Restore: 111 Ukraine markets × 672 cutoffs × 4 conditions × 7 horizons

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+data/unlp-ukraine-D_UA-train.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,77 @@
+---
+license: mit
+language:
+- en
+tags:
+- prediction-markets
+- forecasting
+- polymarket
+- ukraine
+- unlp
+size_categories:
+- 1K<n<10K
+---
+
+# UNLP Ukraine Forecasting Dataset
+
+Prediction market forecasting dataset for **111 resolved Ukraine-related Polymarket markets** with **672 frozen cutoff points** and **7 fixed prediction horizons** (6h, 12h, 1d, 2d, 3d, 5d, 7d).
+
+Built by walking the exact same code path as the UNLP 2026 evaluation harness (`eval/harness.py:build_prompt()`), but emitting dataset rows instead of making LLM API calls. Every prompt, context block, and ground-truth label is identical to what the eval pipeline produces.
+
+## Conditions
+
+Each cutoff is rendered under 4 information conditions:
+
+| Condition | Context provided |
+|---|---|
+| **A** | Price summary only (blind baseline) |
+| **C** | A + news context (48h/7d/30d article windows) |
+| **D** | A + chart + news + cross-market context + comments |
+| **D_UA** | D + Ukrainian sources (Telegram bloggers, casualties, Militarnyi) |
+
+## Schema
+
+Each JSONL row contains:
+
+**Identity:** `market_id`, `question`, `event_slug`, `cutoff_timestamp`, `hours_remaining`, `condition`, `split`
+
+**Prompt (model sees):** `prompt` (full rendered V4 prompt, same as the harness sends to LLMs), `has_chart`, `has_war_map`
+
+**Ground truth:** `current_price`, `labels` (dict mapping horizon labels to actual prices: `{"6h": 0.35, ..., "7d": 0.42}`), `outcome` (binary), `end_date`
+
+**Baselines:** `baseline_nochange` (predict current price unchanged), `baseline_momentum` (7-day linear regression extrapolated)
+
+**Volume:** `volume_at_cutoff_24h` (USD notional in 24h before cutoff)
+
+## Horizons
+
+Fixed 7 prediction horizons from each cutoff:
+
+| Label | Hours forward |
+|---|---|
+| 6h | 6 |
+| 12h | 12 |
+| 1d | 24 |
+| 2d | 48 |
+| 3d | 72 |
+| 5d | 120 |
+| 7d | 168 |
+
+Ground truth is the actual market price at each horizon timestamp (6-hour tolerance for nearest available price point).
+
+## Scoring
+
+**MAE** (Mean Absolute Error): `|predicted_price - actual_price|` at each horizon. Compare against:
+- No-change baseline: `predicted = current_price`
+- Momentum baseline: 7-day linear regression extrapolated forward
+
+## Stats
+
+- **111 markets**, 672 cutoffs, 4 conditions = **2,688 total examples**
+- **Train/test split:** 89/22 markets (temporal by resolution date)
+- **All 7 horizons have labels** for every example (7.0/7 avg)
+- **99% of examples** have 24h volume > 0
+
+## Source
+
+From the [OpenBabylon/signal](https://github.com/OpenBabylon/signal) repo. Built using `eval/build_unlp_dataset.py` which calls the same `build_prompt()` and `compute_labels()` as the evaluation harness.

data/unlp-ukraine-D_UA-train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de790e57e70999942cb21e0c318df3d378686dd311f50ae4f53fb7f801135127
+size 10904603

metadata.json

@@ -0,0 +1,31 @@
+{
+  "created_at": "2026-04-10T23:04:19.858257+00:00",
+  "benchmark": "ukraine",
+  "source": "eval/benchmarks/ukraine.yaml",
+  "pipeline": "eval/harness.py build_prompt() \u2014 same code path as the UNLP eval",
+  "prompt_version": "v4",
+  "horizons": {
+    "6h": "6h",
+    "12h": "12h",
+    "1d": "24h",
+    "2d": "48h",
+    "3d": "72h",
+    "5d": "120h",
+    "7d": "168h"
+  },
+  "conditions_built": [
+    "A",
+    "C",
+    "D",
+    "D_UA"
+  ],
+  "train_fraction": 0.8,
+  "total_examples": 2688,
+  "markets": 111,
+  "cutoffs": 672,
+  "scoring": "MAE: |predicted_price - actual_price| at each horizon",
+  "baselines": {
+    "nochange": "predicted = current_price (no change)",
+    "momentum": "7-day linear regression extrapolated forward"
+  }
+}