Chapter 36 · The Staff
Reading a Finding
A finding is a card in the rail on the right of the editor, and every part of it answers one question: should you change the sentence? This chapter reads a card top to bottom, explains the pills on it, and covers what the buttons do and do not do.
The card, top to bottom
A fact-check finding has six parts, in this order:
| Part | What it is |
|---|---|
| Problem pill | CONTRADICTED or MISLEADING, in small capitals. |
| Confidence pill | A percentage, coloured green, amber or red. |
| Quote | Your own sentence, in the manuscript's serif, in quotation marks. |
| Explanation | What the evidence says instead. |
| Sources | Up to three links, one per line. |
| Buttons | Jump, Resolve, Dismiss. |
Nothing on the card edits your document. Storskarvia never writes a word for you, and that holds here more plainly than anywhere else: the checker found the problem, quoted it, and stopped.
Contradicted or misleading
Only two verdicts ever reach you from a fact-check, and only after the Skeptic failed to argue them away.
- Contradicted. The evidence says the opposite. A date, a name, a figure, or an attribution that the sources put somewhere else.
- Misleading. Defensible as written and wrong in effect. True of a different year, true of a different country, or true in a way a reader will not take it.
A claim the checker could not settle never becomes a card. Verdicts of "supported" and "insufficient" are dropped before the list is built, which is why a check on forty claims hands back three findings and no summary of the other thirty-seven.
The percentage
The second pill is how well the correction was corroborated, not how strongly the sentence is worded. Hover it and the tooltip says so: Corroboration-calibrated: high needs two independent sources.
| Colour | Range | What stands behind it |
|---|---|---|
| Green | 80% and above | Two independent sites agreed. |
| Amber | 50% to 79% | Usually one source. A single site caps confidence near seventy. |
| Red | Below 50% | Sources disagreed with each other, or the correction's own figures could not be found in them. |
Treat a red pill as a lead worth ten minutes rather than a verdict, and a green one as something you would still want to have seen with your own eyes. Continuity findings carry no percentage at all: there is no outside evidence to corroborate them against, because they are your manuscript disagreeing with itself.
The quote, and Jump
The quote is lifted from your draft as it stood when the check ran. Jump finds it again: the passage is selected, scrolled to the middle of the window, and flashed so your eye lands on it.
The search ignores capital letters and works on the first 80 characters of the quote, inside a single paragraph. Two things follow:
- If you have already rewritten or deleted the sentence, Jump does nothing at all. No message, no error. There is simply nothing left to select. This is the most common confusion about the findings rail, and it usually means you have already fixed the thing.
- Splitting the sentence across a paragraph break after the check ran also puts it beyond reach. Find will still get you there.
The explanation
One or two sentences on what the evidence actually said. It is written to be checkable, so it names the year or the figure it is correcting to rather than telling you that something is wrong.
Sometimes it ends with a bracketed warning:
[Note: the specific figure(s) 1937, 12,400 could not be confirmed in the retrieved sources — verify before relying on them.]
That note guards against the worst thing a checker can do, which is hand you a confident replacement fact that nothing supports. Every year and number the explanation asserts is checked against the text of the retrieved sources. Any that do not appear there are listed, and the finding's confidence is halved. The flagged sentence may still be worth a look; the number offered in its place is not evidence.
The sources
Up to three links sit below the explanation, each opening in a new tab. Storskarvia keeps up to six per finding and shows the three it leaned on.
A finding you cannot trace is a rumour, and the fastest way to settle a card is to open the first link and read the paragraph the checker read. A card with no links at all came from a lane that returned nothing citable, and the sourcing note above the list will normally say so.
Continuity cards are a different shape
Continuity findings appear in their own Continuity tab, with a violet left edge and a CONTINUITY pill. They carry two quotes rather than one:
- The passage in question.
- A second line beginning conflicts with:, holding the earlier or later mention that disagrees with it.
The buttons change to match: Jump here and Jump to other, so you can put the two mentions side by side before deciding which one is wrong. The second button only appears when a conflicting passage was identified. No sources and no percentage appear on these cards.
Resolve, Dismiss, and the counters
Every finding is open, resolved, or dismissed. The latter two both fade the card to about half strength and leave it in the list, and neither changes a character of your manuscript.
| Button | Say it when |
|---|---|
| Resolve | You changed the draft. |
| Dismiss | The finding is wrong, or the line is deliberate. |
Storskarvia treats the two identically. The distinction is for you, and for whoever reads the project roll-up in three weeks.
The small number beside Facts and Continuity counts the findings still open in that tab, and only that tab, so a tab you are not looking at can say it still has work in it. It disappears at zero.
When the status will not save, the card stays open and the status bar says "That did not save. The finding is still open." for five seconds. This matters: the card used to grey out first and ask questions later, which produced a punch list a writer believed they had cleared and which came back untouched on the next open. If you see that line, the card is still open and your click did not land.
Statuses survive. Close the document and reopen it and the last check of each kind is loaded back into its tab with resolved and dismissed cards still faded. Running the same kind of check again replaces that tab wholesale with the new list, and the two tabs never clear each other.
The notes above the list
Up to three notes can sit above the cards, and each one changes how to read what is underneath.
| Note | What it changes |
|---|---|
| ⚠️ Weaker sourcing than usual | Retrieval ran degraded. Lower every card's weight, and do not read an empty list as a clean bill of health. |
| Checked the first 150 of… | The claim cap. Claims past that point were never looked at. |
| ⚠️ Checked by your own AI agent | The check came through the agent interface rather than Storskarvia's own pipeline. Storskarvia supplied the sources and your agent made the judgments, so the list is a first pass rather than the vetted result. |
| Checked without a Story Bible | Continuity only. It inferred the cast and places itself, which catches less. Building the Story Bible sharpens it. |
Every finding in a project at once
Working document by document is fine for one chapter and poor for a revision pass. On the project page, the third section, Revise, gathers every open finding and every unresolved comment across the whole binder, grouped by document, under a count reading 9 open findings · 4 unresolved comments · 3 documents.
Each row shows the problem, the confidence, the quote, and the start of the explanation. Clicking it opens that document, switches the rail to the right tab, and jumps to the passage, so a revision pass runs top to bottom without you keeping a list of your own.
The roll-up reads only the most recent completed check of each kind per document, so findings left behind by a superseded run do not haunt it.
The next chapter takes up the Skeptic, the reviewer that argues against every finding before you ever see it, and explains why the list in front of you is shorter than the one the checker produced.
Updated 2026-09-12 · Full contents · Something wrong or missing?