Dual Track Mistakes Teams Make and How to Spot Them
Dual track mistakes teams make fall into nine patterns, all failures of protection and follow-through. Learn the symptom of each one and the first fix to try.
The dual track mistakes teams make come down to nine patterns, and every one of them is a failure of protection and follow-through, not of method. That is good news, because the method is not the hard part and never was.
This post is a diagnosis guide. Instead of repeating a list, it starts from what you can observe in your own team, and works back to the mistake behind it.
Dual track mistakes teams make: the nine at a glance
All nine turn the practice into vocabulary. Every one is recoverable. They group naturally by where they show up.
| Where it shows up | Mistake |
|---|---|
| Structure | Splitting into two teams |
| What goes in discovery | Putting features in the discovery track |
| What gets tested | Testing the convenient assumption; running tests you would ignore |
| Capacity | Doing discovery "when there is time" |
| Outputs | Never killing anything; not writing down what was learned |
| The boundary | Letting the handoff become a stage gate |
| After release | Never checking what shipped |
Read the symptoms below the way a doctor reads them: find the one you recognise, then look at the likely cause.
If findings arrive as documents and builders seem surprised
The likely mistake is splitting into two teams. It is the most common failure. Insight arrives as a specification, the builders were absent from the reasoning, and everything that made it actionable is lost in the handoff.
The first fix is structural and cheap: make sure the engineers and designers who will build the work are present in the discovery that shapes it. Not for every interview, but enough that the reasoning is shared. Then say so out loud, more than once, because this is also the most common misreading of the practice.
If every discovery test confirms the plan
Two mistakes produce this symptom, and it pays to separate them.
- Putting features in the discovery track. Validating a pre-decided solution is not discovery. It is an approval process, and it reliably produces the confirmation you went looking for.
- Running tests you would ignore. If a negative result would not stop you, the test manufactures false confidence at real cost.
To tell them apart, look at the wording of the items. If they read as things to build, the first mistake is present. If they read as questions but nothing could have changed the plan, it is the second. The fix for the first is to rewrite each item as an assumption or risk. The fix for the second is to ask, before each test, whether a negative result would really stop you, and to skip the test if not. The method for both is in how to run product discovery.
If launches keep surprising you with the same problem
The mistake is testing the convenient assumption. Usability is comfortable and measurable. Whether anyone cares is uncomfortable and decisive. Teams test the first and discover the second at launch.
The symptom is a pattern: things work as designed and nobody wants them. The fix is to rank assumptions by how damaging it would be to be wrong, and to test the most dangerous one first.
If discovery never seems to happen
The mistake is doing discovery "when there is time". There is never time. Unprotected capacity is unallocated capacity, and it goes to whatever is loudest.
The symptom is easy to spot: a blank where the month's discovery decisions should be. The fix is to name a share of the team's time, write it down, and agree in advance what happens to it when delivery is under pressure. That is the subject of how to split capacity between discovery and delivery.
If the queue is full and no one can say why
This points to never killing anything. When discovery has never decided against building, it is not filtering, and what reaches delivery has not been examined.
The symptom shows up as a delivery queue full of items with no evidence behind them and nobody able to say what outcome they serve. The fix is to treat the decision not to build as a real output, record it with what was learned, and expect to see some. A track with no kills should make you suspicious of its tests.
If old ideas keep coming back
The mistake is not writing down what was learned. Teams relearn the same thing annually, and the same rejected idea returns every six months with no memory of why it was dropped.
The fix is a four-line record per question that anyone can find: what we asked, did, found and decided. It costs almost nothing at the moment of writing. For a format, see how to keep a discovery record.
If graduating work feels like paperwork
The mistake is letting the handoff become a stage gate. A template, an approval and a queue turn the practice into waterfall with new words, and the team will abandon it within two quarters.
The test is a comparison of effort: if graduating an item takes longer than running a small test, the process has become the point. The fix is to return to a short document and a conversation between people who were already involved. Details are in how to hand off discovery to delivery.
If nobody knows which releases worked
The mistake is never checking what shipped. Without the loop, the team never learns whether its judgement is good, which is how a decade of experience becomes one year lived ten times.
The fix is to check every release against the outcome it predicted, and to make that check somebody's job. The practice is covered in how to close the loop after shipping.
The symptom-to-fix table
| What you see | The likely mistake | First fix |
|---|---|---|
| Findings arrive as specifications | Two teams | Put the builders in the discovery |
| Discovery items read as things to build | Features in the discovery track | Rewrite them as assumptions or risks |
| Launch reveals nobody cares | The convenient assumption | Test the most damaging one first |
| Negative results would change nothing | Tests you would ignore | Skip them and build |
| A blank month for discovery | Discovery "when there is time" | Name and protect a share |
| Nothing has ever been dropped | Never killing anything | Record kills as output |
| The same idea returns | No written record | Four lines per question |
| Hand-offs need a meeting and a template | A stage gate | Return to a short note |
| No one knows what worked | No loop after shipping | Assign the check |
The pattern underneath all nine
Look at the right-hand column and a theme appears. Not one fix requires a new technique. Each asks the team to protect something it already believes in, or to follow through on something it already started: protect the capacity, protect the bar, protect the boundary, follow through on the record, follow through on the check.
That is why the guide calls them failures of protection and follow-through. Understanding the idea takes ten minutes. The difficulty is holding the two tracks together while a quarter is under pressure.
How the mistakes feed each other
The nine are not independent. In a real team they arrive in clusters, and seeing the links helps you decide where to start.
| If this is happening | It tends to cause |
|---|---|
| Two teams | Findings that arrive as specifications, and a boundary that needs paperwork to cross |
| Features in the discovery track | Tests that confirm, because the answer was chosen first |
| Discovery only when there is time | No kills, because nothing was looked into long enough to be dropped |
| No record | The same idea returning, and the same assumption tested twice |
| No check after shipping | No calibration, so the same errors repeat |
Notice that several of these share a root. If the capacity is unprotected, the tests get skipped, the record never gets written and nothing is killed. Fixing the root often relieves three or four symptoms at once, which is why the structural decisions at the start matter so much. Most teams that abandon the practice abandoned it because of an unmade decision at the start.
Where to start if you recognise several
If more than one symptom sounds familiar, resist the urge to fix everything. A sensible order follows the guide's own sequence.
- Structure first. Confirm it is one team running two tracks, with the builders present in the discovery.
- Then what goes where. Make sure discovery items are assumptions and delivery items have already earned their evidence.
- Then the bar and the capacity. Write down what validated means, and protect a share of time.
- Then the boundary. Keep the hand-off short and name who decides.
- Then the loop. Assign the check after shipping and bring the result back to the goals.
Each step makes the next one easier. A team that is still split in two will find that every later fix leaks.
What recovery looks like
The guide calls every one of these mistakes recoverable, and recovery is usually quiet. There is no relaunch. A team says out loud that it is one team. It rewrites a few items as questions. It names a share of capacity. It writes its first four-line entry. It drops one idea and records why. After a month the direct question, what did discovery decide and what did delivery ship, has an answer on both sides.
That is all it takes to move from vocabulary to practice. The practice does not need to be elaborate. It needs to be protected.
A monthly self-diagnosis
You do not need an audit to catch these early. Once a month, ask a few direct questions.
- What did discovery decide this month, and what did delivery ship?
- Did we kill anything?
- Can a newcomer find what we learned?
- Which shipped work have we measured?
- Is it still one team?
Two consecutive months with a blank on either track is the signal. When you find one, start with the fix in the table above and not with a redesign of the whole process.
Frequently asked questions
Which of the nine mistakes is the most common?
Splitting into two teams. The insight then arrives as a specification, the builders were not part of the reasoning, and the context is lost in the handoff.
Can a team recover once it has made these mistakes?
Yes. The guide says every one is recoverable, and since they are failures of protection and follow-through, the fixes ask for discipline rather than new methods.
How can I tell whether discovery is a real filter?
Look for decisions not to build. If discovery has produced none, it is not filtering, and the delivery queue is being fed unexamined.
Should we fix all nine at once?
No need. Start with the symptom you recognise, apply the first fix, and revisit at the next monthly check. Small, protected changes are more likely to last than a redesign.
The full method, with the checklists, is in The Dual-Track Playbook, a free 34-page playbook.