full dataset: 400-case benchmark (test) + 1200-case training split, verified disjoint

README.md

@@ -15,24 +15,34 @@ tags:
 - workflow-evaluation
 - synthetic
 configs:
-- config_name: customer_service
+- config_name: agent_trace_observability
   data_files:
+  - split: train
+    path: agent_trace_observability/train-*.parquet
   - split: test
-    path: customer_service/test-*.parquet
-- config_name: security_incidents
+    path: agent_trace_observability/test-*.parquet
+- config_name: customer_service
   data_files:
+  - split: train
+    path: customer_service/train-*.parquet
   - split: test
-    path: security_incidents/test-*.parquet
+    path: customer_service/test-*.parquet
 - config_name: invoice_processing
   data_files:
+  - split: train
+    path: invoice_processing/train-*.parquet
   - split: test
     path: invoice_processing/test-*.parquet
-- config_name: agent_trace_observability
+- config_name: security_incidents
   data_files:
+  - split: train
+    path: security_incidents/train-*.parquet
   - split: test
-    path: agent_trace_observability/test-*.parquet
+    path: security_incidents/test-*.parquet
 - config_name: all
   data_files:
+  - split: train
+    path: all/train-*.parquet
   - split: test
     path: all/test-*.parquet
 ---
@@ -73,12 +83,12 @@ part of the input: a bare label list is a different, easier task.
 
 ## Workflows
 
-| Workflow | Decision | Cases |
-|---|---|---|
-| `customer_service` | Determine the appropriate assistant response and action from a customer thread and account state. | 100 |
-| `security_incidents` | Decide whether a security alert should be closed, investigated, or contained, from alert data and machine history. | 100 |
-| `invoice_processing` | Review a vendor bill against the order and delivery to determine payment, hold or rejection. | 100 |
-| `agent_trace_observability` | Assess an agent run to decide whether human review is needed, and how urgent it is. | 100 |
+| Workflow | Decision | Train | Test |
+|---|---|---|---|
+| `agent_trace_observability` | Assess an agent run to decide whether human review is needed, and how urgent it is. | 300 | 100 |
+| `customer_service` | Determine the appropriate assistant response and action from a customer thread and account state. | 300 | 100 |
+| `invoice_processing` | Review a vendor bill against the order and delivery to determine payment, hold or rejection. | 300 | 100 |
+| `security_incidents` | Decide whether a security alert should be closed, investigated, or contained, from alert data and machine history. | 300 | 100 |
 
 Each case asks **5 questions over one shared state**.
 
@@ -108,7 +118,8 @@ request, so a row can be replayed without reshaping.
 from datasets import load_dataset
 import json
 
-ds = load_dataset("codelion/typed-decisions", "customer_service", split="test")
+ds = load_dataset("codelion/typed-decisions", "customer_service", split="test")   # benchmark
+tr = load_dataset("codelion/typed-decisions", "customer_service", split="train")  # training data
 row = ds[0]
 
 state = json.loads(row["state"])
@@ -200,14 +211,17 @@ text, not drawn from real support queues, SIEMs or accounts-payable systems.
 Results here describe architecture and systems behaviour, not real-world
 accuracy.
 
-**It is small.** 400 cases, 2000 typed decisions -- all of them
+**It is small.** 1600 cases, 8000 typed decisions -- all of them
 held out. Enough for decision-level metrics; thin for strong per-workflow claims.
 The generator is seeded and extensible, so the benchmark can grow without
 invalidating these cases.
 
-**It is a benchmark, not a training corpus.** Every case is a held-out test
-case. Generate training or calibration data separately, with a different seed;
-do not carve a train split out of this one.
+**Train and test are generated separately and verified disjoint.** The `test`
+split is the benchmark; the `train` split comes from a different seed with
+namespaced case ids. The packaging step checks that no case id and no state
+appears in both and refuses to build if either does -- that check exists because
+the v0.1 prototype leaked exactly this way and nobody noticed until models had
+already been benchmarked on it.
 
 ## What the audit checks
 

agent_trace_observability/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7b2f78fe539517c6a66a90c68ed671c6133194a88d27dbff5d84713af46b901
+size 157731

all/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6c777b77cb1b45fd85ed83d1bdfdb328bb728b2bb7d05170a34a42b5f4228565
-size 222129
+oid sha256:4f294f218ea1da27f3efef936359389c62ea4d3973a41457732990f1d31b647c
+size 222140

all/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46a58d63edfd86e23229c78afe8b72307bb4ca9fb0e8df180cabb3c67ec9dcd5
+size 598824

customer_service/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e3abc2a9307bc641f60a6f9042627897ca01c145c3206211dfa9c926ba8721c
+size 267280

invoice_processing/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a5cfcd12e77e707cd9861107868a5adbb7df566ced3162bb9472cd3e348a598
+size 177403

security_incidents/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d9dcda135003e6812bec87461fcf53591e8b5426b077836616cde871d26f8dda
-size 87540
+oid sha256:4d8bc1879a0b67e5d275edea21a694b06136ca2a2c3355215a7d8adea6b0c75f
+size 87929

security_incidents/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3aae379ac25fc6434d5f5675f04b06f4310d1aac3056a0e30bb0f779aa2f0a99
+size 201079