FEAT: Scan results redesign and live scan attacks in Chat

doc/gui/0_gui.md

@@ -118,14 +118,14 @@ Decided attack outcomes use **True** and **False**; execution errors remain sepa
 Groups and objectives show success rates and completion counts separately.
 The results page does not include a paired-outcome comparison.
 
-**Watch live conversation** opens saved messages while an attack runs. This view
-is read-only and refreshes saved messages; it has no send controls. **Cancel scan**
-stops remaining scan work and keeps saved results. Ordinary manual chats do not
-have scan cancellation controls.
-After cancellation, saved messages remain available in the read-only chat.
-Saving these conversation links does not add attack outcomes or change success rates.
-
-**Rerun** opens the scenario launch page with saved settings filled in. It does
+**Watch live conversation** opens a running attack in Chat. While the scan runs it,
+Chat is read-only, shows **Attack in progress** in place of the prompt bar, and
+refreshes its messages and conversations; when the attack finishes, Chat shows its
+saved result on the same page. **Cancel** stops remaining scan work and keeps saved
+results; saved messages remain available in the read-only chat. Opening these
+conversations does not add attack outcomes or change success rates.
+
+**Rerun with changes** opens the scenario launch page with saved settings filled in. It does
 not start work or change the original run. Review or edit the settings before
 launching. Recorded target/scorer identities and the scenario version are checked.
 Settings that cannot be restored, including older records without a required identity,

frontend/README.md

@@ -118,8 +118,59 @@ rows and keyboard focus. Selected conversation URLs, return-to-scan links and
 the originating History filters survive paging and reloads.
 On narrow screens, the saved viewer scrolls as one surface so loaded evidence
 and continuation controls remain reachable without squeezing the transcript.
-Selecting another saved conversation starts at its beginning; continuation
-keeps the current reading position.
+There, selecting another saved conversation starts at its beginning;
+continuation keeps the current reading position.
+
+## Scan results and live scan attacks
+
+A scan's run page shows its Operation as parent context above the title. Errors
+and scan-level notices (paused updates, overload, an incomplete plan, unmatched
+executions) stay explicit; a single attack's own error is not repeated as a run
+error. While the scan is active, the progress area shows elapsed time, an
+estimate of the time remaining, and **View live attacks**. When the scan
+finishes, the progress area collapses in place without moving the results.
+
+A run of one atomic attack shows that attack directly as one card, from
+preparing through completion. Other runs list atomic attack groups, datasets
+and objectives, and the scenario configuration; each result row opens the
+attack details. Attack details list the objective and outcome (with its score),
+the technique, objective scorer and dataset identities, and any findings that
+link the attack, followed by its conversation and **Attach to findings**.
+
+Every scan conversation opens in Chat at
+`/attack/chat/{attack_id}/conversations/{conversation_id}?scenarioResultId={run_id}`,
+whether the attack is still running or finished. Older
+`/scanner/history/{run_id}/conversations/{conversation_id}` links redirect there.
+While a scan is running the attack, Chat replaces the prompt bar with an
+**Attack in progress** bar and a link back to the scan, keeps the attack
+read-only, and refreshes its metadata, transcript, and conversation list at the
+scan's polling interval. A failed refresh keeps the loaded messages and offers
+**Retry**. When the attack finishes, Chat reloads its saved result in place on
+the same URL.
+
+An interrupted scan can retain recorded conversations without a saved attack
+result. Chat verifies that run, attack and conversation together and reads the
+stored evidence through the run-scoped API, including after reload or an older
+scan link. These conversations stay read-only; no result, score, main
+conversation or message count is invented. Result-dependent actions remain
+unavailable.
+
+Live and manual transcripts follow new messages only while the reader is within
+48 px of the end; scrolling further up keeps the reading position. Historical
+saved evidence keeps its bounded paging and narrow-screen reading behavior
+described above.
+
+**Evaluation metrics** in attack details shows the official evaluation quality
+of the recorded objective-scorer configuration, including configurations no
+longer in the registry. This is separate from the individual attack score in
+the outcome popover and the scan's attack-success rate. Missing evaluation
+metrics have an explicit unavailable state and a link to the scorer-metrics guide.
+
+The Scorer Registry keeps its compact comparison table; each scorer's
+**Evaluation metrics** opens the complete supplied metrics, including standard
+errors and sample counts. Runtime identity and behavioral evaluation identity
+remain separate. Registry names are not used to substitute metrics for a
+recorded run.
 
 ## Dataset catalog