A discovery is a decision with a deadline

Every backlog has a spike in its third sprint. It started with energy — "evaluate the new search engine", "look into the vendor API" — and it is still open, because nothing about it defines what closed looks like. Nobody killed it. Nobody concluded it. It just runs, at low intensity, absorbing an afternoon here and a standup mention there.

That is not discovery. That is a parking lot with better branding.

Research is open-ended on purpose. Discovery is not.

Research is a legitimate activity whose defining property is that you do not know where it ends. A delivery organisation needs some of it and should label it honestly.

Discovery is a different thing: it exists in service of a specific pending decision. Somewhere upstream, a real request is waiting — should we commit to this, at what size, in what shape — and the honest answer was "we do not know enough yet". Discovery is what you run instead of guessing. Which means it inherits the decision's obligations: it must end, and it must end with an answer. A discovery that outlives its question has quietly become research on the delivery budget.

The anatomy

Four parts, and only one of them is optional:

• A goal. What this discovery exists to find out — one or two sentences, tied to the pending decision.
• Questions, optionally. The specific unknowns, written down so that "done" is checkable against them rather than felt.
• A time-box. How much effort this is allowed to consume. Learning has diminishing returns, and the time-box is the admission that a cheaper, earlier answer usually beats a thorough, late one.
• A target date — required. The calendar day on which this discovery will conclude and the decision will be made. Not "about two weeks": a date. Without it, nothing distinguishes a discovery in progress from one that has silently died, and no signal ever fires to say the decision is now due.

The target date is the load-bearing part, and it is the part most teams leave out. A time-box without a date bounds the effort but not the elapsed time — a "two-day spike" can spread its two days across a quarter and technically stay inside its box.

It ends with a decision, not a document

On the target date, two things get recorded. First the findings — what was actually learned, attached to the demand it was learned for. Second the decision, and there are exactly four: plan the work, re-open the decision with what was learned, park it, or decline it. Parking and declining require a reason; all four keep the findings on the record.

"We need more time" is conspicuously not on the list. It can be the honest answer — in which case the decision is made explicitly, with a new date, rather than by the old date sliding past unremarked. The difference matters: one is a recorded judgement that the question is worth more investment, the other is a spike becoming furniture.

A worked example

A data platform EM at a 900-person company is asked to move product search off a struggling self-hosted cluster. The request is real, the impact is clear, and the capacity question is unanswerable — nobody knows whether the migration is a quarter or a year. Committing now would need a guess dressed as an estimate; a decision-ready demand it is not.

So the decision is a discovery. Goal: establish whether the managed service can hold current query load and what the migration would cost. Three questions: latency under production traffic, index rebuild time, egress cost at current volume. Time-box: six person-days. Target date: three weeks out. On that date the findings say the latency is fine, the rebuild is the hard part, and the cost is a third higher than hoped — and the recorded decision is to plan a two-team migration next quarter, sized with numbers that came from somewhere. Total spend: six days. The alternative was a commitment built on adjectives.

How DeliverySheet does it

Running a discovery is one of the four exits at the Decide step, and the product will not open one without a target date — the goal, the questions, the time-box and the date are recorded together, and a date in the past is rejected outright. Concluding is equally strict: findings are required, the outcome is one of the four decisions above, and a park or a decline must carry its reason. Which demands deserve this treatment at all is a rigor question — see the delivery path — and the wider argument about keeping plans honest continues in the planning overview .