What Belongs in Discovery Before a Paper-Trading Sprint

A paper-trading sprint fails in discovery or it fails in week three. Week three is more expensive. We sell a one-to-two-week discovery before any fixed-price build that touches quotes, positions, or customer accounts. Public 2026 guides price that kind of technical discovery in the low thousands, sometimes credited toward the build. The number matters less than the artifact. If we cannot hand you a written scope at the end, we will not invent a build price to keep the conversation warm.

Discovery is paid. It is not a free audit dressed up as a proposal. You get an architecture sketch, a risk register, and a work order you can accept or walk away from. We get enough truth to estimate without a cushion that exists to cover our own ignorance.

The data contract, in writing

Market data is not "realtime" as a mood. The contract has to say:

• Which symbols exist, and who adds one.
• Delayed, realtime, or last-sale only, and the vendor who is allowed to say so.
• Whether we may redistribute those prices to a phone that is not our office. Paper trading does not waive a display license. If the price is real, the license is a line item, often budgeted in 2026 guides at a few hundred to a few thousand dollars a month at retail scale, scaling with users.
• What a correction looks like, and what a corporate action does to history.
• The session calendar. Early closes and halts are not edge cases. They are the calendar.

The UI is forbidden to invent a field the contract does not guarantee. That single rule prevents the class of bug that looks like a chart library and is actually a missing corporate-action policy. Latency gets one sentence, chosen on purpose. Either we are on a delayed tape and we label it, or we write a budget from vendor to screen and account for every hop. Mixing "live" in the with a 15-minute feed in production is how demos lie. A 100 ms story and a delayed contract cannot share a sprint. Pick one.

The ledger, and the list of what it will not do

A paper account is an accounting system that happens to have a chart. Discovery names the books: cash, positions, a fills journal, and the reconciliation that proves they balance. We want a fuzz sketch in the risk register, not necessarily the fuzzer itself. Buy, sell, split, dividend. State what should be true afterward. If we cannot say it in discovery, we cannot test it in the sprint.

Out of scope is part of the document. Options, fractional shares, short sales, multicurrency, and tax lots are the usual visitors. Each one changes the ledger. Each one is a different work order. Writing "phase two" without a definition is how phase two becomes the actual product and the fixed price becomes theater. We list them as not-in-this-sprint, with a one-line reason, so a stakeholder cannot smuggle them in during a standup.

Order routing is the brightest line. Discovery states whether any message leaves the building toward a venue or a brokerage API. For a paper sprint the answer is no. The day the answer becomes yes, counsel is in the room and the estimate is a different document. We will not bury a route inside a paper ticket "just in case."

The compliance boundary, without playing lawyer

We are not your counsel, and discovery is not a legal opinion. It is a boundary the counsel can review. Paper only. No customer cash. No order routing. Disclosures on the surface that say the fills are simulated. If a lawyer reads that boundary and says it is wrong for your entity, we stop. We do not negotiate the statute inside a sprint.

What we will estimate is the engineering that sits behind a boundary someone qualified has accepted: authentication, the ledger, the feed adapter, the staging deploy, and a handoff. KYC widgets, money-transmitter analysis, and audit reports are named as out of scope unless they are explicitly in the order. 2026 cost guides keep showing that this layer, once you touch real money, costs more than the screens. Discovery exists so we do not find that out in week three.

Definition of done

The work order that falls out of discovery fits on a page.

• One feed, named, with the delay and the license status.
• A paper ledger that reconciles under the script we described.
• A watchlist and a ticket that talk to that ledger and no other.
• A staging URL, notes a new engineer can follow, and a handoff call.
• A price and a date. If scope grows, a new order. Not a quiet overrun.

Then the sprint is two or three weeks, and both sides can tell whether it shipped. That is the whole point of refusing hourly work. The clock is a bad referee. A staging URL is a good one.

If you want to see the shape of the thing discovery is protecting, open Stock Picks and click a paper ticket. That client is the reference we argue from: a feed, a ledger, no pretend venue. Bring us the ways your product is not that. Those differences are the discovery. Everything else is a sprint we already know how to price.