Silent-Launch Failure-Layer Inspection
“It didn’t work” is four different problems with four different fixes. This walks the funnel in order and names the layer that died — then refuses to diagnose anything below it, because the people who got further had already survived something broken.
In the worked example, one launch has a want rate of exactly 20.0% against a 20.0% floor, on 120 people — well above its own minimum — and the inspection still refuses to credit it. The step above it died, so those 120 are a self-selected remnant. Their rate describes them, not your offer.
instant download · .xlsx · yours to keep
deterministic · offline · grades launches, never people
The problem
Four different failures wearing the same shrug.
Exactly 20.0% against a 20.0% floor, on 120 people. Refused, because the step above it died and those 120 are the survivors of a broken read.
Of 28 transitions across seven launches, eleven cannot support a conclusion — three because the denominator is too thin, eight because something upstream failed.
Nothing is scored and nothing is added up. A post-mortem has an order, not a set of dimensions, and the order is the finding.
The morning after a quiet launch, the honest answer is usually “we don’t know” — and the dishonest one is a conversion rate computed on nine people. The four failures need opposite fixes. If nobody was reached, the copy is irrelevant. If they read it and did not want it, more traffic makes the same problem larger. This kit tells you which one you have, and is equally clear about the transitions your numbers cannot speak to.
See it work
Watch a rate that meets its floor get refused anyway.
L-02 is selected. Look at its third row: a want rate of exactly 20.0% against a 20.0% floor, on 120 people — and it still reads UNKNOWABLE, because the step above it died.
| Launch | Shown | Diagnosed | Verdict |
|---|---|---|---|
| L-01 Product launch: Channel Assay | 4,800 | 1/4 | NEVER REACHED |
| L-02 Newsletter launch: the distribution issue | 6,200 | 2/4 | REACHED BUT UNREAD |
| L-03 Webinar: killing a channel on purpose | 5,000 | 3/4 | READ BUT UNMOVED |
| L-04 Bundle launch: the distribution stack | 3,000 | 2/4 | REACHED BUT UNREAD |
| L-05 Beta invite to the private list | 3,000 | 1/4 | TOO SMALL TO SAY |
| L-06 Free tool: reachability sweep | 9,000 | 4/4 | THE LAUNCH CONVERTED |
| L-07 Quick Kit launch: cold-start caliper | 7,000 | 4/4 | WANTED BUT UNBOUGHT |
L-02 — Newsletter launch: the distribution issue
1. Shown → reached6,200 → 310
5.0%floor 3.0%HELD
2. Reached → read310 → 120
38.7%floor 50.0%DIED HERE
3. Read → wanted120 → 24
20.0%floor 20.0%UNKNOWABLE
A transition above this one died, so this population is self-selected. Its rate describes the survivors, not the offer.
4. Wanted → bought24 → 7
29.2%floor 25.0%UNKNOWABLE
A transition above this one died, so this population is self-selected. Its rate describes the survivors, not the offer.
Launch verdict
REACHED BUT UNREAD
2 of 4 transitions diagnosed.
Where they die
REACH 1 · READ 2 · WANT 1 · BUY 1
5 died · 1 converted · 1 too small
Not diagnosed
11
3 thin · 8 downstream
Tie-break
The mode, ties to the earliest layer. An upstream failure is more upstream and more actionable.
Across the launches
DIES AT READ
The mode, not the worst or the best. One launch dying at reach is bad luck.
Your counts and your own floors — no benchmark. It grades launches, never people, and connects to nothing. Not financial advice.
L-02 is selected. Its third transition reads 20.0% against a 20.0% floor and is marked UNKNOWABLE. Now raise its read count from 120 to 200 — the second transition holds, the boundary lifts, and the third transition becomes diagnosable for the first time. Then drop L-05’s reached count below its own minimum and watch the other refusal appear.
The model
An ordered walk, a boundary, and two reasons for silence.
Shown to reached, reached to read, read to wanted, wanted to bought. Each is marked HELD, DIED HERE or UNKNOWABLE against a floor you set. The launch verdict is the name of the first layer that died — NEVER REACHED, REACHED BUT UNREAD, READ BUT UNMOVED or WANTED BUT UNBOUGHT — or TOO SMALL TO SAY, or THE LAUNCH CONVERTED.
A denominator below the minimum you set. Four people out of nine is not a 44% conversion rate. The boundary is monotone: once it is set, nothing below it is diagnosed, and the verifier proves that across all 256 abstract states.
A transition above it died, so this population is self-selected. A buy rate measured on the people who survived a broken read step describes those people. The sample contains a step that meets its floor on a sufficient denominator and is refused anyway; 57 of the 256 abstract states have that shape.
Across several launches the Dashboard reports the modal failure layer, not the worst row and not the best. A worst-of rollup would always name the deepest failure and a best-of the shallowest, and neither is the pattern. One launch that died at reach is bad luck; three that died at the same layer is the thing to fix, and it is the only pattern a single post-mortem can never show you. Ties break to the earliest layer, because an upstream failure is both more upstream and more actionable.
Verified, not asserted
The workbook is checked against a reference engine.
You get the .xlsx — that is the product. It is built against a separate reference implementation of the same model, and every formula in the sheet is asserted equal to it before release: 139 cell-level checks on the worked example, 256 abstract boundary states, a 486-row count grid, and all 1,024 tallies of five deaths across four layers. The boundary column gets its own check on every row, because a recurrence written down a column is exactly the formula that works on row 5 and breaks on row 31. Below is that engine’s verbatim output on the same seven launches the workbook ships with, captured programmatically and diffed against this page.
==============================================================================
SILENT-LAUNCH FAILURE-LAYER INSPECTION - SLF-049 - RedHub AI LLC
==============================================================================
No score and no weights. Four transitions, walked in order, with a
boundary below which nothing is diagnosed. Your counts and your own
floors - no benchmark. Grades launches, never people.
------------------------------------------------------------------------------
LAUNCH L-01 [Product launch: Channel Assay]
------------------------------------------------------------------------------
Counts shown 4800 -> reached 96 -> read 61 -> wanted 9 -> bought 2
Minimum denominator you set 100
Shown -> reached 4800 -> 96 2.0% (your floor 3.0%) DIED HERE
Reached -> read 96 -> 61 63.5% (your floor 50.0%) UNKNOWABLE
a transition above this one died, so this population is
self-selected and its rate describes the survivors, not the offer
Read -> wanted 61 -> 9 14.8% (your floor 20.0%) UNKNOWABLE
a transition above this one died, so this population is
self-selected and its rate describes the survivors, not the offer
Wanted -> bought 9 -> 2 22.2% (your floor 25.0%) UNKNOWABLE
a transition above this one died, so this population is
self-selected and its rate describes the survivors, not the offer
VERDICT NEVER REACHED
Transitions diagnosed 1 of 4
------------------------------------------------------------------------------
LAUNCH L-02 [Newsletter launch: the distribution issue]
------------------------------------------------------------------------------
Counts shown 6200 -> reached 310 -> read 120 -> wanted 24 -> bought 7
Minimum denominator you set 100
Shown -> reached 6200 -> 310 5.0% (your floor 3.0%) HELD
Reached -> read 310 -> 120 38.7% (your floor 50.0%) DIED HERE
Read -> wanted 120 -> 24 20.0% (your floor 20.0%) UNKNOWABLE
a transition above this one died, so this population is
self-selected and its rate describes the survivors, not the offer
Wanted -> bought 24 -> 7 29.2% (your floor 25.0%) UNKNOWABLE
a transition above this one died, so this population is
self-selected and its rate describes the survivors, not the offer
VERDICT REACHED BUT UNREAD
Transitions diagnosed 2 of 4
------------------------------------------------------------------------------
LAUNCH L-03 [Webinar: killing a channel on purpose]
------------------------------------------------------------------------------
Counts shown 5000 -> reached 250 -> read 180 -> wanted 20 -> bought 6
Minimum denominator you set 100
Shown -> reached 5000 -> 250 5.0% (your floor 3.0%) HELD
Reached -> read 250 -> 180 72.0% (your floor 50.0%) HELD
Read -> wanted 180 -> 20 11.1% (your floor 20.0%) DIED HERE
Wanted -> bought 20 -> 6 30.0% (your floor 25.0%) UNKNOWABLE
a transition above this one died, so this population is
self-selected and its rate describes the survivors, not the offer
VERDICT READ BUT UNMOVED
Transitions diagnosed 3 of 4
------------------------------------------------------------------------------
LAUNCH L-04 [Bundle launch: the distribution stack]
------------------------------------------------------------------------------
Counts shown 3000 -> reached 150 -> read 60 -> wanted 20 -> bought 6
Minimum denominator you set 100
Shown -> reached 3000 -> 150 5.0% (your floor 3.0%) HELD
Reached -> read 150 -> 60 40.0% (your floor 50.0%) DIED HERE
Read -> wanted 60 -> 20 33.3% (your floor 20.0%) UNKNOWABLE
a transition above this one died, so this population is
self-selected and its rate describes the survivors, not the offer
Wanted -> bought 20 -> 6 30.0% (your floor 25.0%) UNKNOWABLE
a transition above this one died, so this population is
self-selected and its rate describes the survivors, not the offer
VERDICT REACHED BUT UNREAD
Transitions diagnosed 2 of 4
------------------------------------------------------------------------------
LAUNCH L-05 [Beta invite to the private list]
------------------------------------------------------------------------------
Counts shown 3000 -> reached 90 -> read 70 -> wanted 20 -> bought 6
Minimum denominator you set 100
Shown -> reached 3000 -> 90 3.0% (your floor 3.0%) HELD
Reached -> read 90 -> 70 77.8% (your floor 50.0%) UNKNOWABLE
90 is below the 100 you set - this is not a low rate, it is no rate
Read -> wanted 70 -> 20 28.6% (your floor 20.0%) UNKNOWABLE
70 is below the 100 you set - this is not a low rate, it is no rate
Wanted -> bought 20 -> 6 30.0% (your floor 25.0%) UNKNOWABLE
20 is below the 100 you set - this is not a low rate, it is no rate
VERDICT TOO SMALL TO SAY
Transitions diagnosed 1 of 4
------------------------------------------------------------------------------
LAUNCH L-06 [Free tool: reachability sweep]
------------------------------------------------------------------------------
Counts shown 9000 -> reached 540 -> read 400 -> wanted 120 -> bought 40
Minimum denominator you set 100
Shown -> reached 9000 -> 540 6.0% (your floor 3.0%) HELD
Reached -> read 540 -> 400 74.1% (your floor 50.0%) HELD
Read -> wanted 400 -> 120 30.0% (your floor 20.0%) HELD
Wanted -> bought 120 -> 40 33.3% (your floor 25.0%) HELD
VERDICT THE LAUNCH CONVERTED
Transitions diagnosed 4 of 4
------------------------------------------------------------------------------
LAUNCH L-07 [Quick Kit launch: cold-start caliper]
------------------------------------------------------------------------------
Counts shown 7000 -> reached 420 -> read 300 -> wanted 120 -> bought 24
Minimum denominator you set 100
Shown -> reached 7000 -> 420 6.0% (your floor 3.0%) HELD
Reached -> read 420 -> 300 71.4% (your floor 50.0%) HELD
Read -> wanted 300 -> 120 40.0% (your floor 20.0%) HELD
Wanted -> bought 120 -> 24 20.0% (your floor 25.0%) DIED HERE
VERDICT WANTED BUT UNBOUGHT
Transitions diagnosed 4 of 4
==============================================================================
ACROSS THE LAUNCHES
==============================================================================
Launches inspected 7
Converted 1
Too small to say 1
Died somewhere 5
Died at REACH (Shown -> reached ) 1
Died at READ (Reached -> read ) 2
Died at WANT (Read -> wanted ) 1
Died at BUY (Wanted -> bought ) 1
Transitions not diagnosed 11 (3 thin, 8 downstream)
ROLLUP DIES AT READ
The rollup takes the MODE, not the worst or the best. One launch that
died at reach is bad luck. Three launches that died at the same layer
is the thing to fix, and it is the only pattern a single launch
post-mortem can never show you. Ties break to the earliest layer.
==============================================================================What it is — and isn't
It reads counts you already have. It does not go and get them.
- A structural post-mortem on launches you have already run.
- Deterministic and offline — same counts, same verdicts, every time.
- A named model you can argue with: the four transitions, both refusal reasons and the tie-break are all on this page.
- Dateless. Add each launch as you run it; the mode gets more useful the more rows it has.
- Not connected to anything — no analytics, no ad account, no tag. You type five counts.
- Not a statistical test. The minimum denominator is your judgement, not a confidence interval.
- Not a fix. It names the layer; what to do about it is the next conversation.
- Not financial advice, and it grades launches rather than people.
Scope note: the floors and the minimum denominator are your numbers, so the verdicts are relative to your own expectations rather than to anyone’s benchmark. The minimum is a judgement about how few people you would accept a rate from, not a statistical confidence threshold. No verdict guarantees any revenue or growth outcome. This is not financial advice.
Pairs well with
The Inspection names the layer. These three are what you do about it.
When the verdict is READ BUT UNMOVED, the problem is the offer and the page it sits on. This is the treatment for that diagnosis.
ViewWhen the verdict is REACHED BUT UNREAD, they arrived and left. This is the quality guardrail for what they landed on.
ViewThis inspection tells you which counts cannot support a conclusion. That kit is the wider methodology for deciding when a result is reportable at all.
ViewCommon questions
The mechanics, stated plainly.
One spreadsheet
Name the layer.
- Three tabs, 30 rows, live formulas, a worked seven-launch example.
- A funnel-validity check, so a count that widens cannot pass unnoticed.
- Opens in Excel, Google Sheets or Numbers. Nothing to install.
- Deterministic and offline. It grades launches, never people.
Sold by RedHub AI LLC · Secured by Stripe · redhub.ai