Import batch-07 and batch-04 as model-drafted labels

README.md

@@ -156,7 +156,7 @@ item can be highlighted in the original advisory.
 | 1 · Feasibility | Pinned ingestion, corpus statistics, label schema, offset-preserving preprocessing, pilot sample, annotation tooling | ✅ built |
 | 1 · Pilot | 100-document pilot (`pilot-2`: 22 hand-labelled, 78 from model drafts), [decision gate](docs/pilot_gate.md), guidelines v0.2 frozen, [acceptance targets](docs/targets.md) declared | ✅ done |
 | 2 · Baselines | Rules baseline, CRF + logistic-regression attributes and groups, evaluator with bootstrap CIs, evidence viewer | ✅ built · preliminary pilot dev scores |
-| 3 · Neural models | Frozen leakage-safe splits, annotation batches (01–03, 06: model-drafted), DistilBERT and DeBERTa-v3 with span / role / group heads, experiment runner, CRF ablations and learning curve | 🟨 trained on model drafts (run 3) · DeBERTa ≈ CRF on held-out agreement, no clear winner |
+| 3 · Neural models | Frozen leakage-safe splits, annotation batches (01–04, 06, 07: model-drafted), DistilBERT and DeBERTa-v3 with span / role / group heads, experiment runner, CRF ablations and learning curve | 🟨 trained on model drafts (run 3) · DeBERTa ≈ CRF on held-out agreement, no clear winner |
 | 4 · Reliability | Calibration and review policy, formatting robustness (rules: 100%), challenge-set and error-catalogue tooling | 🟨 tooling ready · thresholds await labels |
 | 5 · Application | FastAPI API, review UI, SQLite (WAL) with append-only history, exports, Docker Compose stack, release manifests, tested rollback, CI image smoke test | ✅ built |
 | 6 · Production | Benchmark harness, ONNX/INT8 export (FP32 parity exact), local proxy benchmarks and load tests, monitoring; CRF v0.1.0 measured and deployed (`app verify` passed) on the 2 vCPU / 4 GB target host | 🟨 CRF meets every serving target on the target host · transformers not yet measured there |
@@ -174,7 +174,7 @@ split has not been touched.
 | CRF (`crf-pilot-2`, pilot train split) | 0.231 [0.062, 0.417] | 0.231 [0.062, 0.417] | 0.444 [0.000, 0.667] |
 | CRF (`crf-pilot-2-batch-01`, + 100 model-drafted `batch-01` records) | 0.444 [0.133, 0.667] | 0.444 [0.133, 0.667] | 0.300 [0.000, 0.750] |
 
-**Label source.** Every annotation batch after the pilot (472 advisories annotated, 22 by hand)
+**Label source.** Every annotation batch after the pilot (710 advisories annotated, 22 by hand)
 was submitted from model drafts, by the owner's decision. Accuracy is therefore measured only on
 the 10 hand-labelled dev documents above. Later results are reported as *agreement with the
 drafting model*, never as accuracy, and the release gates in [`docs/targets.md`](docs/targets.md)

annotations/README.md

@@ -27,6 +27,8 @@ advisory-database commit it was taken from, and the SHA-256 of the annotated tex
 | `batches/batch-02.jsonl` | Batch-02 annotations (train pool, enriched), 100 records, guidelines v0.2. |
 | `batches/batch-03.jsonl` | Batch-03 annotations (dev pool, natural), 112 records, guidelines v0.2. |
 | `batches/batch-06.jsonl` | Batch-06 annotations (dev pool, natural), 60 records, guidelines v0.2. |
+| `batches/batch-07.jsonl` | Batch-07 annotations (dev pool, natural), 100 records, guidelines v0.2. |
+| `batches/batch-04.jsonl` | Batch-04 annotations (test pool, natural), 138 records, guidelines v0.2. |
 | `batches/batch-*_report.md` | Batch import reports. |
 
 ## Origin
@@ -71,5 +73,12 @@ hand-labelled dev documents. `evaluate --gold-origin manual` excludes it, and `c
 recorded. On 2026-11-05 the owner decided to continue with model-drafted labels, so results
 against these batches are reported as agreement with the drafting model, not accuracy.
 
+`batch-07` (dev pool, 100 records) and `batch-04` (test pool, 138 records), guidelines v0.2,
+were also drawn for hand labelling and were submitted from model drafts on 2026-11-09, by the
+owner's decision: every record is `model_draft_accepted` (`model-batch-07-guidelines-0.2`,
+`model-batch-04-guidelines-0.2`), with no annotation time recorded. The drafting process read
+the batch-04 test texts; no model was developed, tuned or scored on them. CI checks batch-04
+records for provenance only.
+
 Records follow the schema in `src/triggermap/schema/models.py` and the rules in
 `docs/annotation_guidelines.md`.

annotations/batches/batch-04_report.md

@@ -0,0 +1,49 @@
+# Annotation import report
+
+Generated by `triggermap annotate import`; do not edit by hand.
+
+- Tasks in export: 138
+- Records written (no issues): 138
+- Tasks with issues: 0
+
+## Label counts (clean records)
+
+| Document status | Count |
+|---|---:|
+| HAS_CONDITIONS | 132 |
+| NO_EXPLICIT_CONDITION | 6 |
+
+| Category | Count |
+|---|---:|
+| ACCESS | 89 |
+| CONFIGURATION | 55 |
+| PLATFORM | 39 |
+| USAGE | 157 |
+
+| Role | Count |
+|---|---:|
+| EXCEPTION | 11 |
+| NOT_REQUIRED | 15 |
+| REQUIRED | 284 |
+| UNCERTAIN | 30 |
+
+| Group operator | Count |
+|---|---:|
+| ALL | 13 |
+| ANY | 1 |
+
+Conditions per condition-bearing document: 2.0
+
+## Origin (clean records)
+
+| Origin | Count |
+|---|---:|
+| model_draft_accepted | 138 |
+
+- Draft model `model-batch-04-guidelines-0.2`: 138 records
+
+Records with origin `model_draft_accepted` carry a model's labels unchanged, and `model_draft_revised` records were changed with no annotation time recorded (not evidence of human editing). Scores against either measure agreement with the drafting process, not accuracy; report them separately from `manual` records.
+
+## Pace
+
+No lead times recorded.