Initial import: XAUUSD news sentiment models, evals, and data tooling

TF-IDF baseline (news_sentiment.py, models/xauusd_sentiment.joblib) and the
two-head DeBERTa-v3-small transformer (transformer/, news_transformer.py)
with its Google News crawler and DeepSeek labeller (data_collect/), the
golden-set and archive evals (eval/), and the 2026-09-22 and 2026-09-27
results and progress notes.

Transformer weights (model.pt, model.onnx) and the crawl database are not
tracked; rebuild with transformer/train.py and transformer/export_onnx.py.
The DeepSeek label cache in data/ is tracked.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

eval/run_all.sh

@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+# Runs the three evaluations with the ForexBot2 venv (it has requests,
+# pandas and the live news code; the model loads identically there).
+# Total DeepSeek spend on a first run: ~4 calls (golden) + ~4 (reprobe)
+# + one call per 25 fresh headlines (live). Everything is cached, so a
+# second run makes no calls unless new headlines arrived.
+set -euo pipefail
+PY=/home/harwinder/Development/ForexBot2/venv/bin/python
+cd "$(dirname "$0")"
+
+echo "### 1/3 golden set (independent human labels)"
+"$PY" eval_golden.py "$@"
+
+echo
+echo "### 2/3 archive time-split + DeepSeek self-agreement"
+"$PY" eval_archive.py --reprobe 100 "$@"
+
+echo
+echo "### 3/3 live shadow run (24 h of headlines)"
+"$PY" eval_live.py --max-age-hours 24 "$@"