact large lengths/counts use decimal strings. Original record indexes and code-unit positions are zero-based; span ends are exclusive unless stated otherwise.
Alternative or Complementary methods
Choose a generator whose support matches the intended experiment and retain source-position provenance. Use a separate statistically justified process model or cryptographic source where those properties matter.
References
Similar or alternative tools
- Random Block Order Permutation
Uniform whole-block position permutation preserving internal order and duplicate display labels.
- Random Bounded Chunk Cut Positions
Uniformly sample bounded positive chunk-length compositions of an explicit total with fixed chunk count using exact completion-count unranking; infeasible support rejects.