“It won’t start” is two different complaints wearing one phrase — and each one needs its own diagnostic tree. Here is how to tell which lane you are in from one start attempt, why running the wrong tree proves nothing, and the one observation that decides.
Conceptual definition
A crank-no-start and a starts-then-dies describe two different physical situations. In a crank-no-start, the starter turns the engine over and the engine never catches — spark, fuel, and compression never line up at the same time. In a starts-then-dies, the engine catches, reaches idle-character running (audibly running under its own power, RPM up at idle speed), and then dies. That difference is everything: the engine that ran briefly proved all three legs of combustion momentarily and then lost something. The engine that never caught proved no such thing.
The boundary between them is a single observation: did the engine reach idle character at any point during the start attempt — while the key was still held, or after it was released? Yes means it belongs in the starts-then-dies lane. No means it belongs in the no-start lane. One observation, two trees, no guessing.
Why it varies
The same four words — “it won’t start” — cover both situations because what the driver wants is the same: the car running. But what the engine did is different, and diagnosis follows what the engine did, not what the driver wanted. That is why the boundary is the observation, never the complaint wording.
The boundary sits at idle character rather than at key release for a reason: some engines catch while the starter is still engaged. The question is not “did it keep running after you let go” — it is “did it ever run under its own power, at any point.” A run of one second still proved the three legs. Duration decides which stall branch follows, not which lane you are in.
What it is not
Not a severity scale — a stall after one second and a no-start are not “almost the same thing.” They fail in different systems and get tested differently.
Not decided by cranking speed alone — both lanes crank normally. A slow or absent crank is a starting-system question, not a lane question.
Not decided by how it sounds “like last time” — the lane is decided per observed attempt. Behavior can change between attempts, and the lane can change with it.
Not a run if it only sputtered during crank — popping, bucking, or firing while the starter is still turning the engine is cranking-speed combustion, not a run. This is the number-one misclassification, and it sends people down the wrong tree.
Not diagnosed from a vague memory — “it did it once last week, I can’t say what it did” never classifies. Record the conditions and observe it fresh when it happens.
Not the same question as “what’s wrong with it” — the boundary only picks the tree. The tree picks the fault.
Failure modes
Running the no-start tree on a stall. The no-start tree tests the three legs — spark, fuel, compression — one by one. On an engine that ran briefly, all three already proved themselves momentarily. Testing them again burns time and produces “everything checks out” results that prove nothing about the stall.
Running the stall tree on a no-start. The stall tree chases what the engine lost after it proved itself. On an engine that never caught, there is nothing to lose — chasing fuel delivery or idle controls for an engine that never fired produces conclusions about a failure that never happened.
The sputter misread. Popping and bucking during crank feels like “it almost started,” and “almost” gets read as the stall lane. It is not almost — firing during crank without ever reaching idle character is a no-start. The feeling is the misclassification.
The remote-start misread. A commanded shutdown — a door opened, the brake pressed without the key, the remote runtime expiring — is observationally identical to a stall, and it is not a fault at all. An observation made on a remote or phone-app start cannot feed the boundary until it is re-run on the key.
The band force-fit. The stall lane splits by run duration, and the bands are definitions, not suggestions. A stall after a much longer run does not belong in either band — forcing it there makes the stall tree’s conclusions invalid.
How SpeedNeeds uses it
Observe first, one attempt, with nothing but ears and eyes. Crank it and let go of the key. Did the engine reach idle character at any point — while the key was still held, or after release? That single answer picks the tree.
If yes: it is a starts-then-dies. Note about how long it ran on its own before it died — in round terms, about 1–3 seconds or about 5–30 seconds — plus the run character (smooth or rough) and which key was used. Those observations are the stall tree’s starting evidence. If it died at exactly the same point every time, like clockwork, note that too — a metronome stall has its own suspects.
If no: it is a true no-start. Note the crank character, any sputter or bucking during crank (as evidence of firing during crank, not as a run), and the conditions. Those observations are the no-start tree’s starting evidence — the three-leg split owns the rest.
And the discipline that keeps the boundary honest: one attempt, one lane. If a later attempt behaves differently, the lane can change — the first observation never locks out the second. Re-observe; never inherit.
Caution and edge cases
A vague memory never classifies. If it will not do it now, record when it happens — cold start, hot soak, first start of the day — and run the observation when it happens.
The duration bands belong to the stall tree’s own definitions. Between about 3 and 5 seconds, round to the nearer band and carry the uncertainty — the stall tree re-bands on its own observation.
Longer than about 30 seconds then dies is outside both bands — neither stall branch covers it. Do not force-fit; note the run length and treat it as its own question.
Hybrids and pure EVs do not play by cranking phenomenology — the boundary question does not apply to them.
Push-button start: one press is one attempt. Diesel and start-stop: the same discriminator applies — did it run under its own power after the starter disengaged?
Never started with the remote or the phone app for this observation — re-run it on the key. A clean key retest does not prove there is no fault; it only supplies a classifiable observation.
The tach can back you up if it is readable — a needle jumping well above cranking speed and then falling means it ran. But the tach never overrules your ears, and a needle jump during crank without idle character means nothing.
Closing clarity
“It won’t start” forks at one observation: did it reach idle character, at any point, during the attempt? Yes is a stall tree; no is a no-start tree. The trees test different things because the engines proved different things. Get the fork right and every test downstream starts from the truth — get it wrong and the best test in the world proves nothing. Watch one start attempt honestly, and the boundary hands you the right tree for free.
Stay in the loop
Get new drops in your inbox
New QTS builds, tools, and garage stories — straight to your inbox. No spam. Unsubscribe anytime.