feat(ui): false-positive penalty (severity-scaled, default -0.5)

Adds an editable 'false-positive penalty ×' to the dashboard. A false positive carries no graded severity, so it's penalized by the severity the model CLAIMED (its lens verdict / raw_severity, mapped onto the curve: Blocking->high, Minor->small). points(net) = confirmed points + Σ penalty×points[claimed], so a model with a few good finds but many false positives nets down — even negative — and sorts to the bottom. Adds an 'fp pen' column; net points/pts-min/pts-run shown red when negative. Client-side only; the store stays point-free.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

README.md

@@ -131,6 +131,12 @@ points curve (default `trivial=1, small=3, medium=5, high=8, critical=20`) and c
 `points = Σ weight[severity]·count` and `value/min = points / minutes` on the fly — retune it without
 touching stored data.
 
+There's also an editable **false-positive penalty ×** (default `-0.5`). A false positive has no
+graded severity, so it's penalized by the severity the model **claimed** (its lens verdict —
+Blocking→high, Minor→small): `penalty × points[claimed]`. So a Blocking-claimed FP at `-0.5` costs
+`high(8) × -0.5 = -4`, and a model with the odd good find but many false positives nets *down* —
+even negative — instead of coasting on its hits.
+
 Auth: the `/ui` shell is public (it holds no data); paste the store token into its **connect** box,
 or open `/ui?token=<token>` once (remembered in `localStorage`). Prefer your own dashboard? Point
 Grafana/Metabase/etc. at the SQLite file or the same `/export` + `/scoreboard` + `/runs` JSON.

ui.html

@@ -79,6 +79,7 @@
         <span class="small mut">medium</span><input type="number" id="p_medium" value="5">
         <span class="small mut">high</span><input type="number" id="p_high" value="8">
         <span class="small mut">critical</span><input type="number" id="p_critical" value="20">
+        <span class="small mut" style="margin-left:18px">false-positive penalty ×</span><input type="number" id="fp_mult" value="-0.5" step="0.5" title="A false positive scores this × the severity the model CLAIMED (its lens verdict). e.g. a Blocking-claimed FP at -0.5 = high(8) × -0.5 = -4 pts.">
       </div>
     </div>
   </div>
@@ -164,6 +165,17 @@ function curve(){
   for (const s of SEVS) c[s] = parseFloat(document.getElementById("p_"+s).value) || 0;
   return c;
 }
+function fpMult(){ const v = parseFloat(document.getElementById("fp_mult").value); return isNaN(v) ? 0 : v; }
+// A false positive has no graded severity, so penalize it by the severity the
+// MODEL claimed — its lens verdict (raw_severity) — mapped onto the curve. The
+// louder the wrong cry, the bigger the penalty.
+function rawToSevKey(raw){
+  const s = (raw||"").toLowerCase();
+  if (s.includes("blocking")) return "high";
+  if (s.includes("minor")) return "small";
+  if (s.includes("no material")) return "trivial";
+  return "medium"; // unknown / "Reviewed"
+}
 function filters(){
   return {
     from: document.getElementById("from").value,
@@ -204,7 +216,7 @@ function aggregate(f){
   const c = curve();
   const M = new Map();
   const get = m => { if(!M.has(m)) M.set(m, {model:m, provider:"", runs:0, minutes:0, inTok:0, outTok:0,
-    findings:new Set(), confirmed:new Set(), fp:new Set(), ungraded:new Set(), sev:Object.fromEntries(SEVS.map(s=>[s,new Set()]))}); return M.get(m); };
+    findings:new Set(), confirmed:new Set(), fp:new Map(), ungraded:new Set(), sev:Object.fromEntries(SEVS.map(s=>[s,new Set()]))}); return M.get(m); };
 
   for (const r of RUNS){ if(!runMatch(r,f)) continue; const m=get(r.model); m.runs++; m.minutes += (r.duration_secs||0)/60;
     m.inTok += r.input_tokens||0; m.outTok += r.output_tokens||0; if(r.provider) m.provider=r.provider; }
@@ -213,17 +225,20 @@ function aggregate(f){
   for (const r of rows){ const m=get(r.model); if(r.provider) m.provider=m.provider||r.provider;
     m.findings.add(r.finding_id);
     if (r.graded && r.is_real === true){ m.confirmed.add(r.finding_id); if (r.severity) m.sev[r.severity].add(r.finding_id); }
-    else if (r.graded && r.is_real === false){ m.fp.add(r.finding_id); }
+    else if (r.graded && r.is_real === false){ m.fp.set(r.finding_id, rawToSevKey(r.raw_severity)); }
     else { m.ungraded.add(r.finding_id); }
   }
 
+  const fpm = fpMult();
   const out = [...M.values()].map(m => {
     const sevCounts = Object.fromEntries(SEVS.map(s=>[s, m.sev[s].size]));
-    const points = SEVS.reduce((a,s)=> a + c[s]*sevCounts[s], 0);
+    const confirmedPoints = SEVS.reduce((a,s)=> a + c[s]*sevCounts[s], 0);
+    let fpPen = 0; for (const k of m.fp.values()) fpPen += (c[k]||0) * fpm; // negative when fpm<0
+    const points = confirmedPoints + fpPen; // NET of the false-positive penalty
     const findings = m.findings.size, confirmed = m.confirmed.size;
     return { model:m.model, provider:m.provider, runs:m.runs, minutes:m.minutes,
       inTok:m.inTok, outTok:m.outTok, findings, confirmed, fp:m.fp.size, ungraded:m.ungraded.size,
-      sev:sevCounts, points,
+      sev:sevCounts, confirmedPoints, fpPen, points,
       ptsPerMin: m.minutes>0 ? points/m.minutes : null,
       ptsPerRun: m.runs>0 ? points/m.runs : null,
       confirmedPct: findings>0 ? confirmed/findings*100 : null };
@@ -236,7 +251,8 @@ const COLS = [
   {k:"runs", t:"runs"}, {k:"minutes", t:"min", fmt:v=>v.toFixed(1)},
   {k:"findings", t:"findings"}, {k:"confirmed", t:"real"}, {k:"fp", t:"FP"}, {k:"ungraded", t:"ungr"},
   {k:"confirmedPct", t:"real%", fmt:v=>v==null?"—":v.toFixed(0)+"%"},
-  {k:"points", t:"points", fmt:v=>v.toFixed(0)},
+  {k:"fpPen", t:"fp pen", fmt:v=>v?v.toFixed(1):"0"},
+  {k:"points", t:"points (net)", fmt:v=>v.toFixed(0)},
   {k:"ptsPerMin", t:"pts/min", fmt:v=>v==null?"—":v.toFixed(2)},
   {k:"ptsPerRun", t:"pts/run", fmt:v=>v==null?"—":v.toFixed(1)},
   {k:"sev", t:"by severity", l:true, fmt:sev=>SEVS.filter(s=>sev[s]).map(s=>`<span class="sev" style="background:${SEVCOLOR[s]}">${s[0].toUpperCase()}${sev[s]}</span>`).join(" ")||"—"},
@@ -268,8 +284,9 @@ function render(){
       const td = document.createElement("td"); if (col.l) td.className="l";
       const v = m[col.k];
       td.innerHTML = col.fmt ? col.fmt(v) : (v==null?"—":v);
-      if (col.k==="ptsPerMin" && v!=null) td.classList.add("good");
-      if (col.k==="fp" && v>0) td.classList.add("bad");
+      if ((col.k==="ptsPerMin" || col.k==="ptsPerRun" || col.k==="points") && v!=null) td.classList.add(v<0 ? "bad" : "good");
+      if (col.k==="fpPen" && v<0) td.classList.add("bad");
+      if (col.k==="fp" && v>0) td.classList.add("warn");
       tr.appendChild(td);
     }
     mb.appendChild(tr);