Spaces:
Running
Running
Shared page code: labels from data (for DeepSWE annotated)
Browse files- common.js +1 -1
- home.js +14 -3
- hub-link.js +1 -1
- row.js +6 -5
- style.css +1 -0
common.js
CHANGED
|
@@ -76,7 +76,7 @@
|
|
| 76 |
}
|
| 77 |
const problemChip = p => { const c = el("span", "fl fl--" + (p.level === "bad" ? "bad" : "warn"), p.text); c.title = p.help; return c; };
|
| 78 |
|
| 79 |
-
/* The attempt strip: one column per task,
|
| 80 |
function strip(r, opts = {}) {
|
| 81 |
const s = el("div", "strip" + (opts.large ? " strip--lg" : ""));
|
| 82 |
D.tasks.forEach(task => {
|
|
|
|
| 76 |
}
|
| 77 |
const problemChip = p => { const c = el("span", "fl fl--" + (p.level === "bad" ? "bad" : "warn"), p.text); c.title = p.help; return c; };
|
| 78 |
|
| 79 |
+
/* The attempt strip: one column per task, a square per attempt. */
|
| 80 |
function strip(r, opts = {}) {
|
| 81 |
const s = el("div", "strip" + (opts.large ? " strip--lg" : ""));
|
| 82 |
D.tasks.forEach(task => {
|
home.js
CHANGED
|
@@ -55,6 +55,7 @@
|
|
| 55 |
return out;
|
| 56 |
}
|
| 57 |
const hasLb = D.rows.some(r => r.lb.source !== "recount");
|
|
|
|
| 58 |
const ESTIMATE_HELP = "our estimate: rewarded trials, less review leads (an unexcused high or critical finding) and trials partly run by a fallback model";
|
| 59 |
|
| 60 |
const compactN = n => (n >= 1e6 ? (n / 1e6).toFixed(1) + "M" : n >= 1e3 ? Math.round(n / 1e3) + "k" : String(n));
|
|
@@ -67,8 +68,8 @@
|
|
| 67 |
};
|
| 68 |
function board() {
|
| 69 |
const cols = [
|
| 70 |
-
{ name: "
|
| 71 |
-
{ name: hasLb ? "Leaderboard" : "Score", sort: "score", help: hasLb ? "published
|
| 72 |
{ name: "Our estimate", sort: "adj", help: ESTIMATE_HELP },
|
| 73 |
{ name: "Leads", sort: "leads", help: "rewarded trials with an unexcused high or critical finding" },
|
| 74 |
{ name: "Notes", cls: "l" },
|
|
@@ -81,7 +82,7 @@
|
|
| 81 |
tr.append(rowHead(r));
|
| 82 |
const s = el("td", "l"); s.append(strip(r, { leadsOnly: leadsOnly.checked })); tr.append(s);
|
| 83 |
const sc = el("td"); sc.append(num(fmt1(r.lb.accuracy), "n big"), el("span", "pm", "±" + fmt1(r.lb.ci95)));
|
| 84 |
-
sc.title = r.scan.rewarded + " of " + r.scan.trials + " trials rewarded"; tr.append(sc);
|
| 85 |
const ad = el("td");
|
| 86 |
const delta = adj.accuracy - r.lb.accuracy;
|
| 87 |
ad.append(num(fmt1(adj.accuracy), "n big" + (delta <= -1 ? " hot" : "")));
|
|
@@ -292,5 +293,15 @@
|
|
| 292 |
notes.append(el("dt", null, r.agent), el("dd", null, note));
|
| 293 |
});
|
| 294 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 295 |
board(); accounting(); detail(); families(); checks();
|
| 296 |
})();
|
|
|
|
| 55 |
return out;
|
| 56 |
}
|
| 57 |
const hasLb = D.rows.some(r => r.lb.source !== "recount");
|
| 58 |
+
const SITE = { attempts: 5, ...(D.site || {}) };
|
| 59 |
const ESTIMATE_HELP = "our estimate: rewarded trials, less review leads (an unexcused high or critical finding) and trials partly run by a fallback model";
|
| 60 |
|
| 61 |
const compactN = n => (n >= 1e6 ? (n / 1e6).toFixed(1) + "M" : n >= 1e3 ? Math.round(n / 1e3) + "k" : String(n));
|
|
|
|
| 68 |
};
|
| 69 |
function board() {
|
| 70 |
const cols = [
|
| 71 |
+
{ name: D.tasks.length + " tasks × " + SITE.attempts + " attempts, easiest first", cls: "l" },
|
| 72 |
+
{ name: hasLb ? "Leaderboard" : "Score", sort: "score", help: hasLb ? "published score ± 95% half-width" : "rewarded trials ± 95% half-width over the attempt rounds" },
|
| 73 |
{ name: "Our estimate", sort: "adj", help: ESTIMATE_HELP },
|
| 74 |
{ name: "Leads", sort: "leads", help: "rewarded trials with an unexcused high or critical finding" },
|
| 75 |
{ name: "Notes", cls: "l" },
|
|
|
|
| 82 |
tr.append(rowHead(r));
|
| 83 |
const s = el("td", "l"); s.append(strip(r, { leadsOnly: leadsOnly.checked })); tr.append(s);
|
| 84 |
const sc = el("td"); sc.append(num(fmt1(r.lb.accuracy), "n big"), el("span", "pm", "±" + fmt1(r.lb.ci95)));
|
| 85 |
+
sc.title = r.scan.rewarded + " of " + r.scan.trials + " trials rewarded" + (r.lb.passAt4 ? " · pass@4 " + r.lb.passAt4 + "%" : ""); tr.append(sc);
|
| 86 |
const ad = el("td");
|
| 87 |
const delta = adj.accuracy - r.lb.accuracy;
|
| 88 |
ad.append(num(fmt1(adj.accuracy), "n big" + (delta <= -1 ? " hot" : "")));
|
|
|
|
| 293 |
notes.append(el("dt", null, r.agent), el("dd", null, note));
|
| 294 |
});
|
| 295 |
|
| 296 |
+
if (!notes.children.length) { notes.previousElementSibling.hidden = true; notes.hidden = true; }
|
| 297 |
+
|
| 298 |
+
/* Counts in the lede, and leaderboard rows without published trajectories. */
|
| 299 |
+
const fill = (sel, text) => document.querySelectorAll(sel).forEach(n => { n.textContent = text; });
|
| 300 |
+
fill("[data-rows]", String(D.rows.length));
|
| 301 |
+
fill("[data-trials]", D.rows.reduce((n, r) => n + r.scan.trials, 0).toLocaleString("en"));
|
| 302 |
+
const missing = SITE.unpublished || [];
|
| 303 |
+
fill("[data-unpublished]", missing.length ? "Not here, as Datacurve publishes no trajectories for them: " +
|
| 304 |
+
missing.map(u => "#" + u.rank + " " + u.model + " · " + u.effort + " (" + u.accuracy + "%)").join(", ") + "." : "");
|
| 305 |
+
|
| 306 |
board(); accounting(); detail(); families(); checks();
|
| 307 |
})();
|
hub-link.js
CHANGED
|
@@ -10,7 +10,7 @@
|
|
| 10 |
a.className = "annot-hub";
|
| 11 |
a.target = "_blank";
|
| 12 |
a.rel = "noopener noreferrer";
|
| 13 |
-
a.textContent = "Trial on Harbor Hub ↗";
|
| 14 |
a.hidden = true;
|
| 15 |
nav.append(a);
|
| 16 |
let last = null;
|
|
|
|
| 10 |
a.className = "annot-hub";
|
| 11 |
a.target = "_blank";
|
| 12 |
a.rel = "noopener noreferrer";
|
| 13 |
+
a.textContent = window.TB21_HUB_LABEL || "Trial on Harbor Hub ↗";
|
| 14 |
a.hidden = true;
|
| 15 |
nav.append(a);
|
| 16 |
let last = null;
|
row.js
CHANGED
|
@@ -12,7 +12,7 @@
|
|
| 12 |
const hubScript = document.createElement("script");
|
| 13 |
hubScript.src = rowPath(r) + "hub.js";
|
| 14 |
document.head.append(hubScript);
|
| 15 |
-
document.title = "#" + r.rank + " " + title(r) + " · " + r.agent + " · " + (r.pr ? "TB2.1 submissions" : "TB2.1 annotated");
|
| 16 |
const h = r.scan.holes, n = r.scan.trials;
|
| 17 |
const full = Object.values(trialInfo[r.id]).filter(t => t.h).length;
|
| 18 |
|
|
@@ -37,7 +37,7 @@
|
|
| 37 |
el("small", null, OUTCOME[t.c] || ""));
|
| 38 |
if (t.h) li.append(link(fullHref(r, vid), "full ↗", "full"));
|
| 39 |
const hub = (window.TB21_HUB || {})[vid];
|
| 40 |
-
if (hub) { const a = link(hub, "Hub ↗", "full"); a.target = "_blank"; a.rel = "noopener noreferrer"; li.append(a); }
|
| 41 |
ol.append(li);
|
| 42 |
});
|
| 43 |
d.append(ol);
|
|
@@ -62,12 +62,13 @@
|
|
| 62 |
/* Header. */
|
| 63 |
const head = el("header", "row-head");
|
| 64 |
head.append(el("p", "kicker", (r.pr ? "Terminal-Bench 2.1 submission · PR #" + r.pr + " (" + r.prState.toLowerCase() + ")"
|
| 65 |
-
: "#" + r.rank + " on the Terminal-Bench 2.1 leaderboard") + " · " +
|
| 66 |
head.append(el("h1", null, title(r)));
|
| 67 |
const meta = el("p", "row-meta");
|
| 68 |
meta.append(el("span", "harn", r.agent), el("span", null, [r.modelOrg].filter(Boolean).join(" · ")));
|
| 69 |
-
meta.append(link(r.source, r.pr ? "PR #" + r.pr + " ↗" : "Submission ↗", "ext"));
|
| 70 |
-
if (r.jobs.length
|
|
|
|
| 71 |
else meta.append(el("span", null, r.jobs.length + " Hub jobs: each trial links its own"));
|
| 72 |
head.append(meta);
|
| 73 |
if (r.why) head.append(el("p", "why-line", r.why));
|
|
|
|
| 12 |
const hubScript = document.createElement("script");
|
| 13 |
hubScript.src = rowPath(r) + "hub.js";
|
| 14 |
document.head.append(hubScript);
|
| 15 |
+
document.title = "#" + r.rank + " " + title(r) + " · " + r.agent + " · " + (r.pr ? "TB2.1 submissions" : (D.site || {}).short || "TB2.1 annotated");
|
| 16 |
const h = r.scan.holes, n = r.scan.trials;
|
| 17 |
const full = Object.values(trialInfo[r.id]).filter(t => t.h).length;
|
| 18 |
|
|
|
|
| 37 |
el("small", null, OUTCOME[t.c] || ""));
|
| 38 |
if (t.h) li.append(link(fullHref(r, vid), "full ↗", "full"));
|
| 39 |
const hub = (window.TB21_HUB || {})[vid];
|
| 40 |
+
if (hub) { const a = link(hub, (D.site || {}).trialLabel || "Hub ↗", "full"); a.target = "_blank"; a.rel = "noopener noreferrer"; li.append(a); }
|
| 41 |
ol.append(li);
|
| 42 |
});
|
| 43 |
d.append(ol);
|
|
|
|
| 62 |
/* Header. */
|
| 63 |
const head = el("header", "row-head");
|
| 64 |
head.append(el("p", "kicker", (r.pr ? "Terminal-Bench 2.1 submission · PR #" + r.pr + " (" + r.prState.toLowerCase() + ")"
|
| 65 |
+
: "#" + r.rank + " on the " + ((D.site || {}).board || "Terminal-Bench 2.1 leaderboard")) + (r.date ? " · " + r.date : "")));
|
| 66 |
head.append(el("h1", null, title(r)));
|
| 67 |
const meta = el("p", "row-meta");
|
| 68 |
meta.append(el("span", "harn", r.agent), el("span", null, [r.modelOrg].filter(Boolean).join(" · ")));
|
| 69 |
+
meta.append(link(r.source, r.pr ? "PR #" + r.pr + " ↗" : (D.site || {}).sourceLabel || "Submission ↗", "ext"));
|
| 70 |
+
if (!r.jobs.length) { /* trials link their own pages */ }
|
| 71 |
+
else if (r.jobs.length <= 3) r.jobs.forEach(j => meta.append(link("https://hub.harborframework.com/jobs/" + j, "Hub job ↗", "ext")));
|
| 72 |
else meta.append(el("span", null, r.jobs.length + " Hub jobs: each trial links its own"));
|
| 73 |
head.append(meta);
|
| 74 |
if (r.why) head.append(el("p", "why-line", r.why));
|
style.css
CHANGED
|
@@ -242,3 +242,4 @@ details.trials summary .label { font-weight: 650; }
|
|
| 242 |
.notes .why { display: block; color: var(--muted); font-weight: 500; }
|
| 243 |
.why-line { margin: 6px 0 0; color: var(--muted); font-size: 14px; }
|
| 244 |
.key .mx, .explain-line .mx { display: inline-block; width: 14px; height: 8px; border-radius: 2px; vertical-align: 0; margin-right: 3px; }
|
|
|
|
|
|
| 242 |
.notes .why { display: block; color: var(--muted); font-weight: 500; }
|
| 243 |
.why-line { margin: 6px 0 0; color: var(--muted); font-size: 14px; }
|
| 244 |
.key .mx, .explain-line .mx { display: inline-block; width: 14px; height: 8px; border-radius: 2px; vertical-align: 0; margin-right: 3px; }
|
| 245 |
+
.canary { font: 500 10px var(--code-f); color: var(--muted); letter-spacing: .02em; }
|