If you are building a prediction market, the first product decision is not whether to use an AMM or an order book. It is the question itself: what real-world fact moves money, how you will know when that fact is settled, and what happens when reality refuses to cooperate. Sloppy event text turns every later engineering choice into a different way to lose trust. The chapters on blockchain risks showed how ambiguous rules invite bond wars and angry traders; this chapter is where you stop inheriting those mistakes and start authoring rules that a stranger can score without a committee meeting.
You are not listing a ticker. You are shipping a legal-economic machine whose only output is a payout flag.
The audit mindset from venue risk chapters applies in reverse: before the first dollar trades, ask whether a skeptical stranger with money on the line could score your market from the PDF alone. If the answer is no, you are not ready for mechanism debates or liquidity budgets.
The resolution spec as your real product
Treat every market as a short Resolution Spec traders can open in one click. The marketing title can be punchy; the spec must be boring and precise. At minimum it should name the outcome set (including whether INVALID is possible), define capitalized terms like “certified” or “official,” list a primary source hierarchy (government table before newswire, never “community consensus”), spell edge cases (recounts, delays, restatements, withdrawals), and pin clocks in UTC: when trading stops, when an outcome may be asserted, how long challenges run, and what happens if the source never publishes.
Resolution clarity is what makes “price equals probability” conditional on honesty—not magic. If two honest readers can read your spec and pick different winners, rewrite before you spend a dollar on liquidity.
Outcome geometry matters early
Binary yes/no markets are the easiest to explain and defend. Categorical markets need mutually exclusive outcomes and a clear rule when a named option drops out. Scalar markets need the exact series, rounding, and whether revisions count. Conditional children should not launch while a parent is still wide open unless your spec says exactly what happens when the parent resolves NO. Bundles and trees multiply dispute risk: one ambiguous leg poisons the whole payout story.
Match contract shape to the question you are actually asking. A debate “dominance” market fails because “dominate” has no scorer; a measurable prop—“net favorability ≥ +3.0 in a named poll fielded on specific dates”—might be defensible if the poll definition is pinned.
Three clocks traders confuse
Trading end is when new risk should stop because the event is effectively knowable or about to be. Real-world resolution is when the fact exists off-chain (certification published, jobs report printed). Oracle finality is when your system locks a flag and collateral can move. Headlines on election night are sentiment; redemption follows the certification clock if that is what you promised.
Ending trading too late invites sniping after deterministic data leaks. Ending too early leaves a stale mid that misleads journalists. Align halts with information flow, not with when you want marketing to stop.
Source hierarchy without folklore
Good specs read like journalism standards, not vibes. Priority one is usually a government statistical release or certified electoral authority. Priority two might be a regulator filing or court order. Priority three might be a named newswire only when primary sources lag—and only if your spec says what happens when the wire retracts. “Obvious to a reasonable person,” “viral,” and “community consensus” are dispute magnets; they convert forecasting into litigation.
Publish the same document at list time and resolve time. Hash or version the PDF so traders know the rules did not drift while they held risk.
Worked example: clean election binary
Suppose you list: “Will Candidate A win the 2028 general election per FEC-certified result?” Outcomes are YES and NO. Primary source is FEC certified totals for President. Trading ends at 23:59 UTC on election night. Assertion is eligible only after FEC certification is published. A forty-eight-hour challenge window follows an optimistic oracle pattern you will choose in the next chapter. INVALID triggers if no certification exists by a stated spring deadline; recounts wait for certified totals, not cable calls.
A trader watching a 62¢ mid on election night is pricing mood and partial information. A trader redeeming after certification is pricing your spec. Keep those stories separate in comms and in UI copy.
Bundles and joint payoffs
When multiple legs pay together, specify whether legs are independent, whether one invalid voids all, and how partial data delays work. Traders will model correlation; if your spec does not, they will model suspicion instead.
Worked example: scalar CPI
“March 2026 CPI year-over-year ≥ 3.0% per BLS series CUUR0000SA0, first print.” The measure is year-over-year to one decimal on the first published value; revisions after settlement are ignored. Trading ends one minute before the release; assertion follows the 08:30 print. INVALID applies if BLS skips the release. At 72¢ YES, traders are pricing the threshold event; your job is feed match, not debate aesthetics.
Worked example: the prop you should not ship
“Will Party A dominate the debate?” fails on every layer: “dominate” is undefined, no scorer exists, trading end is unclear, and social clips will move prices without a settlement path. Replace with a measurable prop tied to a named instrument and field dates, or do not list. Curators exist to protect the catalog from becoming a bond-war factory.
INVALID, void, and refunds
Decide refund policy before liquidity arrives. Cancellation, duplicate listings, ambiguous outcomes, and missing data sources are not edge cases—they are routine on long-tail catalogs. Forcing YES or NO when your spec cannot score the world converts a wording bug into a reputation event. High INVALID rates on decentralized venues often mean bad authoring, not “decentralization working.”
Traders price invalid tails even when they never happen. If you do not document the path, they price suspicion into the mid.
Pre-launch review (what good operators actually do)
Read the title aloud to a non-expert. If they ask “what does win mean?” you are not done. Simulate two disagreeing asserters: if each can defend a different winner, fix the hierarchy table. List headlines that break your rules—restatement, postponement, death, tie—and write the remedy for each. Pin the spec URL and version hash so the document at resolve matches the document at trade. Screen category and geography with counsel before creators promote the market. Sanity-check dispute bonds against maximum manipulation profit on open interest.
Compliance filters which questions you may ship, not only who may click buy. A beautiful mechanism cannot rescue a question you were never allowed to list.
Common failures
Marketing copy that disagrees with the legal spec creates permanent anger. Competing feeds without order produce false asserts. Trading that continues after deterministic data invites oracle snipes. Too many correlated outcomes split seed capital into dust. Bundles without independence mislead traders about joint probabilities.
Another quiet failure is conditional sprawl: launching a general-election child while the primary parent is still at forty cents duplicates dispute surfaces. Gate children in the spec, not in Discord announcements.
Misconceptions
“Lawyers can fix it at resolve time” is false—dispute systems pick among readings of what you wrote, they do not invent precision you never shipped. “INVALID is failure” is false—INVALID is often the honest output when the world cancels or data never arrives. “Shorter titles convert better” may be true for ads and false for trust; the spec can be long if it is clear.
Conditional and parent–child markets
When a child market only matters if a parent resolves YES, say so in the parent spec first. If the parent fails or resolves NO, the child might auto-resolve NO, INVALID, or void—pick one and publish it. Launching a general-election contract while a primary parent still trades near forty cents duplicates dispute surfaces and splits liquidity. Traders are not confused because they are dumb; they are confused because you gave them two overlapping clocks.
Transparency traders expect
Pin the spec where the trade button lives, not in a footer link. Version it when you amend edge cases. If you change wording after volume arrives, you are effectively rewriting open bets—treat amendments like exchange rule changes with notice and clocks. The transparency chapter on venue trust applies here: the PDF is part of the product surface.
Key ideas
Event definition is product design; mechanism and oracle only execute what you wrote. Use outcome geometry early, align three clocks to real information arrival, and treat INVALID as a first-class outcome. A stranger should score your spec without a committee—if they cannot, neither will your oracle.
What comes next in this module
This chapter locked the market DNA: the fact, the outcomes, the clocks, and the remedies. Everything downstream—matching engine, oracle, liquidity, growth—only executes what you wrote.
The next chapter chooses how prices form once the spec is tight: continuous limit-order book, automated market maker, or a deliberate hybrid path as volume matures.
Before you ship, archive a one-page “scorecard” answer key: for ten realistic world states (certified win, recount, cancellation, data delay), which outcome token pays whom. If any row is TBD, you are not done.
Next: Choosing a Market Mechanism (AMM vs. Order Book)