How to Keep a Discovery Record People Can Find
How to keep a discovery record: four lines per question, killed ideas logged as output, a timebox on each question and a place anyone can find it later.
To keep a discovery record, write four lines for every question, what you asked, what you did, what you found and what you decided, record killed ideas as output, and store it where someone who was not there can find it. It costs almost nothing when you write it and saves a great deal later.
Discovery has a particular weakness: its most useful results are decisions and understanding, neither of which leaves anything visible behind. This post is about making them visible without building a bureaucracy.
How to keep a discovery record in four lines
The format is deliberately small. For each question the team looks into, write these four things.
| Line | What goes in it |
|---|---|
| What we asked | The assumption or risk, phrased as a question |
| What we did | The test, and who it involved |
| What we found | The result, including the unwelcome parts |
| What we decided | Build, drop, or the one thing to test next |
That is the whole record. If it takes longer than a few minutes to write, it has become heavier than it needs to be. The compounding value comes from consistency and findability, not from length.
Why four lines are enough
A short record is more likely to be written, and a written record is the only kind that helps. The guide gives three reasons it is worth the small effort.
- It stops teams relearning the same thing every year. Without a record, a question looked into once is looked into again by people who never knew the answer existed.
- It lets a new joiner understand why the product is shaped as it is. The record holds the reasoning that would otherwise live only in someone's memory.
- It shows, over a year, where your assumptions are systematically wrong. One result is an anecdote. Many results side by side reveal a pattern in how the team guesses.
The last point is the least obvious and the most useful. A team that can see its own pattern of errors can improve its judgement. A team without a record has experience, but not the compounding kind.
Keep it findable
The guide adds a second condition: the record must be findable by someone who was not there. A record that exists but cannot be located is nearly as useless as one that was never written.
Test it from the other side. Ask whether a colleague who joined last week could locate the answer to a question the team looked into last quarter, and whether they would understand it without asking anyone. If not, the record is not yet doing its job.
The habit that makes it pay off is the guide's last one: check the record before starting something that feels familiar. If it does feel familiar, the odds are that someone has been here before.
Record the kills, and treat them as output
The most valuable discovery outcome is frequently a decision not to build something, because it returns a quarter of capacity to better uses. But teams do not naturally celebrate this, so it goes unrecorded. Meanwhile delivery produces visible output every week, and next to it discovery begins to look unproductive.
The fix is to record killed ideas explicitly, with what was learned. The guide names three effects.
| Effect | What it prevents |
|---|---|
| It changes the culture | Stopping an idea comes to feel like a result, not a failure |
| It protects the track | The accusation that discovery produces nothing |
| It keeps the memory | The same idea returning in six months with no one recalling why it was dropped |
There is also a diagnostic value. When discovery never produces a decision against building, nothing is being filtered and the delivery queue goes unexamined. A record with no kills in it is a warning sign worth acting on.
Timebox every question
Each entry in the record should also carry a limit. Discovery expands to fill available time as reliably as any other work, and it has a particular hazard: it always feels responsible to learn a little more.
Set the limit before you start, a few days or a week, and write the date beside the question. When the date arrives, decide with what you have. If the honest answer is that you still do not know, the decision line says so and names the one thing you would test next. That counts as an entry and as progress. What does not count is an entry that never closes.
Open-ended research is how a discovery track becomes an internal department that ships nothing. A timebox plus a four-line entry is the smallest structure that prevents it. The wider routine for working through questions in this way is set out in how to run product discovery.
What goes in the record and what does not
| Belongs in the record | Does not belong |
|---|---|
| The question, as an assumption or risk | A solution waiting for approval |
| The result, including a negative one | Only the findings that supported the plan |
| The decision and the date | An open-ended note with no closing line |
| The one thing to test next, if still unsure | A long report nobody will read |
Record negative results with the same care as positive ones. A record that only holds good news cannot show where assumptions are wrong, and that is one of the main reasons for keeping it.
How the record feeds the rest of the loop
A discovery record is not an archive. It is an input to three other routines.
- The hand-off. When an item graduates, the record supplies most of the note the builders need: what was assumed, tested and found. See how to hand off discovery to delivery.
- The retrospective. The record lets you ask which assumptions were wrong, where you were overconfident, which tests changed a decision and which were theatre, and how long items really took compared with what you expected.
- The next question. Checking the record first stops the team spending a week on something already settled.
The record also provides the evidence for the bar you set earlier. If you want to see what a written standard of evidence looks like, read what does validated mean in discovery.
What the record does for discovery's reputation
Because the record can be read by people who were not in the room, it also helps the wider team understand what discovery is doing. The accusation that discovery produces nothing is hard to sustain when a clear trail of questions, results and decisions sits in plain view.
It also keeps the work honest from the inside. An entry has a decision line, so a question cannot quietly stay open for ever, and a result has to be written down in plain words, so a disappointing one cannot be quietly forgotten.
How a record changes the way the team guesses
The longer view is where the record earns its keep. Over a year, entries side by side show which kinds of assumption the team tends to get wrong. It may be that the team is routinely too hopeful about whether people will change their behaviour, or about how long discovery takes. Either way, the pattern is invisible without the record and obvious with it.
| Time horizon | What the record gives you |
|---|---|
| This week | A check that you are not repeating a question already answered |
| This quarter | A clear answer to anyone asking what discovery produced |
| This year | A view of where your assumptions are systematically wrong |
| Across new joiners | The reasoning behind why the product is shaped as it is |
This is the same compounding argument that runs through the whole guide. A small cost at the moment of writing buys a benefit that grows with every entry.
Common ways a record fails
- It is never started. Writing it feels optional, so it is dropped the first busy week.
- It is written only for wins. Killed ideas and negative results go missing, and the record loses most of its value.
- It cannot be found. The entries exist but nobody outside the original conversation could locate them.
- It is never consulted. Nobody checks it before starting something familiar, so the team relearns the same thing anyway.
- Questions never close. There is no timebox and no decision line, so entries sit open and the track drifts toward open-ended research.
Each failure has a direct remedy in the checklist below. None needs new tools.
A checklist for the record
- Each question has four lines: asked, did, found, decided.
- Killed ideas are recorded explicitly, with what was learned.
- The record is findable by someone who was not there.
- We check it before starting something that feels familiar.
- Each question has a timebox and a date to decide.
- Negative results are recorded as carefully as positive ones.
Frequently asked questions
How detailed should a discovery record be?
Four lines per question: what we asked, what we did, what we found, what we decided. It costs almost nothing at the moment of writing, and detail beyond that tends to stop it from being written.
Why record ideas we decided not to build?
Because the decision not to build is often the most valuable output, and because it prevents the same idea from returning with no memory of why it was dropped. It also answers the charge that discovery produces nothing.
What if a question runs past its timebox?
Decide with what you have. If you still do not know, record that, name the one thing you would test next and set a new limit. Do not leave the entry open.
Who should be able to find the record?
Someone who was not there. A new joiner should be able to locate an entry and understand why the product is shaped as it is.
The full method, with the checklists, is in The Dual-Track Playbook, a free 34-page playbook.