diff --git a/main.js b/main.js
index 286bd7e..4796004 100644
--- a/main.js
+++ b/main.js
@@ -1,5 +1,5 @@
-process.env.UV_THREADPOOL_SIZE = process.env.UV_THREADPOOL_SIZE || '64';
-const { monitorEventLoopDelay } = require('perf_hooks');
+process.env.UV_THREADPOOL_SIZE = process.env.UV_THREADPOOL_SIZE || '8';
+const { monitorEventLoopDelay, PerformanceObserver } = require('perf_hooks');
 const { app, BrowserWindow, ipcMain, dialog, clipboard, nativeTheme, Tray, Menu, nativeImage } = require('electron');
 nativeTheme.themeSource = 'dark';
 const path = require('path');
@@ -32,6 +32,19 @@ _eventLoopDelay.enable();
 let _eldLastLog = 0;
 let _lastCpu = process.cpuUsage();
 let _lastCpuT = Date.now();
+let _gcCount = 0;
+let _gcTotalMs = 0;
+let _gcMaxMs = 0;
+try {
+  const _gcObserver = new PerformanceObserver((list) => {
+    for (const entry of list.getEntries()) {
+      _gcCount++;
+      _gcTotalMs += entry.duration;
+      if (entry.duration > _gcMaxMs) _gcMaxMs = entry.duration;
+    }
+  });
+  _gcObserver.observe({ entryTypes: ['gc'] });
+} catch {}
 
 let mainWindow;
 let _lastImportPath = null;
@@ -212,11 +225,20 @@ function _maybeLogEventLoopDelay(activeJobs) {
       const d = process.cpuUsage(_lastCpu);
       const wall = now - _lastCpuT;
       const pct = wall > 0 ? Math.round((d.user + d.system) / 1000 / wall * 100) : 0;
-      const rss = Math.round(process.memoryUsage().rss / 1048576);
-      cpuStr = ` cpu=${pct}%core rss=${rss}MB`;
+      const mem = process.memoryUsage();
+      const rss = Math.round(mem.rss / 1048576);
+      const heap = Math.round(mem.heapUsed / 1048576);
+      const ext = Math.round(mem.external / 1048576);
+      const ab = Math.round((mem.arrayBuffers || 0) / 1048576);
+      cpuStr = ` cpu=${pct}%core rss=${rss}MB heap=${heap}MB ext=${ext}MB ab=${ab}MB`;
       _lastCpu = process.cpuUsage();
       _lastCpuT = now;
     } catch {}
+    let gcStr = '';
+    try {
+      gcStr = ` gc=${_gcCount} gcTotal=${_gcTotalMs.toFixed(0)}ms gcMax=${_gcMaxMs.toFixed(0)}ms`;
+      _gcCount = 0; _gcTotalMs = 0; _gcMaxMs = 0;
+    } catch {}
     let upStr = '';
     try {
       if (uploadManager && typeof uploadManager.getDiagnostics === 'function') {
@@ -225,7 +247,7 @@ function _maybeLogEventLoopDelay(activeJobs) {
         upStr = ` active-by-hoster={${byHoster}} transient-errs=${d.transientErrors || 0} pending=${d.pending || 0}`;
       }
     } catch {}
-    logInfo(`eventloop-delay active=${activeJobs} mean=${mean}ms p99=${p99}ms max=${max}ms stddev=${stddev}ms threadpool=${process.env.UV_THREADPOOL_SIZE}${cpuStr}${resStr}${upStr}`);
+    logInfo(`eventloop-delay active=${activeJobs} mean=${mean}ms p99=${p99}ms max=${max}ms stddev=${stddev}ms threadpool=${process.env.UV_THREADPOOL_SIZE}${cpuStr}${gcStr}${resStr}${upStr}`);
     _eventLoopDelay.reset();
   } catch {}
 }
diff --git a/package.json b/package.json
index 7d0e75d..c092181 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "multi-hoster-uploader",
-  "version": "3.3.96",
+  "version": "3.3.97",
   "description": "Upload files to doodstream, voe, vidmoly, byse simultaneously",
   "main": "main.js",
   "scripts": {
diff --git a/tasks/lessons.md b/tasks/lessons.md
index 5a1125f..bdb77ed 100644
--- a/tasks/lessons.md
+++ b/tasks/lessons.md
@@ -1,5 +1,20 @@
 # Lessons
 
+## 2026-06-21 — Histogram-Korrelation beweist KEINE Kausalrichtung; rss-Mathe als Sanity-Check
+**Symptom:** ELD-Spikes korrelierten exakt mit hohem `FSReqCallback` (File-Reads in flight) → ich wollte
+sofort ein Read-Concurrency-Semaphore über 5 Dateien bauen.
+**Root cause / Korrektur (Advisor):** (1) Korrelation ≠ Kausalität: hohe in-flight-Reads können auch SYMPTOM
+sein — ein aus ANDEREM Grund blockierter Loop drained die Read-Completions nicht, also stapeln sie sich im
+Snapshot. (2) rss-Mathe widerlegte meine „Read-Buffer ballonen den Speicher"-These: 70 Streams × 256KB ≈
+18MB, aber rss schwang ~300MB → das ist Heap-/Objekt-Churn (GC), nicht die Read-Buffer.
+**Regel:** Bevor ich auf Basis einer Histogramm-Korrelation einen Multi-File-Refactor baue: (a) Kausalrichtung
+mit einem BILLIGEN reversiblen 1-Zeilen-Hebel testen (hier UV_THREADPOOL_SIZE 64→8), (b) die Größenordnung
+gegenrechnen (passt die vermutete Quelle zahlenmäßig zur beobachteten Wirkung?), (c) den fehlenden Co-Faktor
+(GC) erst MESSEN, bevor ich ihn aus- oder einschließe. Ein Build kann gleichzeitig Kandidaten-Fix UND
+Diskriminator sein.
+**Wie anwenden:** Bei „X korreliert mit Y, also fixe X": erst fragen „könnte Y → X statt X → Y?" und „passt
+die Magnitude?". Wenn nein/unklar → erst der billige reversible Knopf + Messung, dann der teure Refactor.
+
 ## 2026-04-21 — DOM-Doppelrender bei Bulk-State-Changes
 **Symptom:** User klickt auf "Erneut versuchen" mit 500+ Jobs → App hängt sekundenlang.
 **Root cause:** `retrySelectedJobs()` ruft `renderQueueTable + updateQueueActionButtons + updateStatusBar` auf, `startSelectedUpload()` ruft direkt danach genau dieselben Funktionen nochmal auf.
diff --git a/tasks/todo.md b/tasks/todo.md
index cbbea97..a492c75 100644
--- a/tasks/todo.md
+++ b/tasks/todo.md
@@ -1,3 +1,32 @@
+# v3.3.97 — DECISIVE ELD finding: file-read phase-flip + threadpool 64→8 + GC instrument
+
+The v3.3.96 `eventloop-delay` logs gave the decisive signal. At CONSTANT active-count, the system flips
+between two regimes:
+- HEALTHY (ELD ~11ms, rss 268–308MB): `SimpleWriteWrap ≈ active`, `FSReqCallback ≈ 0–1` (write/network-bound)
+- BLOCKED (ELD 49–217ms, rss 540–610MB): `FSReqCallback ≈ active` (62–71 file reads in flight), `SimpleWriteWrap ≈ 0–4`
+ELD spike + rss balloon both track `FSReqCallback` → file-read path through the libuv threadpool, NOT crypto,
+NOT renderer, NOT GC-alone. All 5 uploaders read identically (256KB createReadStream); byse/dood/voe run
+through the GENERIC uploadFile in hosters.js (no dedicated module).
+
+Advisor caveats baked into the build (do NOT skip on re-measure):
+1. Causation UNPROVEN — high FSReqCallback could be a SYMPTOM (blocked loop can't drain read-completions).
+2. rss math kills "read buffers ballooned": 70×256KB ≈ 18MB, but rss swings ~300MB → heap/object churn (GC).
+3. Cheapest discriminator already wired: `UV_THREADPOOL_SIZE` 64→8 (1 line, reversible, NOT an upload cap;
+   8×256KB reads ≈ 100MB/s ≫ 41MB/s aggregate). Suspect tp=64 made it WORSE (removed read-serialization).
+
+Shipped v3.3.97 = candidate-fix + discriminator in one build:
+- main.js:1 `UV_THREADPOOL_SIZE` 64→8.
+- ELD line now also logs `heap=`(heapUsed) `ext=`(external) `ab=`(arrayBuffers) `gc=`/`gcTotal=`/`gcMax=`ms
+  (PerformanceObserver entryTypes:['gc'], reset per window).
+
+DECISION RULE for the next user log:
+- ELD drops with tp=8 → read over-parallelism confirmed → keep 8 or productionize a DEDICATED read-semaphore.
+- ELD high + gcTotal/gcMax align with spikes → heap churn → hunt the allocator (semaphore would be wasted).
+- ELD high + gc flat → causation reversed (symptom) → pivot.
+WAIT for the next `eventloop-delay` log before any read-path refactor. NO upload cap (user rejected it).
+
+---
+
 # v3.3.94 — comprehensive measurement build (user: "mach alles messen was man messen kann")
 
 Localization so far (each step EMPIRICAL, not by elimination — advisor caught the elimination-leap):