Chapter 32: Who Stayed
The readers who found me were fewer; the readers who stayed had reached for Carys, not the corpse.
Later on Tuesday, PageTurn opened Saltmere's anonymous funnel in the same clean-room session. The report remained labelled EARLY COHORT.
No names, email addresses or payment cards sat behind the bars.
I could see where grouped readers entered, paused and left. No bar represented fewer than the platform's privacy threshold.
The opposing project appeared only as a comparison band.
Its individual users and comments remained closed.
My strongest continuation point came immediately after Carys submitted the correction under her own staff number. The heat line rose again at the audit alert.
The second rose when she carried the ledger through the exit.
The corpse reveal I had planned for episode three did not yet exist on the live page.
It dominated my draft anyway.
I had built a storm-drain discovery, a concealed injury report and an elegant false identity.
The sequence was efficient enough to sell in one sentence. My marketing notes had already placed it in bold.
It also moved Carys back into the role of competent observer.
The sample contained too few readers for prophecy.
It contained enough behaviour for a revision test. I could change my own draft without pretending the sample spoke for everyone.
I opened SALTMERE_EP3_V1 beside the funnel.
The corpse appeared at word three hundred and twelve.
Carys's next decision appeared at word nine hundred.
I selected the reveal.
For several seconds, the scene remained blue and commercially reassuring. Removing it felt like throwing away the easiest trailer.
Then I moved it to the later-story file.
Episode three lost its high-concept surprise. The word count fell, and the empty space forced Carys forwards.
In the new space, Carys returned to work without the ledger.
Her closest colleague asked why her coat had torn at the lining.
Carys could tell the truth and lose the only person still speaking to her.
Instead, she hid the ledger in a disused dues cupboard and lied about the tear.
The lie protected the record and damaged the relationship. It also gave her colleague a reason to act in the next scene.
That consequence belonged to her rather than the puzzle machinery.
I wrote the revision reason into my independent editorial log.
ANONYMOUS FUNNEL: CONTINUATION RISES AFTER CARYS CHOOSES.
I did not call it proof that readers preferred character to mystery. That claim would turn a small cohort into a slogan.
I called it the testable reason for version two.
The platform generated a new heat map when I uploaded the internal revision sample.
The Carys sections held attention more evenly than the removed reveal.
The result was provisional, not miraculous. PageTurn preserved the test group separately from live reader behaviour.
I signed the change and accepted the smaller promotional hook.
If readers wanted another body immediately, I had just made the book harder to pitch. The removed reveal stayed in my later-story file, not the live episode.
If they wanted Carys, I had finally stopped interrupting her.
At 16:18, PageTurn archived my funnel export and the editing rationale separately.
The raw grouped data remained with the platform. I could not export device IDs, contact fields or individual paths.
My log retained only percentages, version references and the decision I had made.
A yellow triangle appeared on both project summaries.
TRAFFIC QUALITY REVIEW.
The new source had produced a sudden burst of clicks after 15:57. Its volume rose too quickly to match the normal discovery curve.
Its median reading time was four seconds.
The same source line appeared against Larkspur through the equal-summary view.
Neither team could see the underlying visitors while the review was open.
PageTurn paused the source before it could swell either headline number.
The alert named no buyer, author or suspected culprit.
It said only that one abnormal referrer was feeding low-duration clicks into both books.