Generator: pilot analysis and fixes; G0002-G0021 tasks; handover notes merged

- Static bundle checks (name length, seed type, reserved words, contract test classes,
  testclasses_file), local abaplint parser check before SAP, max_tokens, robust JSON parse
- Runner: G2 for CDS views with parameters, contract_check detail in report
- G0020 and G0013 fixed by review and revalidated (oracle 100, null 0)
- Pilot results in docs/faz1-tasarim.md 11e; handover notes merged into CLAUDE.md and docs

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

CLAUDE.md

@@ -1,8 +1,15 @@
 # ABAP LLM harness — instructions for Claude Code
 
-Talk to the user (Kral) in Turkish. Keep answers short; remove words that add no value.
-When you write English text (specs, prompts, docs for the model), use ASD-STE100 Simplified Technical English.
-Ask before an action that uses much cloud budget.
+## 0. Working with Kral
+
+- Talk in Turkish. Short answers. Remove words and sentences that add no value.
+- English text (specs, prompts, docs for the model): ASD-STE100 Simplified Technical English.
+- Ask before an action that costs much (cloud budget) or that changes the A4H system outside harness runs.
+- For a series of updates to one document, deliver the result directly as a Markdown file; do not ask each time.
+- Record each decision with its date in `docs/yol-haritasi.md` or `docs/faz1-tasarim.md`.
+- Commit the current changes before a new generation or test batch.
+- Kral has no time for a full review. Eval review = automatic gates + empirical filter (2–3 models) +
+  Claude review with a checklist; Kral reviews only flagged tasks + one per category (~10 tasks).
 
 ## 1. Project
 
@@ -12,6 +19,12 @@ Ask before an action that uses much cloud budget.
 - Contract of the model: "ABAP task as text + generic ABAP MCP interface".
   The plan format (plan-agent.md) gives the best result, but it is not mandatory.
 - Training data: never Claude output. Eval tasks never go into training data.
+- Input: the plan format gives the best result but is not mandatory. The old playbook format is not used.
+- Generic ABAP MCP interface (core: search_object, read_source, object_structure, where_used, create_object,
+  write_source, activate, syntax_check, run_unit_tests, run_atc). Small tool name/schema variation in training.
+- Category K: incomplete or free-text input + other tool schema.
+- Object type mix: CLAS/INTF 35 %, FUNC 15 %, PROG 10 %, DDIC 10 %, CDS 25 %, MSAG + exception 5 %.
+- Teacher and task author: DeepSeek V4.1 Flash (Ollama cloud). Budget calendar in `docs/yol-haritasi.md`.
 - Read first: `docs/yol-haritasi.md` (roadmap, current position) and `docs/faz1-tasarim.md` (design).
   When a decision or a result changes, update these files.
 
@@ -76,21 +89,27 @@ python3 -c "from harness.ledger import spent; print(spent())"
 - Done: harness on Mac mini; test include creation in the server; teardown; FUNC/PROG/DDLS support;
   task generator; ledger and budget guard.
 - Results: T01 oracle 100, Qwen 27B 41.7 (old run 103), DeepSeek 98.5. T13 DeepSeek 85, T14 DeepSeek 100.
-- Running: pilot (`harness.pilot`, 20 tasks) → `runs/gen/pilot.json`, log `runs/gen/pilot.log`.
-  A macOS notification shows when it ends.
-- First generated task G0001: not accepted after 3 attempts (activation error, seed save error,
-  reference failed 1 hidden test). Feedback now includes the reference write/activation errors.
+- Pilot (G0002–G0021) done: 16/20 accepted, 0.16 USD per accepted task (`runs/gen/pilot.json`,
+  analysis in `docs/faz1-tasarim.md` 11e). Generator improved after the pilot: static checks
+  (name length, seed type, reserved words, contract/test classes), local abaplint parser check before SAP,
+  max_tokens 100k, G2 detail in the repair feedback. Runner: G2 works for CDS views with parameters.
+  The improved generator is not yet tested with new tasks.
+- G0020 (rejected) passes with oracle 100 after one fix (remove parameter `default`). G0013 (accepted) has a test
+  class in its contract: fix in review.
 
 ## 7. Next steps
 
-1. Analyze the pilot: acceptance rate, failure causes, cost per task. Improve the generator prompt.
+1. Test the improved generator with a small batch (~5 tasks, ~0.7 USD); compare repairs per task with the pilot (2.75 calls).
 2. Step D: mutation check (hidden tests must fail on a broken reference).
 3. Step F: eval set of 110 tasks: ~150 candidates, empirical filter (2–3 models), Claude review with a
    checklist, Kral spot-checks ~10 flagged tasks.
 4. Scoring for stop tasks (category H) and category K (free-text input, other tool schema).
 5. Step 1.3b: generic ABAP MCP interface (spec `abap-mcp-arayuz.md`, later). The proxy is the first adapter.
 
-## 8. Open items for Kral (server)
+## 8. Open items for Kral (EPOD server)
 
+- When a write fails, run a syntax check and return its messages. Now the server says only "save failed"
+  (for example for `TYPE c LENGTH n` in a method signature). The model cannot see the cause and cannot repair;
+  this hurts eval and training. Workaround in the generator: local abaplint parser check.
 - Concurrency: queue calls per RFC connection, or use a connection pool.
 - BDEF creation (needed for RAP tasks).