unpitchdocs
Troubleshooting

When results and your draft disagree

Understand why findings dim after an edit, what a partial read is claiming, and how to avoid pasting a rewrite over work you have since done.

You have a verdict or a reaction on screen, you have carried on writing, and now the two do not match. Sometimes the panel says so and sometimes it only hints. This page covers the three ways a result and a draft come apart, and what each one means for the rewrite you are about to do.

Results from a version you have moved past

Both Quality Check and Simulation freeze the exact text they were given and describe that text forever. Edit the draft afterwards and a strip is pinned under the result reading Document changed with Results are from the previous version. The inline highlights in the draft dim at the same time.

Nothing has gone wrong. The findings are still true of the version that was read. They just no longer point at words that exist, which is why the highlights fade rather than moving.

The strip's button is Re-run. In Quality Check it re-reads the current saved text. In Simulation it starts a fresh run for 10 credits, and after a page reload it degrades to Back to readers, because the app can no longer re-admit the reader that was frozen into the old run. Pick that reader again from the chooser to run them on the new text.

Copy with all feedback copies the version that was read, not what is in the editor now. The footer says which: the Simulation scope line changes from Tested message to Simulated version. If you paste that package into an assistant and then paste its rewrite back over your draft, every edit you made since the run is quietly gone.

Reads that did not cover the whole draft

A Quality Check can stop before it reaches the end of your text. When it does, a scissors line under the headline says how far it got, followed by The read was cut off partway through your draft. and a scope warning: Everything below — including anything marked clear — is only about the part that was read.

The single most useful consequence is at the top. A clean verdict is retitled from Reads clean to Clean in what was read, and the Checked and clear section stops being a claim about the whole draft. A pinned strip headed Only part of the draft was read offers the fix, and it is an honest one: The reader stopped early. Running it again may reach further.

A related state has its own strip, Part of the read was lost. Here the read did cover the draft, but some of what it noticed could not be recorded, and a line says how many observations were dropped. Re-running is again the only remedy, and again it may land differently.

Findings that quote text you cannot find

Two smaller mismatches confuse people more than they should.

  • A finding can quote real words and still highlight nothing. When a phrase repeats and the app cannot be certain which occurrence a finding meant, it paints no highlight at all rather than marking the wrong one. The finding is still valid.
  • With more than twenty findings, only the twenty most severe are listed up front and the rest sit in a drawer labelled with a count, such as 7 more findings · the rest of this read. Step headers then read like 2 of 8 annotations, because the draft still highlights all eight even though the panel is showing two.

Copying can also refuse on an old result, with a message ending and feedback cannot be copied from this check. Run a new check to copy them. Run a fresh check and the package copies normally.

Where the old version still lives

Every Quality Check freezes a version, so running one adds a snapshot to the document's history, and the findings stay attached to that snapshot. Open the history rail and you can read the verdict or the reaction exactly as it was, including its Copy with all feedback footer, whose scope line names the checked version rather than your current draft.

History is read-only throughout. There is no Run or Re-run anywhere in it, and a snapshot can also honestly report that it has nothing: No Quality Check for this version, or No Simulation for this version with No saved reader reaction is attached to this snapshot.

The same freezing explains one thing on the Personas screen. A persona's What they did with it section shows the lines that persona actually reacted to during a completed run, not the document as it stands, so it will not move when you edit the draft afterwards.

Read the findings against the version they were made on, not against what is in front of you now.

Make your edits, using Copy with all feedback immediately if you want the package, before the draft moves further.

Press Re-run when the text has changed enough that you want a verdict about it, remembering that a Simulation re-run costs another 10 credits.