Replace trajectory audit with isolated prefix scoring

README.md

@@ -72,6 +72,7 @@ README.
 swe_data_processing/
 ├── pyproject.toml
 ├── README.md
+├── changelog.md             # Versioned strategy changes
 ├── .env.example
 ├── src/swe_data_processing/
 │   ├── cli.py                 # Command-line entry points
@@ -203,18 +204,28 @@ The classifier returns one of:
 Training use is tracked separately as `SFT_FULL`, `SFT_STEP_ONLY`,
 `DPO_REJECTED`, `ERROR_ANALYSIS`, `HOLD`, or `DROP`.
 
-### 4. Audit educational process quality
+### 4. Locate a safe prefix and score it independently
 
-This stage does not repair trajectories. It identifies defensible erroneous or
-inefficient calls, scores six process-quality dimensions, and optionally finds
-a causal first-bad assistant turn for prefix-only learning:
+`audit` uses two isolated GLM calls for failed and unknown trajectories:
+
+1. The boundary call receives the trajectory as roughly 32-turn blocks without
+   splitting an assistant action from its immediate tool result. It cannot see
+   `resolved`, model patches, or reference patches. It selects the earliest
+   unrepaired major/critical assistant turn, keeps a safe full process trace,
+   or returns `HOLD` when neither decision is supported.
+2. Python slices the exact messages before that assistant turn. The quality call
+   receives only this materialized prefix, so suffix behavior cannot affect its
+   score.
+
+Successful trajectories skip boundary selection and are scored as complete
+trajectories. The command remains simple:
 
 ```bash
 swe-qc audit \
   --input samples/sample_20_seed_20260805.jsonl \
   --output qc_outputs/sample20.audits.jsonl \
   --errors qc_outputs/sample20.audit.errors.jsonl \
-  --workers 20 \
+  --workers 5 \
   --resume
 ```
 
@@ -223,15 +234,23 @@ remain serialized in the main thread, so each completed record is appended
 atomically even when API requests run in parallel. Output order follows request
 completion order; `sample_id` remains the stable join key.
 
-The deterministic score combines weighted process dimensions with penalties
-for minor, major, and critical behavior issues. Failed exploratory calls are
-not penalized when the agent interprets them correctly and recovers.
+The gateway previously returned frequent HTTP 429 responses at 20 workers, so
+five workers is the practical default for long runs unless the service limit is
+raised.
 
-The causal audit payload excludes the reference patch and every derived signal,
-including reference file names, patch length, and model/reference size ratios.
-The immutable outcome is used only to select the mode: failed or unknown
-trajectories can become process-prefix candidates, but never full-trajectory
-SFT candidates.
+The quality call lists concrete erroneous and inefficient assistant actions and
+scores five 0-20 dimensions: planning, investigation, tool use and observation,
+progress, and clarity/efficiency. Python recomputes issue counts, sums the five
+dimensions, and assigns:
+
+- `HIGH`: 80-100
+- `MEDIUM`: 60-79
+- `LOW`: 40-59
+- `REJECT`: below 40, or an unrepaired major/critical problem remains
+
+Incomplete prefixes can still be useful. Normal failed experiments are retained
+when the agent later diagnoses, repairs, and verifies them. A prefix ending in a
+complete tool result is valid process-SFT structure.
 
 ### 5. Create static repair plans