Free tool · runs in your browser

The Wording Clinic.

Paste a market or contract question. It's checked against published wording-failure patterns from the standing research agenda — the same patterns that make a resolution get disputed. Not AI: every rule is named, fixed in advance, and listed in full below.

Your question

How this works

Everything above stays in your browser. Nothing is sent to a server or stored anywhere.
Each check is a named pattern, listed alongside the result — not a model's judgment.
A clean result does not mean the wording is safe. It means no listed pattern matched.
This is not legal advice, and it is not a determination the Institute stands behind for any specific contract.

What gets checked

· Subjective language
· Compound condition
· No resolution source named
· No resolution deadline stated
· No dispute or tie-break procedure
· No halt or suspension criteria
· Numeric threshold without a rounding rule

Start typing, or load an example, to see a result here.

The rules

All seven checks, and where each one comes from.

Each pattern maps to one of the six standing research tracks, and carries a stable identifier — PMI-WP-001 and its siblings — so a flag can be cited in a dispute rather than described. This tool is a small, mechanical slice of that agenda, not a substitute for reading it.

Registry version 1.0.0 · last revised 2026-08-17 · the registry

PMI-WP-001
Subjective languageStructural
The text contains at least one evaluative term whose application to a given state of the world depends on the reader — “significant”, “major”, “materially” and their kin. Two resolvers acting in good faith can reach opposite conclusions without either misreading the contract.
Replace the term with a number, a date, or a named source.
Contract design & wording
PMI-WP-002
Compound conditionStructural
More than one fact is bundled into a single resolution, joined by “and” or “or”, with no stated rule for the case where the halves resolve differently. The contract has no defined outcome for a partial truth.
Split it into separate questions, or state up front how a partial truth resolves.
Contract design & wording
PMI-WP-003
No resolution source namedStructural
The text does not name whose account of the world settles the question. Absent a named source, resolution falls to whoever is holding the pen when the outcome is contested.
Name the source and the specific publication it appears in.
Settlement & resolution governance
PMI-WP-004
No resolution deadline statedWording
The text states no date or bounded period after which the question stops being open. An unbounded question cannot be settled against, and cannot be shown to have failed to settle.
State the date the question closes, and what happens if the event has not occurred by then.
Contract design & wording
PMI-WP-005
No dispute or tie-break procedureWording
The text does not say what happens when two readers disagree about the outcome: who decides, on what grounds, and within what window. A procedure designed after the first dispute is a procedure designed by the dispute.
State the decision-maker and the appeal window before the market opens.
Settlement & resolution governance
PMI-WP-006
No halt or suspension criteriaNote
The text states no basis for halting or suspending trading if the underlying event becomes contested or the resolution source stops publishing mid-market. Silence here is common and is not always a defect — venue policy may already cover it.
Confirm the venue policy that applies, or state the halt criteria in the contract.
Market integrity
PMI-WP-007
Numeric threshold without a rounding ruleNote
A numeric comparison is stated without the precision and rounding convention needed to evaluate it identically twice. “More than 5%” resolves differently at 5.004% depending on where the reader rounds.
State the measurement source, its precision, and the rounding rule.
Measurement & calibration

Scope

What this tool is not.

Everything you type stays in your browser. There is no network call, nothing is sent to a server, and nothing is stored.
Each check is a named pattern, listed above — not a model's judgment, and not a score.
A clean result does not mean the wording is safe. It means no listed pattern matched.
It is not legal advice, and it is not a determination the Institute stands behind for any specific contract.

Put it on your own site

The clinic is embeddable, free, and needs no key.

Same component, same rules, same client-side guarantee — nothing typed into an embedded copy reaches us either. It carries a link back so a reader can see the full rule list.

<iframe src="https://www.predictionmarketadvisors.com/embed/wording-clinic"
        title="Wording Clinic — Prediction Markets Institute"
        width="100%" height="720" style="border:1px solid var(--pmi-line);border-radius:8px"
        loading="lazy"></iframe>

Preview the embed

For machines

Check a book of contracts, or fail your own pull request.

The endpoint below runs the same registry this page publishes and returns every flag with its identifier and a permanent URL — so a result can be cited rather than described. Nothing posted to it is stored, and no third party is in the path.

Limits, in full: 2 MB of body, 100 items per request, 8000 characters each, unique ids, and roughly 60 requests a minute. That last one is approximate and the page says so because the endpoint cannot honestly say otherwise — it is counted in one server instance's memory, and there is more than one instance. It stops a runaway loop; it is not a quota. Batch your book into requests of a hundred rather than pacing single calls against it.

curl -sX POST https://www.predictionmarketadvisors.com/api/clinic \
  -H 'content-type: application/json' \
  -d '{"items":[{"id":"MKT-104","text":"Will the FOMC cut rates before year end?"}]}'

# → { "registry": { "version": "1.0.0", … },
#     "results": [ { "id": "MKT-104", "worst": "structural",
#                    "flags": [ { "id": "PMI-WP-003", "url": "https://www.predictionmarketadvisors.com/patterns/PMI-WP-003", … } ] } ] }

If you would rather not send contract wording to someone else’s server at all — a reasonable position, and the one this page is built around — the CI action below runs the same engine inside your own runner and makes no network call, and the registry is published as data at /patterns.json for anything building against it.

# .github/workflows/contracts.yml
- uses: kalepasch1/prediction-markets-institute/wording-clinic-action@v1
  with:
    files: contracts/**/*.md
    fail-on: structural
    registry-version: '1.0.0'   # optional

It annotates the line the wording is on, links each flag to its published definition, and exits non-zero on anything structural. Pinning registry-version is optional: leave it off to track the registry, set it to be told the day it moves rather than surprised by a flag you did not ask for.

Where these rules come from

The same checklist the Institute holds its own work to.

Read the research agenda