Read-only samplers that test a market thesis with data, not vibes
Scheduled, read-only samplers that build a time-series of Polymarket and Kalshi order books to test whether hourly prediction markets are ever actually market-makeable, with hard no-trade and no-write rails baked into every task.
A hypothesis-driven data pipeline where the safety rail is the point: a sampler stays a sampler. It is a research instrument built to answer one question with collected data, not a trading bot, and it is designed so it could never become one.
The problem
Hourly prediction markets look like they might be market-makeable, but "look like" is not an answer. Settling that question honestly needs a record of how the order books actually move over time, not a hunch and not a backtest on data that was never captured. The risk in building anything near a live market is obvious: the moment a tool can act, the temptation and the blast radius both appear. So the question had to be answered by observation alone, with collection that is structurally incapable of placing a trade or writing to a market.
Approach
The design is a set of scheduled samplers that read public order-book data on a cadence and accumulate a time-series. The whole thing is built hypothesis-first: it exists to test one claim, whether hourly markets are ever market-makeable, and to let the data answer rather than a narrative. The defining decision is what the samplers are forbidden to do. They read, they record, and they cannot trade or write, because that rail is baked into every task rather than left to discipline.
- Sample on a schedule, read-only. Scheduled tasks pull Polymarket and Kalshi order books at intervals and build a time-series. The leash: every task is read-only, so collection can never turn into action.
- Test a hypothesis, do not chase a result. The pipeline is designed around one question and lets the collected data answer it. The leash: it is a research instrument, so there are no invented results to report, only what the data shows over time.
- Respect the source. The samplers pull at a measured cadence rather than hammering the endpoints. The leash: they stay within rate limits, so observation never tips into abuse.
What was built
A set of scheduled sampler tasks that capture Polymarket and Kalshi order books over time, accumulating the time-series needed to test the market-makeable question. The design is hypothesis-driven: the instrument is shaped by the question it exists to answer, and the no-trade, no-write rails are part of every task by construction. It is a research instrument first, so the outcome is the dataset and the method, not a claimed return.
Guardrails
What the system is structurally not allowed to do. This is the through-line: capability, then leash.
- NoIt cannot place a trade. A hard no-trade rail is baked into every task by design.
- NoIt cannot write to a market. The samplers read and record only, never act.
- NoIt does not ignore rate limits. Collection stays within the limits the sources set.
Stack and tools
My role
I designed and built this research instrument, the scheduled samplers, the hypothesis-driven design, and the no-trade and no-write rails that bound every task.
Links and verification
This is an open-source project. The repository is public, so the architecture and guardrails described here can be inspected directly.