Build · Create · Analysis & Finance
Boundary-case test set for a contested term
You get a boundary-case test set for one contested term: a set of edge cases, an in-or-out verdict for every case under every rival definition you supplied, at least one case that the rival definitions actually split on, and a named residue of cases your own preferred definition leaves undecided.
You receive: A JSON boundary-case record: the contested term, a numbered case set (each case with a short phenomenon description and the feature statements it contests), a complete verdict matrix using exactly the three allowed verdict values "in", "out", and "unsettled", a four-criterion assessment (operational, essential, widelyApplicable, succinct) for each supplied definition using exactly the three allowed values "meets", "fails", and "unclear", and a residue list of the case ids your preferred definition marks "unsettled".
Part of Win Market
Example
A sample of what this play produces. Your result is generated for your inputs.
- Term
- AI agent
Cases
- Case Id
- C01
- Phenomenon
- A coding agent given 'fix issue 212' that plans, edits files, runs the tests, and opens a pull request unattended
- Is Control
- true
- Case Id
- C02
- Phenomenon
- A thermostat that senses room temperature and switches the furnace to hold a setpoint
- Is Control
- false
- Contests
- A
- C
- Case Id
- C03
- Phenomenon
- A single-turn chat completion that answers one blank-prompt question and stops
- Is Control
- false
- Contests
- A
- B
- Case Id
- C04
- Phenomenon
- An RPA script that follows a fixed click-path to copy invoice fields between two systems every night
- Is Control
- false
- Contests
- A
- Case Id
- C05
- Phenomenon
- An LLM with tool calling where a human reviews and approves every individual step before it executes
- Is Control
- false
- Contests
- A
- D
- Case Id
- C06
- Phenomenon
- A recommendation engine that continuously reranks a product feed from click signals
- Is Control
- false
- Contests
- C
- Case Id
- C07
- Phenomenon
- A cron-scheduled ETL pipeline that moves data on a timer with no goal representation
- Is Control
- false
- Contests
- A
- C
- Case Id
- C08
- Phenomenon
- A STUD play run: a short intake fills from workspace facts, an AI runs multi-step work, and the deliverable passes the play's acceptance check before it lands
- Is Control
- false
- Contests
- D
- Case Id
- C09
- Phenomenon
- An open-ended multi-model swarm that explores and generates without any stated objective
- Is Control
- false
- Contests
- C
- Case Id
- C10
- Phenomenon
- A self-driving route planner that takes a destination, plans, and executes the drive end to end
- Is Control
- false
- Contests
- B
- Case Id
- C11
- Phenomenon
- A human analyst steering a general-purpose chat assistant turn by turn to draft a memo
- Is Control
- false
- Contests
- D
- Case Id
- C12
- Phenomenon
- A deep-research run that plans its own queries, browses sources, and compiles a cited report against a stated question
- Is Control
- false
- Contests
- B
Verdicts
| Case Id | Definition Id | Verdict |
|---|---|---|
| C01 | D1 | in |
| C01 | D2 | in |
| C01 | D3 | in |
| C01 | D4 | in |
| C02 | D1 | in |
| C02 | D2 | out |
| C02 | D3 | out |
| C02 | D4 | out |
| C03 | D1 | out |
| C03 | D2 | out |
| C03 | D3 | out |
| C03 | D4 | out |
| C04 | D1 | in |
| C04 | D2 | in |
| C04 | D3 | out |
| C04 | D4 | out |
| C05 | D1 | in |
| C05 | D2 | unsettled |
| C05 | D3 | out |
| C05 | D4 | unsettled |
| C06 | D1 | in |
| C06 | D2 | out |
| C06 | D3 | out |
| C06 | D4 | out |
| C07 | D1 | in |
| C07 | D2 | out |
| C07 | D3 | out |
| C07 | D4 | out |
| C08 | D1 | in |
| C08 | D2 | in |
| C08 | D3 | in |
| C08 | D4 | in |
| C09 | D1 | in |
| C09 | D2 | in |
| C09 | D3 | unsettled |
| C09 | D4 | unsettled |
| C10 | D1 | in |
| C10 | D2 | in |
| C10 | D3 | out |
| C10 | D4 | in |
| C11 | D1 | out |
| C11 | D2 | out |
| C11 | D3 | out |
| C11 | D4 | out |
| C12 | D1 | in |
| C12 | D2 | in |
| C12 | D3 | in |
| C12 | D4 | in |
Definition Assessments
| Definition Id | Operational | Essential | Widely Applicable | Succinct |
|---|---|---|---|---|
| D1 | meets | fails | meets | meets |
| D2 | unclear | meets | meets | fails |
| D3 | meets | fails | fails | meets |
| D4 | meets | meets | unclear | fails |
Residue Case Ids
- C05
- C09
Opens soon
STUD opens runs for everyone soon, on every play at once.
Scope of the check
What's verified: STUD verifies the STRUCTURE of the boundary-case set only. It checks that the record is about the term you named; that you supplied at least your minimum number of rival definitions and that each one carries a source attribution; that you supplied at least your minimum number of cases with no case repeated; that every case is judged under every supplied definition, exactly once, with nothing left blank; that every verdict is one of exactly these three literal strings: "in", "out", "unsettled" (nothing else passes, including "yes", "no", "maybe", "+", "-", "?"); that at least your minimum number of cases are ones the rival definitions actually split on; that exactly one case is flagged as a control and that all definitions agree on it; that every other case names at least one of your supplied feature statements it contests; that every supplied definition is scored against exactly the four criteria operational, essential, widelyApplicable, and succinct, each using one of exactly these three literal strings: "meets", "fails", "unclear"; and that the residue list is exactly the set of cases your preferred definition marks "unsettled", no omissions and no extras. STUD does NOT verify that a case is well chosen, interesting, or representative of the real edges of your term. STUD does NOT verify that a verdict is CORRECT: whether a given definition truly puts a given case in or out is a reading of that definition, and readings are contested (in the source paper 52% of surveyed professionals contradicted their own stated criteria when judging cases). STUD does NOT check succinctness by word count, because the source names succinctness as a criterion but sets no numeric threshold. STUD does NOT require that every feature statement you supply be contested by some case: the source's own 20-case set left one of its 13 statements (D, on observability) untested by any example, so full coverage is not treated as a law here. STUD does NOT judge whether your preferred definition is better than its rivals. The field or discipline you name is collected as context for the operator and is not itself checked.
Get early access to STUD the day it goes live.