feat: enhance diagnostic dashboard with structured trace and copy functionality

README.md

@@ -235,7 +235,7 @@ Under each score a teacher or a tutor gets an optional note and the button "🚩
 
 - who and what: `email`, `cat`, `phrase`, `he`, `note`;
 - the answer the scorer matched (`ar`, `ipa`), its free decode (`heard_ipa`), `score` and `unrelated`;
-- `result`: the whole score response (every word and phoneme with its score and what took its place, the tips, the issues);
+- `result`: the whole score response, including the structured diagnostic trace (every word and phoneme, pipeline stage, threshold, score, replacement, tip and issue);
 - `model` and `version`, so a flag can be checked against the model and build that produced it;
 - `audio`: always null for new reports; the nullable legacy column is retained so old reports with recordings remain readable.
 
@@ -281,7 +281,7 @@ alter table public.flags
   add column if not exists false_issues jsonb not null default '[]'::jsonb;
 ```
 
-The management panel's **דיווחים** tab lets a teacher confirm or reject each report, classify it, and select the individual sound judgements that were false. It can still play a legacy recording when an older row has one, but new reports deliberately retain only diagnostics. The tab groups confirmed `expected → heard` pairs by reports, reporters, and phrases. Only confirmed reports should be used for calibration; review and aggregation use the stored structured result without needing the user's voice file.
+The management panel's **דיווחים** tab lets a teacher open the existing diagnostic dashboard beside the approval action, confirm or reject each report, classify it, and select the individual sound judgements that were false. The dashboard includes a copyable, identity-free text payload for asking a chat assistant to distinguish acoustic-model, alignment, rule, threshold and feedback failures and suggest regression tests. It can still play a legacy recording when an older row has one, but new reports deliberately retain only diagnostics. The tab groups confirmed `expected → heard` pairs by reports, reporters, and phrases. Only confirmed reports should be used for calibration; review and aggregation use the stored structured result without needing the user's voice file.
 
 ### The lab (מעבדה)
 

app/main.py

@@ -2041,9 +2041,13 @@ def create_app(scorer: Scorer | None = None, tts: SawtakTTS | None = None,
         except ValueError:
             raise HTTPException(400, "could not decode audio")
         refs = app.state.refs.get(phrase_id, phrase) if app.state.refs else []
-        result = app.state.scorer.score(wav, phrase, refs)
+        # Keep a structured pipeline trace for the diagnostic dashboard; the audio itself is discarded.
+        result = app.state.scorer.score(wav, phrase, refs, diagnostics=True)
         if result.get("error"):
             raise HTTPException(400, result["message"])
+        # The trace already contains every value used by the dashboard. The dense heatmap is only a display aid
+        # for the immediate scoring screen and would make every database report unnecessarily large.
+        result.pop("phoneme_heatmap", None)
         return save_flag(phrase, result, note, user)
 
     def save_flag(phrase: dict, result: dict, note: str, user: dict) -> dict:

static/index.html

@@ -4481,7 +4481,15 @@ function drawFlags(){
     if(issues.length) card.append(field); else card.append(mark("meta","בדיווח הזה אין פונמה שסומנה במפורש; עדיין אפשר לסווג ולאשר אותו.","p"));
     const category=document.createElement("select"); for(const [value,label] of Object.entries(FLAG_CATEGORIES)){ const o=document.createElement("option"); o.value=value; o.textContent=label; category.append(o); }
     const note=document.createElement("textarea"); note.maxLength=500; note.rows=2; note.placeholder="הערת בדיקה (לא חובה)";
-    const actions=mark("row",""), confirm=mark("btn primary","אישור הדיווח","button"), reject=mark("btn danger","דחייה","button"), msg=mark("msg","");
+    const actions=mark("row",""), diagnose=mark("btn","לוח אבחון","button"), confirm=mark("btn primary","אישור הדיווח","button"), reject=mark("btn danger","דחייה","button"), msg=mark("msg","");
+    diagnose.disabled=!result.trace;
+    diagnose.title=result.trace?"פתיחת האבחון המלא, כולל טבלת ההגאים":"הדיווח ישן ואינו כולל נתוני trace";
+    diagnose.onclick=()=>{
+      const selected=[...field.querySelectorAll("input:checked")].map(x=>({word:x._issue.word,expected:x._issue.expected,heard:x._issue.heard}));
+      openTrace({...result,phrase:{he:f.he||f.phrase,ar:f.ar||""},_diagnostic:{prompt_he:f.he||f.phrase,answer_ar:f.ar,
+        expected_ipa:f.ipa,model:f.model,version:f.version,report_note:f.note||"",review_category:category.value,
+        reviewer_marked_false_issues:selected}});
+    };
     const save=async status=>{
       confirm.disabled=reject.disabled=true; msg.textContent="שומר…";
       const false_issues=[...field.querySelectorAll("input:checked")].map(x=>({word:x._issue.word,expected:x._issue.expected,heard:x._issue.heard}));
@@ -4491,7 +4499,7 @@ function drawFlags(){
         flagsLoaded=false; await loadFlags(); drawManage();
       }catch(err){ msg.textContent="השמירה נכשלה: "+err.message; confirm.disabled=reject.disabled=false; }
     };
-    confirm.onclick=()=>save("confirmed"); reject.onclick=()=>save("rejected"); actions.append(category,confirm,reject,msg); card.append(note,actions); box.append(card);
+    confirm.onclick=()=>save("confirmed"); reject.onclick=()=>save("rejected"); actions.append(category,diagnose,confirm,reject,msg); card.append(note,actions); box.append(card);
   }
 }
 function drawSystemActivity(box){

static/trace.html

@@ -141,6 +141,9 @@ table.data{border-collapse:collapse;font-size:12px;width:100%}
 table.data th,table.data td{padding:3px 6px;border-bottom:1px solid var(--grid);text-align:right;white-space:nowrap}
 details summary{cursor:pointer;color:var(--ink-2);font-size:12px}
 .scroll{overflow-x:auto}
+.chat-payload{width:100%;min-height:320px;resize:vertical;padding:10px;border:1px solid var(--axis);border-radius:9px;background:var(--page);color:var(--ink);font:12px/1.5 ui-monospace,SFMono-Regular,Menlo,monospace;direction:ltr;text-align:left;white-space:pre}
+.copy-row{display:flex;align-items:center;justify-content:flex-end;gap:10px}
+.copy-status{color:var(--ink-2);font-size:13px}
 </style>
 </head>
 <body>
@@ -188,6 +191,12 @@ details summary{cursor:pointer;color:var(--ink-2);font-size:12px}
   <section class="panel">
     <details><summary>טבלת הגאים (כל המספרים)</summary><div class="scroll"><table class="data" id="table"></table></div></details>
   </section>
+
+  <section class="panel" aria-labelledby="chatPayloadTitle">
+    <div class="head"><h2 id="chatPayloadTitle">מידע להתייעצות על ה־pipeline</h2><span class="sub">אבחון מלא להעתקה לצ׳אט, ללא זהות המשתמש וללא קובץ קול</span></div>
+    <textarea class="chat-payload" id="chatPayload" readonly aria-label="מידע מלא להעתקה לצ׳אט"></textarea>
+    <div class="copy-row"><span class="copy-status" id="copyStatus" role="status"></span><button id="copyPayload">העתקה לצ׳אט</button></div>
+  </section>
 </main>
 <dialog id="excludeDlg" aria-labelledby="excludeTitle">
   <form method="dialog" class="exclude-form">
@@ -221,11 +230,29 @@ let wordAudioUrl=new URLSearchParams(location.search).get("src")==="app"?(localS
 let traceContext={}; try{ traceContext=JSON.parse(localStorage.getItem("ihki.traceContext")||"{}"); }catch{}
 let excludedWords={...(traceContext.excluded_words||{})};
 
+function diagnosticChatText(result){
+  const t=result.trace||{}, context={...(result.phrase||{}),...(result._diagnostic||{})};
+  const trace={...t,
+    tokens:(t.tokens||[]).map(({curves,...token})=>token),
+    frames:t.frames?{count:t.frames.count,ms:t.frames.ms}:null
+  };
+  const publicResult={...result,trace};
+  delete publicResult.email; delete publicResult.audio; delete publicResult.phoneme_heatmap; delete publicResult._diagnostic;
+  return [
+    "Please analyze this Palestinian-Arabic pronunciation-scoring diagnostic.",
+    "Identify the most likely failure point and distinguish between acoustic-model, segmentation/alignment, pronunciation-rule, threshold/calibration, and feedback-generation problems.",
+    "Propose prioritized, concrete pipeline changes and regression tests. Do not infer the speaker's identity.",
+    "",
+    JSON.stringify({context,result:publicResult},null,2)
+  ].join("\n");
+}
+
 /* ---------- loading ---------- */
 function load(result,source){
   if(!result||!result.trace||!result.trace.tokens){ show("בקובץ אין trace. פתחו את הלוח מהמעבדה או מניתוח הפונמות של מורה, או טענו את הדוגמה."); return; }
   if(source!=="מהאפליקציה"){ traceContext={}; excludedWords={}; wordAudioUrl=""; }
   R=result; T=result.trace; PH=(R.words||[]).flatMap(w=>w.phonemes||[]); traceSource=source;
+  $("chatPayload").value=diagnosticChatText(result); $("copyStatus").textContent="";
   $("framePlay").disabled=!wordAudioUrl;
   $("framePlay").title=wordAudioUrl?"נגינת חלון הפריימים עם סמן מסונכרן":"אין אודיו מצורף ל־trace";
   const worst=T.tokens.reduce((b,t)=>t.final<b.final?t:b,T.tokens[0]), asked=+new URLSearchParams(location.search).get("i");
@@ -239,6 +266,11 @@ async function loadMock(){
   catch(e){ show(`הדוגמה לא נטענה (${e.message}). הריצו python scripts/mock_trace.py`); }
 }
 $("mock").onclick=loadMock;
+$("copyPayload").onclick=async()=>{
+  const field=$("chatPayload"), status=$("copyStatus");
+  try{ await navigator.clipboard.writeText(field.value); status.textContent="הועתק."; }
+  catch{ field.focus(); field.select(); status.textContent="הטקסט סומן; העתיקו עם Ctrl/Cmd+C."; }
+};
 $("file").onchange=async e=>{ const f=e.target.files[0]; if(!f) return; try{ load(JSON.parse(await f.text()),f.name); }catch(err){ show(`לא JSON תקין: ${err.message}`); } };
 addEventListener("keydown",e=>{ if(!T||e.target.tagName==="INPUT") return; if(e.key==="ArrowLeft"){ step(1); e.preventDefault(); } if(e.key==="ArrowRight"){ step(-1); e.preventDefault(); } });
 $("next").onclick=()=>step(1); $("prev").onclick=()=>step(-1);

tests/test_scoring.py

@@ -1781,7 +1781,8 @@ def test_a_tutor_flags_a_recording_scored_wrong(monkeypatch):
         assert r.status_code == 200 and r.json() == {"ok": True, "audio": False, "flag": row}, r.text
         assert row["email"] == "t@b.c" and row["cat"] == TOILET["cat"] and row["he"] == TOILET["he"] and row["note"] == "אמרתי ח נכונה"
         assert row["ipa"] == "w eː n | i l ħ a m m aː m" and "χ" in row["heard_ipa"] and row["score"] == row["result"]["score"] < 100
-        assert row["result"]["tips"][0]["key"] == "ħ>x" and "phoneme_heatmap" not in row["result"]
+        assert row["result"]["tips"][0]["key"] == "ħ>x" and row["result"]["trace"]["tokens"]
+        assert "phoneme_heatmap" not in row["result"]
         assert send(c, "t", phrase_id=10**6).status_code == 404
         a.teachers = frozenset({"s@b.c"})   # a teacher flags too
         assert c.get("/api/me", headers={"Authorization": "Bearer s"}).json()["tutor"] is True and send(c, "s").status_code == 200