Building a market-making bot changed how I think about “opportunity.”
At first, more markets looked obviously better.
In practice, every additional market adds another place where state can go wrong:
stale data
thin liquidity
partial execution
uncertain inventory
settlement edge cases
So the system I want isn’t the one that trades the most.
It’s the one that can confidently explain:
why it traded, what actually filled, what it owns now, and how it gets back to a safe state.
Coverage is easy to expand.
Confidence is harder.
HIP-4 is reaching the point where headline volume matters less than where the volume comes from.
Through Oct 2, independent tracking shows ~$465M in matched volume across 6,395 questions.
But the mix is revealing:
Crypto: ~$229M
Sports: ~$215M
Macro: ~$12M
Prediction markets aren’t one market.
Different categories have different information speeds, participant behavior and liquidity patterns.
For builders, “HIP-4 volume is growing” is becoming less useful than asking:
which market structure is actually producing repeatable participation?
“Canceled” does not always mean “nothing happened.”
An immediate-or-cancel order can fill part of its requested size and cancel only the remainder.
If a bot stores just the final label, it can conclude:
order canceled → no position
while the account is already exposed.
The safer record is:
requested → filled → canceled remainder → verified position
Order status is a lifecycle, not a verdict.
A dead man’s switch is one of the cleanest ideas in trading-system safety:
orders stay live only while the bot keeps proving it is healthy.
If that heartbeat stops, the exchange cancels the open orders.
But implementing it safely is more than scheduling one cancel:
arm → verify → renew → monitor → trigger → reconcile
The final step matters.
After the switch fires, the bot still needs to confirm which orders disappeared, which had already filled and what exposure remains.
Fail-safe mechanisms are state machines too.
Hyperliquid’s announced move to reduce the minimum outcome order from $10 to $1 sounds like a simple accessibility change.
It’s also a market-structure experiment.
Smaller orders let traders:
quote probabilities with less capital
test thinner markets
spread activity across more questions
adjust exposure in finer increments
But a lower minimum doesn’t automatically create liquidity.
It lowers the cost of experimentation.
Whether that becomes better price discovery still depends on depth, spreads and execution quality.
A measured bullish case for HIP-4 is that builders don’t need to wait for protocol-level trading fees to experiment with sustainable products.
Outcome fees are currently zero during initial testing, but builder codes can still attribute eligible order flow and route builder fees onchain.
That creates a useful separation:
protocol → provides the market primitive
builder → owns discovery and UX
user → explicitly controls fee permission
If specialized outcome apps can monetize without fragmenting the underlying execution layer, the ecosystem can support more interfaces without requiring a new exchange for each niche.
Multi-outcome markets create an easy engineering trap:
each outcome has its own order book, but the positions aren’t independent.
If exactly one outcome must settle Yes, then the real state lives at the question level—not inside any single book.
A bot can reconcile every order correctly and still misunderstand its total exposure.
The safer model is:
question identity → linked inventories → conversion operations → conservation checks
Local correctness isn’t enough when the contracts share a global invariant.
An outcome market doesn’t end when trading stops.
Settlement is another system boundary that must be verified.
A reliable bot should be able to answer:
What contract did I trade?
What input resolved it?
What fraction did it settle at?
Did the resulting balance match that resolution?
“Market settled” is an event.
Spec → evidence → settlement value → balance reconciliation is an audit trail.
That distinction matters whenever real money depends on an external fact becoming final.
One of the hardest bugs in trading systems starts with a clean abstraction:
“I placed one order.”
In a merged outcome book, that intent may interact with both sides—and an order that partly fills and partly rests can surface as separate records.
The UI can stay simple.
The reconciliation layer cannot.
It needs one canonical intent, linked execution records and a final check against actual economic exposure.
Abstractions are useful.
Reliability comes from knowing where they leak.
One overlooked state machine in trading infrastructure is permission.
A market can move through:
allowed → restricted → close-only → settlement-only
based on jurisdiction, product type or a court decision—not market data.
If compliance is treated as a static login check, the system will eventually behave incorrectly while orders or positions still exist.
Good architecture treats eligibility as live state:
continuously verified, auditable and safely degradable without trapping the user mid-lifecycle.
If outcome markets expand into corporate events, surveillance becomes part of the product.
A clean matching engine isn’t enough.
Builders also need to ask:
Who knew what, and when?
Is activity concentrated before an announcement?
Can related accounts be identified?
What happens when manipulation is suspected?
Trading infrastructure doesn’t end at execution.
In information-sensitive markets, monitoring and auditability are also core reliability systems.
HIP-4 is starting to show an interesting asymmetry:
market creation is becoming abundant, but liquidity is still highly concentrated.
Recent data shows 3 registered permissionless venues launched 3,468 traded outcome tokens and generated ~$62.7M in matched volume.
But one venue accounted for ~95% of that volume.
That tells me the next bottleneck probably isn’t deployment.
It’s distribution, liquidity concentration, and market selection.
Permissionless creation opens the door.
The harder problem is deciding where attention should actually go.
A useful way to evaluate new market infrastructure:
Don’t just ask what can be traded?
Ask:
Who defines the contract?
Who executes it?
Who holds collateral?
Who settles it?
Who handles failure and disputes?
HIP-3 and HIP-4 are interesting because those responsibilities can be separated across protocols, deployers and frontends.
More modularity creates more room for builders.
It also makes every boundary between those modules a reliability problem.
Outcome markets are starting to look less like a standalone category and more like another trading primitive.
https://t.co/IKH7FmXBHW adding Events is a good example.
A trader can move between:
continuous price exposure → event-dependent exposure
without leaving the same broader trading ecosystem.
That changes how I think about prediction markets.
The interesting question may eventually become less:
“Which prediction market wins?”
and more:
“Which execution layer supports the widest set of ways to express a view?”