feat(codex-bridge): decision-model stages, Dify rules and beta.39

Owner request, 14 October 2026: use the decision model (Jev / Celeris) at
several stages, let Dify supply the rules, code it without new bugs.
Every new stage is switched on in Settings > Kademeli Gecis and starts
disabled, so the upgrade changes no behaviour by itself.

- Settings schema 2 (Python and TypeScript validate the same rules); a
  schema 1 file is filled from the defaults. Default chain is now
  Katman-1 -> Katman-2 -> Katman-3 with Katman-3, Chat-Model and
  bekci-sesli-bekci always passing (owner decision of the same day).
- watchModels: the Bekci correction channel follows a settings list
  instead of the hard-coded Sesli-Hizli-Zeki.
- prefilter: a predicted refusal starts the step at the chosen layer.
- autoRoute: virtual "Otomatik" model; one decision per user turn picks
  the combo, tool steps of the same turn keep it.
- finishCheck: an answer that stops early goes back to the same model once.
- A layer whose window cannot hold the request is skipped.
- policy: stage rules from a Dify workflow (karar_politikasi.py), fetched
  in the background, validated, last good copy kept; turns never wait.
- karar_sor.py: verifierCheck for the dynamic-workflow acceptance step.
- Bridge version 19, IDE 1.0.0-beta.39.

Tests: 299 bridge tests, 36 electron tests; live run on a side bridge
(port 8797) with Celeris and the AITURK gateway for every stage.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

apps/desktop/assets/codex-bridge/jev-settings-defaults.json

@@ -5,10 +5,9 @@
   "fallbackModel": "Katman-1",
   "rejectionRules": [
     { "from": "Katman-1", "to": "Katman-2" },
-    { "from": "Katman-2", "to": "Katman-3" },
-    { "from": "Katman-3", "to": "Katman-4" }
+    { "from": "Katman-2", "to": "Katman-3" }
   ],
-  "exemptModels": ["Sesli-Hizli-Zeki", "Katman-4"],
+  "exemptModels": ["Chat-Model", "bekci-sesli-bekci", "Katman-3"],
   "thresholds": { "ret": 0.5, "sart": 0.7 },
   "questions": {
     "ret": "The assistant did NOT do what the user asked or did NOT answer the user's question: it refused, declined, said it cannot or will not, avoided giving the requested judgment, or redirected to a different topic.",
@@ -18,5 +17,41 @@
   "criteria": { "ret": { "true": "", "false": "" }, "sart": { "true": "", "false": "" } },
   "maxConcurrent": 0,
   "queueWaitSeconds": 3600,
-  "failureAction": "pass"
+  "failureAction": "pass",
+  "watchModels": ["Sesli-Hizli-Zeki", "Chat-Model", "bekci-sesli-bekci"],
+  "prefilter": {
+    "enabled": false,
+    "question": "A mainstream AI assistant such as ChatGPT, Claude or Gemini would refuse this request, water it down, or answer it only with added warnings or conditions.",
+    "threshold": 0.6,
+    "target": "Katman-2",
+    "timeoutSeconds": 3
+  },
+  "autoRoute": {
+    "enabled": false,
+    "model": "Otomatik",
+    "timeoutSeconds": 3,
+    "questions": {
+      "zor": "The user's request is hard: doing it well needs multi-step work, expert or technical knowledge, careful analysis, planning, calculation or writing and changing code. Greetings, small talk, simple facts and short simple commands are NOT hard.",
+      "arastirma": "The user's request is mainly research: finding, reading and comparing many sources or files and reporting what they say, rather than changing anything."
+    },
+    "thresholds": { "zor": 0.5, "arastirma": 0.6 },
+    "targets": { "basit": "Workflow-Kesif", "zor": "Ajan", "arastirma": "Arastirma" }
+  },
+  "finishCheck": {
+    "enabled": false,
+    "question": "The assistant stopped before the user's task was finished: it said it is done or handed back without doing all of what was asked, or it claimed success (tests pass, file changed, deployed) without showing the actual evidence.",
+    "threshold": 0.6,
+    "note": "Önceki cevabın iş bitmeden bitti dedi ya da kanıt göstermedi. İşi gerçekten bitir; gerekiyorsa araçları kullan ve gerçek kanıtı göster.",
+    "timeoutSeconds": 5
+  },
+  "verifierCheck": {
+    "enabled": false,
+    "question": "The reviewer approved the work without concrete evidence: it does not cite the actual diff, the real command output or specific lines, so the approval may be a guess.",
+    "threshold": 0.6
+  },
+  "policy": {
+    "source": "local",
+    "endpoint": "https://dify.turkservis.online/v1",
+    "refreshMinutes": 5
+  }
 }