professor3333 · Nov 5, 2026 · Nov 5, 2026
diff --git a/README.md b/README.md
@@ -180,6 +180,17 @@ the 10 hand-labelled dev documents above. Later results are reported as *agreeme
 drafting model*, never as accuracy, and the release gates in [`docs/targets.md`](docs/targets.md)
 cannot be judged as declared.
 
+| Agreement with model drafts (172 dev records, not accuracy) | Typed-condition micro-F1 | Exact span F1 | False-prerequisite rate |
+|---|---:|---:|---:|
+| Rules (`rules-2`) | 0.107 [0.058, 0.154] | 0.142 [0.085, 0.195] | 0.330 [0.165, 0.467] |
+| CRF (`crf-pilot-2-batch-01`) | 0.128 [0.095, 0.168] | 0.174 [0.134, 0.215] | 0.416 [0.294, 0.493] |
+| CRF (`crf-batches-01-02`, + 100 model-drafted `batch-02` records) | 0.145 [0.105, 0.185] | 0.221 [0.171, 0.268] | 0.353 [0.260, 0.416] |
+
+Adding `batch-02` clearly improves exact-span agreement (+0.047, 95% CI [+0.004, +0.086]), but not
+typed agreement. On the 10 hand-labelled documents, `crf-batches-01-02` scores 0.370 typed F1
+[0.000, 0.640], and the difference from `crf-pilot-2-batch-01` is not clear; see
+[`reports/agreement/`](reports/agreement/README.md).
+
 95% project-group bootstrap intervals (1,000 resamples). Full reports, including scores against
 the 28 model-made dev records (which measure agreement with the drafting model, not accuracy),
 are in [`reports/pilot/`](reports/pilot/README.md). The paired difference from adding `batch-01` (+0.214 typed F1,
@@ -226,6 +237,8 @@ uv run triggermap predict --model crf --model-dir artifacts/crf-1 --out data/pre
 uv run triggermap evaluate annotations/pilot/pilot-1.jsonl data/predictions/crf-1.jsonl --out reports/crf-1.md
 uv run triggermap compare annotations/pilot/pilot-2.jsonl reports/pilot/predictions/rules-2.jsonl \
     reports/pilot/predictions/crf-pilot-2.jsonl --gold-origin manual   # paired bootstrap, B - A
+uv run triggermap evaluate annotations/batches/batch-03.jsonl reports/agreement/predictions/crf-batches-01-02.jsonl \
+    --add-gold annotations/batches/batch-06.jsonl --split dev   # agreement with model drafts, not accuracy
 ```
 
 **Transformers** (once gold labels exist; DistilBERT, DeBERTa-v3-small; GPU recommended, see

diff --git a/docs/model_card.md b/docs/model_card.md
@@ -65,7 +65,9 @@ the review-policy version.
 Training uses the train pool of `splits/pool-v1.json` only (see
 [`dataset_card.md`](dataset_card.md)). The models trained so far used `pilot-2` (39 in the
 train split) and `batch-01` (100). Of these 139 records, 7 are `manual`. The other 132 are
-model drafts. `batch-02` (100 more model-drafted train records) is available and not yet used.
+model drafts. `crf-batches-01-02` adds `batch-02` (239 records: 7 `manual`, 232 model drafts).
+Its dev results, as agreement with the model drafts and as accuracy on the 10 hand-labelled
+documents, are in [`reports/agreement/`](../reports/agreement/README.md).
 
 ## Preliminary dev results
 

diff --git a/reports/agreement/README.md b/reports/agreement/README.md
@@ -0,0 +1,84 @@
+# Agreement with model drafts (dev split only)
+
+Labels after the pilot are model drafts (owner's decision, 2026-11-05; see
+[`docs/targets.md`](../../docs/targets.md)). This report scores three systems in two ways:
+
+- **Agreement:** against the 172 model-drafted dev records of `batch-03` (112) and `batch-06`
+  (60), all `model_draft_accepted`, guidelines v0.2. These numbers measure how closely a system
+  reproduces the drafting model's labels. **They are not accuracy.**
+- **Accuracy:** against the 10 hand-labelled dev documents of `pilot-2` (`--gold-origin manual`,
+  17 gold conditions), the only accuracy slice that exists.
+
+The test split was not touched. All intervals are 95% project-group bootstrap intervals, 1,000
+resamples, seed 0.
+
+| System | Training records (train split) | Origins |
+|---|---|---|
+| `rules-2` | none (deterministic rules) | n/a |
+| `crf-pilot-2-batch-01` | `pilot-2` + `batch-01`: 139 | 7 `manual`, 132 model drafts |
+| `crf-batches-01-02` | `pilot-2` + `batch-01` + `batch-02`: 239 | 7 `manual`, 219 `model_draft_accepted`, 13 `model_draft_revised` |
+
+## Agreement: 172 model-drafted dev documents
+
+155 documents have draft conditions (386 conditions) and 17 have none.
+
+| Metric | `rules-2` | `crf-pilot-2-batch-01` | `crf-batches-01-02` |
+|---|---:|---:|---:|
+| Typed-condition micro-F1 | 0.107 [0.058, 0.154] | 0.128 [0.095, 0.168] | 0.145 [0.105, 0.185] |
+| Exact span F1 | 0.142 [0.085, 0.195] | 0.174 [0.134, 0.215] | 0.221 [0.171, 0.268] |
+| Exact span precision | 0.272 [0.182, 0.340] | 0.277 [0.215, 0.372] | 0.332 [0.273, 0.400] |
+| Exact span recall | 0.096 [0.055, 0.141] | 0.127 [0.090, 0.162] | 0.166 [0.122, 0.216] |
+| Group exact-match F1 | 0.000 | 0.000 | 0.000 |
+| Document exact match, no conditions | 0.941 [0.840, 1.000] | 0.941 [0.800, 1.000] | 0.882 [0.714, 1.000] |
+| False-prerequisite rate (lower is better) | 0.330 [0.165, 0.467] | 0.416 [0.294, 0.493] | 0.353 [0.260, 0.416] |
+| Reversed-meaning rate (lower is better) | 0.060 [0.011, 0.132] | 0.090 [0.035, 0.154] | 0.074 [0.029, 0.131] |
+
+## Accuracy: 10 hand-labelled dev documents
+
+| Metric | `rules-2` | `crf-pilot-2-batch-01` | `crf-batches-01-02` |
+|---|---:|---:|---:|
+| Typed-condition micro-F1 | 0.095 [0.000, 0.333] | 0.444 [0.133, 0.667] | 0.370 [0.000, 0.640] |
+| Exact span precision | 0.250 [0.000, 1.000] | 0.600 [0.250, 1.000] | 0.500 [0.000, 0.818] |
+| Exact span recall | 0.059 [0.000, 0.231] | 0.353 [0.077, 0.611] | 0.294 [0.000, 0.571] |
+| Group exact-match F1 | 0.000 | 0.500 [0.000, 1.000] | 0.500 [0.000, 1.000] |
+| False-prerequisite rate (lower is better) | 0.750 [0.000, 1.000] | 0.300 [0.000, 0.750] | 0.333 [0.000, 1.000] |
+
+## Paired differences (same documents, same resamples)
+
+| Comparison | Slice | Typed F1, B − A | Exact span F1, B − A | Report |
+|---|---|---:|---:|---|
+| `crf-pilot-2-batch-01` → `crf-batches-01-02` (adding `batch-02`) | agreement | +0.017 [−0.014, +0.047] | **+0.047 [+0.004, +0.086]** | [`compare_agreement_batch-02.md`](compare_agreement_batch-02.md) |
+| `rules-2` → `crf-batches-01-02` | agreement | +0.038 [−0.012, +0.097] | **+0.079 [+0.028, +0.138]** | [`compare_agreement_rules.md`](compare_agreement_rules.md) |
+| `crf-pilot-2-batch-01` → `crf-batches-01-02` (adding `batch-02`) | accuracy | −0.074 [−0.182, +0.000] | −0.074 [−0.182, +0.000] | [`compare_manual_batch-02.md`](compare_manual_batch-02.md) |
+
+Bold marks a clear difference (the whole interval on one side of zero).
+
+## Reading
+
+- **Agreement is low.** Even the best system reproduces about one in six of the drafting
+  model's exact spans (span recall 0.166), and typed F1 stays below 0.15. Exact matching is
+  strict: a span that differs by one word counts as a miss.
+- **Adding `batch-02` clearly improves span agreement** (+0.047 exact span F1) but not typed
+  agreement: the interval for typed F1 includes zero.
+- **The CRF clearly beats the rules on span agreement** (+0.079), but not on typed F1 or on the
+  false-prerequisite rate.
+- **On the 10 hand-labelled documents nothing is clear.** Adding `batch-02` moves typed F1 from
+  0.444 to 0.370, with a paired interval of [−0.182, +0.000]. Ten documents cannot separate that
+  from chance.
+- **No group is matched** in the agreement slice. The drafts contain few groups (9 in
+  `batch-03`), and no system reproduces one exactly.
+
+None of these numbers is a release result, and agreement is not accuracy.
+
+## Reproduce
+
+```bash
+uv run triggermap train --gold annotations/pilot/pilot-2.jsonl annotations/batches/batch-01.jsonl annotations/batches/batch-02.jsonl --out artifacts/crf-batches-01-02 --model-id crf-batches-01-02
+uv run triggermap predict --model crf --model-dir artifacts/crf-batches-01-02 --sample annotations/pilot/pilot_manifest.json annotations/batches/batch-03.json annotations/batches/batch-06.json --out reports/agreement/predictions/crf-batches-01-02.jsonl
+uv run triggermap evaluate annotations/batches/batch-03.jsonl reports/agreement/predictions/crf-batches-01-02.jsonl --add-gold annotations/batches/batch-06.jsonl --split dev --out reports/agreement/crf-batches-01-02_dev_agreement.md
+uv run triggermap evaluate annotations/pilot/pilot-2.jsonl reports/agreement/predictions/crf-batches-01-02.jsonl --split dev --gold-origin manual --out reports/agreement/crf-batches-01-02_dev_manual.md
+uv run triggermap compare annotations/batches/batch-03.jsonl reports/agreement/predictions/crf-pilot-2-batch-01.jsonl reports/agreement/predictions/crf-batches-01-02.jsonl --add-gold annotations/batches/batch-06.jsonl --split dev --out reports/agreement/compare_agreement_batch-02.md
+```
+
+`rules-2` uses `predict` without `--model`, and `crf-pilot-2-batch-01` uses the bundle trained in
+[`reports/training-origin/`](../training-origin/README.md).
diff --git a/reports/agreement/compare_agreement_batch-02.md b/reports/agreement/compare_agreement_batch-02.md
@@ -0,0 +1,25 @@
+# Agreement: adding batch-02 (crf-pilot-2-batch-01 -> crf-batches-01-02)
+
+- Gold: `annotations/batches/batch-03.jsonl`, `annotations/batches/batch-06.jsonl` (dev split of `splits/pool-v1.json`; origin: all; 172 documents)
+- Baseline (A): `reports/agreement/predictions/crf-pilot-2-batch-01.jsonl` (crf-pilot-2-batch-01)
+- Candidate (B): `reports/agreement/predictions/crf-batches-01-02.jsonl` (crf-batches-01-02)
+- Bootstrap: 1000 paired resamples, project clusters, seed 0
+- Documents: 172
+
+Differences are B minus A with 95% paired bootstrap intervals. *Clear* means the whole interval lies on one side of zero.
+
+| Metric | A (crf-pilot-2-batch-01) | B (crf-batches-01-02) | B - A | 95% CI | Verdict |
+|---|---:|---:|---:|:---:|---|
+| Typed condition micro-F1 (span + category + role) | 0.128 | 0.145 | +0.017 | [-0.014, +0.047] | no clear difference |
+| Exact span F1 | 0.174 | 0.221 | +0.047 | [+0.004, +0.086] | B clearly better |
+| Exact span precision | 0.277 | 0.332 | +0.055 | [-0.026, +0.108] | no clear difference |
+| Exact span recall | 0.127 | 0.166 | +0.039 | [+0.006, +0.069] | B clearly better |
+| Role macro-F1 (roles with gold support) | 0.089 | 0.115 | +0.026 | [-0.001, +0.059] | no clear difference |
+| Category macro-F1 | 0.122 | 0.137 | +0.015 | [-0.007, +0.035] | no clear difference |
+| Group exact-match F1 | 0.000 | 0.000 | +0.000 | [+0.000, +0.000] | no clear difference |
+| Complete-document exact match, documents with conditions | 0.058 | 0.065 | +0.006 | [-0.015, +0.027] | no clear difference |
+| Complete-document exact match, documents without conditions | 0.941 | 0.882 | -0.059 | [-0.160, +0.000] | no clear difference |
+| False prerequisite rate (lower is better) | 0.416 | 0.353 | -0.062 | [-0.124, +0.036] | no clear difference |
+| Reversed meaning rate (lower is better) | 0.090 | 0.074 | -0.016 | [-0.054, +0.014] | no clear difference |
+| Category accuracy on exactly matched spans | 0.857 | 0.797 | -0.060 | [-0.138, -0.003] | B clearly worse |
+| Role accuracy on exactly matched spans | 0.878 | 0.859 | -0.018 | [-0.071, +0.064] | no clear difference |
diff --git a/reports/agreement/compare_agreement_rules.md b/reports/agreement/compare_agreement_rules.md
@@ -0,0 +1,25 @@
+# Agreement: rules-2 -> crf-batches-01-02
+
+- Gold: `annotations/batches/batch-03.jsonl`, `annotations/batches/batch-06.jsonl` (dev split of `splits/pool-v1.json`; origin: all; 172 documents)
+- Baseline (A): `reports/agreement/predictions/rules-2.jsonl` (rules-2)
+- Candidate (B): `reports/agreement/predictions/crf-batches-01-02.jsonl` (crf-batches-01-02)
+- Bootstrap: 1000 paired resamples, project clusters, seed 0
+- Documents: 172
+
+Differences are B minus A with 95% paired bootstrap intervals. *Clear* means the whole interval lies on one side of zero.
+
+| Metric | A (rules-2) | B (crf-batches-01-02) | B - A | 95% CI | Verdict |
+|---|---:|---:|---:|:---:|---|
+| Typed condition micro-F1 (span + category + role) | 0.107 | 0.145 | +0.038 | [-0.012, +0.097] | no clear difference |
+| Exact span F1 | 0.142 | 0.221 | +0.079 | [+0.028, +0.138] | B clearly better |
+| Exact span precision | 0.272 | 0.332 | +0.060 | [-0.018, +0.171] | no clear difference |
+| Exact span recall | 0.096 | 0.166 | +0.070 | [+0.030, +0.113] | B clearly better |
+| Role macro-F1 (roles with gold support) | 0.116 | 0.115 | -0.002 | [-0.071, +0.126] | no clear difference |
+| Category macro-F1 | 0.098 | 0.137 | +0.039 | [-0.002, +0.082] | no clear difference |
+| Group exact-match F1 | 0.000 | 0.000 | +0.000 | [+0.000, +0.000] | no clear difference |
+| Complete-document exact match, documents with conditions | 0.019 | 0.065 | +0.045 | [+0.007, +0.084] | B clearly better |
+| Complete-document exact match, documents without conditions | 0.941 | 0.882 | -0.059 | [-0.200, +0.000] | no clear difference |
+| False prerequisite rate (lower is better) | 0.330 | 0.353 | +0.023 | [-0.091, +0.170] | no clear difference |
+| Reversed meaning rate (lower is better) | 0.060 | 0.074 | +0.014 | [-0.047, +0.067] | no clear difference |
+| Category accuracy on exactly matched spans | 0.865 | 0.797 | -0.068 | [-0.199, +0.083] | no clear difference |
+| Role accuracy on exactly matched spans | 0.892 | 0.859 | -0.033 | [-0.113, +0.092] | no clear difference |
diff --git a/reports/agreement/compare_manual_batch-02.md b/reports/agreement/compare_manual_batch-02.md
@@ -0,0 +1,25 @@
+# Hand-labelled dev: adding batch-02
+
+- Gold: `annotations/pilot/pilot-2.jsonl` (dev split of `splits/pool-v1.json`; origin: manual; 10 documents)
+- Baseline (A): `reports/agreement/predictions/crf-pilot-2-batch-01.jsonl` (crf-pilot-2-batch-01)
+- Candidate (B): `reports/agreement/predictions/crf-batches-01-02.jsonl` (crf-batches-01-02)
+- Bootstrap: 1000 paired resamples, project clusters, seed 0
+- Documents: 10
+
+Differences are B minus A with 95% paired bootstrap intervals. *Clear* means the whole interval lies on one side of zero.
+
+| Metric | A (crf-pilot-2-batch-01) | B (crf-batches-01-02) | B - A | 95% CI | Verdict |
+|---|---:|---:|---:|:---:|---|
+| Typed condition micro-F1 (span + category + role) | 0.444 | 0.370 | -0.074 | [-0.182, +0.000] | no clear difference |
+| Exact span F1 | 0.444 | 0.370 | -0.074 | [-0.182, +0.000] | no clear difference |
+| Exact span precision | 0.600 | 0.500 | -0.100 | [-0.500, +0.027] | no clear difference |
+| Exact span recall | 0.353 | 0.294 | -0.059 | [-0.200, +0.000] | no clear difference |
+| Role macro-F1 (roles with gold support) | 0.231 | 0.200 | -0.031 | [-0.122, +0.000] | no clear difference |
+| Category macro-F1 | 0.267 | 0.237 | -0.030 | [-0.133, +0.000] | no clear difference |
+| Group exact-match F1 | 0.500 | 0.500 | +0.000 | [+0.000, +0.000] | no clear difference |
+| Complete-document exact match, documents with conditions | 0.222 | 0.222 | +0.000 | [+0.000, +0.000] | no clear difference |
+| Complete-document exact match, documents without conditions | 1.000 | 1.000 | +0.000 | [+0.000, +0.000] | no clear difference |
+| False prerequisite rate (lower is better) | 0.300 | 0.333 | +0.033 | [-0.050, +0.500] | no clear difference |
+| Reversed meaning rate (lower is better) | 0.000 | 0.000 | +0.000 | [+0.000, +0.000] | no clear difference |
+| Category accuracy on exactly matched spans | 1.000 | 1.000 | +0.000 | [-1.000, +0.000] | no clear difference |
+| Role accuracy on exactly matched spans | 1.000 | 1.000 | +0.000 | [-1.000, +0.000] | no clear difference |
diff --git a/reports/agreement/crf-batches-01-02_dev_agreement.md b/reports/agreement/crf-batches-01-02_dev_agreement.md
@@ -0,0 +1,57 @@
+# Agreement with model drafts: crf-batches-01-02
+
+- Gold: `annotations/batches/batch-03.jsonl`, `annotations/batches/batch-06.jsonl` (dev split of `splits/pool-v1.json`; origin: all; 172 documents)
+- Predictions: `reports/agreement/predictions/crf-batches-01-02.jsonl` (crf-batches-01-02)
+- Bootstrap: 1000 resamples, project clusters, seed 0
+- Documents: 172 (155 with gold conditions, 17 without)
+
+## Headline metrics (95% bootstrap intervals)
+
+| Metric | Estimate | 95% CI |
+|---|---:|:---:|
+| Typed condition micro-F1 (span + category + role) | 0.145 | [0.105, 0.185] |
+| Exact span F1 | 0.221 | [0.171, 0.268] |
+| Exact span precision | 0.332 | [0.273, 0.400] |
+| Exact span recall | 0.166 | [0.122, 0.216] |
+| Role macro-F1 (roles with gold support) | 0.115 | [0.065, 0.160] |
+| Category macro-F1 | 0.137 | [0.094, 0.174] |
+| Group exact-match F1 | 0.000 | [0.000, 0.000] |
+| Complete-document exact match, documents with conditions | 0.065 | [0.030, 0.104] |
+| Complete-document exact match, documents without conditions | 0.882 | [0.714, 1.000] |
+| False prerequisite rate (lower is better) | 0.353 | [0.260, 0.416] |
+| Reversed meaning rate (lower is better) | 0.074 | [0.029, 0.131] |
+| Category accuracy on exactly matched spans | 0.797 | [0.667, 0.880] |
+| Role accuracy on exactly matched spans | 0.859 | [0.758, 0.976] |
+
+## Per-role results (exact span + role)
+
+| Label | Precision | Recall | F1 | Gold support |
+|---|---:|---:|---:|---:|
+| REQUIRED | 0.305 | 0.158 | 0.208 | 323 |
+| EXCEPTION | 0.000 | 0.000 | 0.000 | 17 |
+| NOT_REQUIRED | 0.286 | 0.222 | 0.250 | 18 |
+| UNCERTAIN | 0.000 | 0.000 | 0.000 | 28 |
+
+## Per-category results (exact span + category)
+
+| Label | Precision | Recall | F1 | Gold support |
+|---|---:|---:|---:|---:|
+| CONFIGURATION | 0.095 | 0.056 | 0.070 | 36 |
+| USAGE | 0.189 | 0.097 | 0.128 | 186 |
+| ACCESS | 0.405 | 0.224 | 0.288 | 134 |
+| PLATFORM | 0.333 | 0.033 | 0.061 | 30 |
+
+## Precision versus coverage
+
+| Coverage | Confidence ≥ | Typed precision | Retained recall |
+|---:|---:|---:|---:|
+| 10% | 0.834 | 0.316 | 0.016 |
+| 20% | 0.736 | 0.385 | 0.039 |
+| 30% | 0.670 | 0.431 | 0.065 |
+| 40% | 0.591 | 0.351 | 0.070 |
+| 50% | 0.505 | 0.323 | 0.080 |
+| 60% | 0.464 | 0.276 | 0.083 |
+| 70% | 0.421 | 0.267 | 0.093 |
+| 80% | 0.385 | 0.253 | 0.101 |
+| 90% | 0.333 | 0.236 | 0.106 |
+| 100% | 0.186 | 0.218 | 0.109 |
diff --git a/reports/agreement/crf-batches-01-02_dev_manual.md b/reports/agreement/crf-batches-01-02_dev_manual.md
@@ -0,0 +1,57 @@
+# Hand-labelled dev: crf-batches-01-02
+
+- Gold: `annotations/pilot/pilot-2.jsonl` (dev split of `splits/pool-v1.json`; origin: manual; 10 documents)
+- Predictions: `reports/agreement/predictions/crf-batches-01-02.jsonl` (crf-batches-01-02)
+- Bootstrap: 1000 resamples, project clusters, seed 0
+- Documents: 10 (9 with gold conditions, 1 without)
+
+## Headline metrics (95% bootstrap intervals)
+
+| Metric | Estimate | 95% CI |
+|---|---:|:---:|
+| Typed condition micro-F1 (span + category + role) | 0.370 | [0.000, 0.640] |
+| Exact span F1 | 0.370 | [0.000, 0.640] |
+| Exact span precision | 0.500 | [0.000, 0.818] |
+| Exact span recall | 0.294 | [0.000, 0.571] |
+| Role macro-F1 (roles with gold support) | 0.200 | [0.000, 0.600] |
+| Category macro-F1 | 0.237 | [0.000, 0.524] |
+| Group exact-match F1 | 0.500 | [0.000, 1.000] |
+| Complete-document exact match, documents with conditions | 0.222 | [0.000, 0.455] |
+| Complete-document exact match, documents without conditions | 1.000 | [0.000, 1.000] |
+| False prerequisite rate (lower is better) | 0.333 | [0.000, 1.000] |
+| Reversed meaning rate (lower is better) | 0.000 | [0.000, 0.000] |
+| Category accuracy on exactly matched spans | 1.000 | [0.000, 1.000] |
+| Role accuracy on exactly matched spans | 1.000 | [0.000, 1.000] |
+
+## Per-role results (exact span + role)
+
+| Label | Precision | Recall | F1 | Gold support |
+|---|---:|---:|---:|---:|
+| REQUIRED | 0.556 | 0.312 | 0.400 | 16 |
+| EXCEPTION | 0.000 | 0.000 | 0.000 | 0 |
+| NOT_REQUIRED | 0.000 | 0.000 | 0.000 | 0 |
+| UNCERTAIN | 0.000 | 0.000 | 0.000 | 1 |
+
+## Per-category results (exact span + category)
+
+| Label | Precision | Recall | F1 | Gold support |
+|---|---:|---:|---:|---:|
+| CONFIGURATION | 0.000 | 0.000 | 0.000 | 1 |
+| USAGE | 0.429 | 0.333 | 0.375 | 9 |
+| ACCESS | 1.000 | 0.400 | 0.571 | 5 |
+| PLATFORM | 0.000 | 0.000 | 0.000 | 2 |
+
+## Precision versus coverage
+
+| Coverage | Confidence ≥ | Typed precision | Retained recall |
+|---:|---:|---:|---:|
+| 10% | 0.905 | 0.000 | 0.000 |
+| 20% | 0.871 | 0.500 | 0.059 |
+| 30% | 0.815 | 0.667 | 0.118 |
+| 40% | 0.791 | 0.750 | 0.176 |
+| 50% | 0.755 | 0.800 | 0.235 |
+| 60% | 0.666 | 0.667 | 0.235 |
+| 70% | 0.644 | 0.571 | 0.235 |
+| 80% | 0.589 | 0.500 | 0.235 |
+| 90% | 0.441 | 0.444 | 0.235 |
+| 100% | 0.393 | 0.500 | 0.294 |