Benchmark Radar Day 111: Your Trading Agent Already Knows Tomorrow’s News
Published:
Day 111 of Benchmark Radar. The scan fetched 1,492 items, published 694, and marked 317 as recommended.
Author: Koutian Wu; GitHub: ktwu01
Benchmark Radar is live. Open the dashboard, star the repository, or read the paper. The Hugging Face dataset provides the data.
rta-bench is a new benchmark that tests language-model trading agents without letting them trade on future information. It replays each headline at its publication minute, fills orders at the next market open, and uses three frozen historical markets. This directly addresses backtests that join news and prices by date, where a 16:05 headline could influence a trade at the earlier 16:00 close. If you are selecting an evaluation for trading agents, event timing and executable order timing are part of the benchmark here, so apparent returns measure agent decisions rather than accidental access to information that was unavailable when the trade supposedly occurred. jacksonwangpro1/rta-bench-llm-trading-benchmark.
Ragas v0.4.4 is a maintenance and security update, not a new evaluation suite. It restores fresh installations broken by a dependency change, makes imports of removed Vertex AI modules optional, fixes a published vulnerability in multimodal metrics, and adjusts requests for current Anthropic models. If you already use Ragas to evaluate current OpenAI or Anthropic models, earlier installations may fail or send parameters that newer models reject, and multimodal evaluation carries the cited security issue. Results produced before and after upgrading may need their Ragas version recorded. vibrantlabsai/ragas v0.4.4.
Why this matters
A backtest that lets a 16:05 headline move a 16:00 trade is not measuring trading skill. It is measuring leakage. And the tool doing the measuring needs its own version recorded, because an unpatched evaluator can fail silently or score a different model than the one you think you ran. Timing is part of the test, and the test harness is part of the result.
Scan coverage
This is a keyword-filtered discovery feed, not a representative survey of AI research. The archived briefing used 140 of 694 captured evidence records. These findings summarize its cited releases; discovery dates do not necessarily equal original publication dates.
Data and briefing: 2026-11-14 archived snapshot.
