feat(sentiment): score news sentiment with capable tier (sonnet-4.6)#21
Open
Zeeechenn wants to merge 1 commit into
Open
feat(sentiment): score news sentiment with capable tier (sonnet-4.6)#21Zeeechenn wants to merge 1 commit into
Zeeechenn wants to merge 1 commit into
Conversation
…titles from LLM input Acts on the 2026-06-15 clean single-provider OOS: sentiment IC measured 0.0735 under sonnet-4.6 vs ~0.020 under the fast/Codex tier — scoring-model quality is the dominant lever for this signal. - config: sentiment_model_tier (default "capable" → claude-sonnet-4-6 on every provider). analyze_news + the m27 backfill tool now score at this tier instead of the hardcoded "fast". NOTE: with AI_PROVIDER=local_cli also set LOCAL_CLI_PREFER_CODEX=false, else the Codex path ignores the model. - analyze_news: the company-evidence check is now a full filter — only company-specific titles (after market-flow + alias relevance) are sent to the LLM and used for the cache key; a window with none returns neutral and skips the call. - backtest news_cache: resolves stock name+code aliases from Stock metadata and passes them through so the relevance filter applies on that path too. Verification: full suite 1225 passed / 6 skipped; ruff + mypy clean. New tests assert only company-specific titles reach the LLM, the backtest path forwards aliases, and sentiment scores at the configured tier. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #20（base =
feat/news-sentiment-pack）。从新闻层 PR 拆出，单独 review/合并，因为它改模型/成本行为，风险类与纯防御的新闻层加固不同。#20 合并后本 PR 自动 retarget 到 main。内容（仅 4 文件）
config.py：新增sentiment_model_tier（默认capable→claude-sonnet-4-6，每个 provider 都映射到它）。analyze_news+ m27 backfill：model_tier从硬编码fast改为读配置。依据
2026-06-15 干净单 provider OOS：sentiment IC 0.0735@sonnet-4.6 vs ~0.020@fast/Codex —— 打分模型质量是该信号的最大杠杆。
AI_PROVIDER=anthropic（云 API 付费、稳，适合 88 股日批）；或AI_PROVIDER=local_cli+LOCAL_CLI_PREFER_CODEX=false（走 claude CLI / CC 订阅免 API 费，但批量有日配额截断风险）。local_cli默认仍走 Codex，本改动不生效。CI：随 PR 运行。
Security/dependency audit的 fail 是cryptography 48.0.0→48.0.1 等依赖 CVE，与本 PR 无关。🤖 Generated with Claude Code