Spaces:

localgate
/

audit-calibration

Running

abullard1 commited on about 21 hours ago

Commit

8b45cc8

verified ·

1 Parent(s): 5da77ee

Deploy calibration @ d2b2018-space

Files changed (4) hide show

app.js CHANGED Viewed

@@ -24,8 +24,8 @@ import { oauthLoginUrl, oauthHandleRedirectIfPresent, uploadFiles, downloadFile,
 // the ?v= stamp — an unversioned nested import could load a stale module.
 import { keyOf, latestByKey, mergeEvents, orderItems, firstUngraded as
          firstUngradedPure, countComplete, stampClass, nextEpoch,
-         standsDownTo, statusOf, classifyProbe, hasScopes }
-  from "./logic.mjs?v=960f4e5-space";
 // Test seam: Playwright installs window.__testHub (an in-memory hub) before
 // any page script runs; production never defines it, so this is one inert
@@ -485,8 +485,11 @@ function renderItem() {
   state.onPushSuccess = null;
   const item = state.items[state.order[state.idx]];
   window.scrollTo(0, 0);                 // a new item always starts at its top
   main.append(el("div", "item-id",
-    `${keyOf(item)} · ${state.idx + 1}/${state.order.length}`));
   if (item.warmup) {
     main.append(el("p", "warmup-note",
       "Warm-up item — discussable with the others; real items start after these and must be graded independently."));

 // the ?v= stamp — an unversioned nested import could load a stale module.
 import { keyOf, latestByKey, mergeEvents, orderItems, firstUngraded as
          firstUngradedPure, countComplete, stampClass, nextEpoch,
+         standsDownTo, statusOf, classifyProbe, hasScopes, displayId }
+  from "./logic.mjs?v=d2b2018-space";
 // Test seam: Playwright installs window.__testHub (an in-memory hub) before
 // any page script runs; production never defines it, so this is one inert
   state.onPushSuccess = null;
   const item = state.items[state.order[state.idx]];
   window.scrollTo(0, 0);                 // a new item always starts at its top
+  const shownId = item.warmup && !CONFIG.opaqueIds
+    ? "warm-up"                          // raw corpus qids never reach the screen
+    : displayId(keyOf(item), state.idx, CONFIG.opaqueIds);
   main.append(el("div", "item-id",
+    `${shownId} · ${state.idx + 1}/${state.order.length}`));
   if (item.warmup) {
     main.append(el("p", "warmup-note",
       "Warm-up item — discussable with the others; real items start after these and must be graded independently."));

config.js CHANGED Viewed

@@ -2,7 +2,10 @@ window.CONFIG = {
   task: "calibration",
   title: "Judge calibration — 100 gradings",
   itemsPath: "calibration_items.json",
-  build: "960f4e5-space",
   landing: "You will grade whether a model's answer states the same fact as a "
     + "reference answer. 100 gradings plus 4 discussable warm-ups; expect 2–3 "
     + "hours total, in as many sittings as you like — progress saves continuously. "

   task: "calibration",
   title: "Judge calibration — 100 gradings",
   itemsPath: "calibration_items.json",
+  // Display-only: keys are "qid:rep", so repeats of one question would be
+  // recognisable on screen. Logs and scoring always use the true key.
+  opaqueIds: "C",
+  build: "d2b2018-space",
   landing: "You will grade whether a model's answer states the same fact as a "
     + "reference answer. 100 gradings plus 4 discussable warm-ups; expect 2–3 "
     + "hours total, in as many sittings as you like — progress saves continuously. "

index.html CHANGED Viewed

@@ -5,8 +5,8 @@
 <meta name="viewport" content="width=device-width, initial-scale=1">
 <meta http-equiv="Content-Security-Policy" content="default-src 'none'; script-src 'self'; style-src 'self'; style-src-attr 'unsafe-inline'; font-src 'self'; connect-src https://huggingface.co https://*.hf.co; img-src 'self' data:; base-uri 'none'; form-action 'none'">
 <title>LocalGate audit</title>
-<link rel="stylesheet" href="style.css?v=960f4e5-space">
-<link rel="stylesheet" href="katex/katex.min.css?v=960f4e5-space">
 </head>
 <body>
 <header>
@@ -22,9 +22,9 @@
 <pre id="rubric-body"></pre></details>
 </div>
 <div id="controls" hidden></div>
-<script src="katex/katex.min.js?v=960f4e5-space"></script>
-<script src="katex/auto-render.min.js?v=960f4e5-space"></script>
-<script src="config.js?v=960f4e5-space"></script>
-<script type="module" src="app.js?v=960f4e5-space"></script>
 </body>
 </html>

 <meta name="viewport" content="width=device-width, initial-scale=1">
 <meta http-equiv="Content-Security-Policy" content="default-src 'none'; script-src 'self'; style-src 'self'; style-src-attr 'unsafe-inline'; font-src 'self'; connect-src https://huggingface.co https://*.hf.co; img-src 'self' data:; base-uri 'none'; form-action 'none'">
 <title>LocalGate audit</title>
+<link rel="stylesheet" href="style.css?v=d2b2018-space">
+<link rel="stylesheet" href="katex/katex.min.css?v=d2b2018-space">
 </head>
 <body>
 <header>
 <pre id="rubric-body"></pre></details>
 </div>
 <div id="controls" hidden></div>
+<script src="katex/katex.min.js?v=d2b2018-space"></script>
+<script src="katex/auto-render.min.js?v=d2b2018-space"></script>
+<script src="config.js?v=d2b2018-space"></script>
+<script type="module" src="app.js?v=d2b2018-space"></script>
 </body>
 </html>

logic.mjs CHANGED Viewed

@@ -103,3 +103,12 @@ export const hasScopes = (grantedString, required) => {
   const granted = new Set(String(grantedString ?? "").split(/\s+/));
   return required.every((scope) => granted.has(scope));
 };

   const granted = new Set(String(grantedString ?? "").split(/\s+/));
   return required.every((scope) => granted.has(scope));
 };
+// On-screen item label. Some tasks must not reveal their key's structure —
+// calibration keys are "qid:rep", so five repeats of one question would be
+// recognisable and put the annotator under consistency pressure the design
+// otherwise avoids. The STORED key is always the real one; this is display
+// only, derived from the annotator's own seeded position so two repeats never
+// share a visible prefix.
+export const displayId = (key, position, opaque) =>
+  opaque ? `${opaque}-${String(position + 1).padStart(3, "0")}` : key;