Skip to content

fix(explorer): label each verifier's results and show check descriptions and penalties #395

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
polakamtejas wants to merge 3 commits into
base: main
Choose a base branch
from
Open

fix(explorer): label each verifier's results and show check descriptions and penalties #395

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e84ca44
fix(explorer): label each verifier's results and show check descripti…
polakamtejas Oct 29, 2026
8cb49b1
fix(explorer): say what a penalty check subtracts as weighted_average…
polakamtejas Nov 3, 2026
f91cce6
Merge branch 'main' into fix/explorer-verifier-panel-labels
polakamtejas Nov 3, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 22 additions & 3 deletions src/agent_env/explorer/ui/src/components/task-instance-viewer.tsx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -490,7 +490,9 @@ export function TaskInstanceViewer({
id: string;
score: number;
result: boolean;
message: string;
message?: string;
description?: string;
weight?: number;
title?: string;
justification?: string;
}[];
Expand Down Expand Up @@ -1084,6 +1086,9 @@ export function TaskInstanceViewer({
>
Score: {(v.score * 100).toFixed(0)}%
</span>
<code className="text-[10px] font-mono text-[var(--muted-foreground)]">
{verifierId}
</code>
</div>

{results.map((r, i) => (
Expand All @@ -1107,6 +1112,14 @@ export function TaskInstanceViewer({
<span className="text-sm font-semibold">
{(r.id ?? `check ${i + 1}`).replace(/_/g, ' ')}
</span>
{typeof r.weight === 'number' && r.weight < 0 && (
<span
className="text-[10px] font-semibold uppercase px-1.5 py-0.5 rounded bg-amber-500/10 text-amber-600"
title={`Penalty check (weight −${Math.abs(r.weight)}): weighted_average subtracts (1 − score) × ${Math.abs(r.weight)} from the weighted sum, ${Number(penaltySubtracted(r.score, r.result, r.weight).toFixed(2))} here`}
>
penalty −{Math.abs(r.weight)}
</span>
)}
{typeof r.score === 'number' && r.score !== 0 && r.score !== 1 && (
<span className="text-xs text-[var(--muted-foreground)]">
score:{' '}
Expand All @@ -1116,9 +1129,9 @@ export function TaskInstanceViewer({
</span>
)}
</div>
{r.message ? (
{(r.message ?? r.description) ? (
<p className="text-xs text-[var(--muted-foreground)] whitespace-pre-wrap">
{r.message}
{r.message ?? r.description}
</p>
) : null}
</div>
Expand Down Expand Up @@ -1387,6 +1400,12 @@ function AggregateScoreCard({
);
}

/** What weighted_average subtracts for a penalty check, as scoring.py computes it: (1 − score) × |weight|, the score
* the check's own or else 1 when it passed and 0 when it didn't. */
function penaltySubtracted(score: number | undefined, result: boolean, weight: number): number {
return (1 - (typeof score === 'number' ? score : result ? 1 : 0)) * Math.abs(weight);
}

/** Sandbox-side verifier outcomes from a `verify_sandbox` step. Per-criterion rows show the criterion
* `type` (probe_file_exists, bash_cmd_succeeds, …) and its `paths`/`cmd`. */
// Sandbox verifier — justification-string parsers. Each probe type emits its own `justification` format:
Expand Down
Loading