Why does Storskarvia say No issues found when the checker flagged something?

Every claim the checker judges contradicted or misleading goes to a second reviewer, the Skeptic, whose only job is to argue the manuscript was right. If it wins that argument for every candidate, nothing reaches the Facts tab and you get the clean line, for example "No issues found in 38 checked claims." The number counts every claim that was checked, including the ones whose flags the Skeptic overturned.

Can I see the findings the Skeptic threw out?

No. An overturned finding is never stored, and neither is the Skeptic's one-sentence reason, so there is no hidden list to open. If you suspect a detail, research it directly from the Research library, or run the check again after the passage has changed.

Does the Skeptic change a finding's confidence percentage?

No. The Skeptic decides only whether a finding survives. The percentage on the card comes from the earlier judgement and how well the evidence corroborated it, and it can still be halved afterwards if the explanation quotes a figure none of the sources contain.

Chapter 37 · The Staff

The Skeptic: Why You See Fewer Findings Than It Found

The Skeptic is the member of staff who argues for your manuscript. Every time the fact-checker decides a sentence is wrong, the Skeptic is handed the case and told to prove the sentence was fine. A finding reaches the Facts tab only if that attempt fails. This chapter covers what it sees, what it is allowed to argue, and what happens to the findings it wins.

Where it sits in a check

A fact-check runs in four stages, and the Skeptic is the last of them.

Stage	What happens
1. Extract	Real-world claims are pulled from the document, up to 150 per check.
2. Gather	Each claim is looked up in Wikipedia and web search.
3. Judge	Each claim is ruled supported, contradicted, misleading or insufficient against that evidence.
4. The Skeptic	Every contradicted or misleading ruling is argued against. Survivors become cards.

Claims ruled supported or insufficient never reach the Skeptic. There is nothing to defend you from. Running a Fact-Check covers the first three stages in detail.

What it is handed

For each candidate finding the Skeptic receives five things, and only these:

Your sentence, quoted from the draft.
The claim as the checker phrased it, already set against the period your pages establish.
The ruling: contradicted or misleading.
The checker's explanation.
The evidence the checker used.

It does not search again. It works from the same sources the checker read, so it cannot rescue a sentence by finding a better page than the one already on the table. If the right answer lives somewhere the search never reached, the Skeptic does not know about it either.

What counts as a defence

The Skeptic works under a strict rule: it may overturn a finding only when the evidence in front of it supports the defence. A plausible story is not enough.

Overturns the finding	Does not
The checker misread the claim.	"Perhaps it aired as a rerun somewhere."
The evidence itself contradicts the checker.	"The sentence never names a year."
The evidence shows your detail is correct as written.	"It could refer to a different time."
A round number or figure of speech whose ordinary reading fits the evidence.	A prototype, test flight or technicality that the ordinary reader would not mean.
The line is a character's hope, prediction or opinion.	Any defence the evidence does not back.

The middle rows of the right-hand column matter for fiction. The question of which year a scene takes place in was settled when the claim was written, using the dates in your text, and the Skeptic is not allowed to reopen it. A radio playing a 1963 single in a chapter headed June 1962 cannot be defended on the grounds that the sentence never says which summer.

The Skeptic has to rule explicitly that a finding stands. An answer that comes back without that ruling counts as overturned, so the pass leans toward saying less.

What it throws away, and where that goes

An overturned finding is discarded entirely. Storskarvia keeps neither the finding nor the Skeptic's reason for killing it, so no hidden list exists and no setting reveals one.

That explains the most common surprise with the Facts tab. The count in the clean line, as in "No issues found in 38 checked claims.", is every claim the check examined. A run where the checker flagged four of those and the Skeptic overturned all four looks identical to a run where nothing was ever flagged.

The trade is deliberate. A checker that raises every doubt produces a list you stop reading by the second page. Storskarvia would rather miss an arguable detail than put a wrong accusation in front of you, and the cost of that choice is that a real error with a plausible, sourced defence can go quiet.

A clean result is only as good as its sourcing. When the Weaker sourcing than usual note sits above the list, the Skeptic had thinner evidence to work with, and the absence of findings means less.

What it does not touch

The Skeptic decides survival and nothing else.

It does not edit the card. The explanation, sources and problem pill are the checker's, word for word.
It does not set the percentage. Confidence comes from how well the evidence corroborated the original ruling.
It does not replace the figure check. After the Skeptic, every year and number in a surviving explanation is looked for in the sources. Any that are missing get listed in a bracketed note and the confidence is halved. Reading a Finding shows that note.
It never touches your text. Like the rest of the staff, the Skeptic reads and rules. Storskarvia never writes a word for you.

When the Skeptic cannot answer

The Skeptic is a model call like the checker, and calls occasionally fail. Each one is tried up to three times. If a candidate still gets no answer, the whole check stops, because handing you a list with one finding silently unreviewed would break the promise the list makes.

The Facts tab shows "Check failed" with the reason, and the checks it cost go back to your allowance. Wait a minute and press Check Facts again.

When you disagree with the result

Situation	What to do
A card survived and you know the line is right	Press Dismiss. The Skeptic had no sourced defence, which is different from your sentence being wrong.
You expected a flag and got a clean line	Highlight the passage and ask the Research library directly. That answers your question with citations and costs one check.
You want a second opinion on the whole document	Change the passage first. Search results are kept for 14 days, so re-running an unchanged document inside that window mostly reads the same evidence and reaches the same rulings. Every run is charged again, one check per 10,000 words.

No switch turns the Skeptic off. Every check from the Check Facts button runs it.

Agent checks and continuity

A check submitted by your own AI agent carries the Checked by your own AI agent note. Storskarvia gives the agent its sources and tells it to argue against its own findings and drop any it cannot defend with a cited source, but the agent runs that step itself, and Storskarvia cannot confirm it did. Read those lists as a first pass.

The continuity editor has a reviewer of its own with different rules: it asks whether two conflicting statements can be reconciled, such as dyed hair or a deliberate change in the story. That pass belongs to continuity checks, covered later in this Part. How the allowance behind all of this is counted is set out in Part IX.

Updated 2026-09-13 · Full contents · Something wrong or missing?

Start writing free