news-reaction-failure-analyzer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

News Reaction Failure Analyzer

新闻反应失败分析器

Overview

概述

Implements step 2 of Jason Shapiro's COT contrarian process: once a market is flagged as crowded (

cot-contrarian-detector

, step 1), check whether it FAILED to react to news that should have rewarded the crowd. A crowded-long market that doesn't rally on genuinely bullish news, or a crowded-short market that doesn't sell off on genuinely bearish news, is the core behavioral tell that the crowd has run out of buying/selling power — this is the confirmation step that turns "crowded" into a contrarian setup candidate (steps 3-5, still manual: price-action confirmation, entry, exit).

Why this isn't a naive failure-ratio check: an earlier design flagged "news failure" whenever fewer than half the relevant events "responded" — but under pure noise, roughly 69% of individual events fail to respond by chance, so that rule would CONFIRM on random noise 48-83% of the time depending on sample size. This skill instead requires the market to have moved significantly against the crowd's favorable news (a drift- significance test with a Monte-Carlo-verified null false-positive bound), never merely "didn't respond enough." See

references/news-failure-patterns.md

for the full statistical rationale.

实现Jason Shapiro的COT反向策略流程的第二步：一旦某市场被标记为拥挤状态（由

cot-contrarian-detector

工具的第一步流程识别），检查该市场是否未能对本应让拥挤头寸获利的新闻做出反应。拥挤多头市场未在真正利好的新闻下上涨，或拥挤空头市场未在真正利空的新闻下下跌，这是表明拥挤群体已耗尽买卖力量的核心行为信号——这一步确认流程将“拥挤状态”转化为反向交易候选机会（后续第3-5步仍需人工操作：价格行为确认、入场、出场）。

为何不采用简单失败率检查：早期设计中，只要不足半数相关事件“做出反应”就标记为“新闻反应失败”——但在纯噪声情况下，约69%的单个事件会随机无反应，因此该规则会在48%-83%的时间内误判为确认状态（取决于样本量）。本工具则要求市场对有利于拥挤头寸的新闻做出显著反向变动（采用经蒙特卡洛验证的零假设假阳性边界的漂移显著性测试），而非仅仅“反应不足”。详细统计原理请见

references/news-failure-patterns.md

。

When to Use This Skill

适用场景

English:

"Did the market shrug off [event] even though [asset] is crowded long/short?"
"Run a news-failure check on [symbol]"
"Is [symbol] confirmed for a Shapiro-style contrarian setup?"
After
```
cot-contrarian-detector
```
flags a market CROWDED_LONG / CROWDED_SHORT and the user wants to move to step 2

Japanese:

「この市場は好材料に反応しなかった？」
「COTで偏っているこの銘柄のニュース失敗を確認して」

Do NOT use when:

The market isn't crowded (NEUTRAL classification) — this skill refuses fail-closed without an explicit
```
--direction
```
override
No curated events JSON exists yet — WebSearch must run first (Phase 2 below); never fabricate events or URLs to get a verdict

英文场景：

"Did the market shrug off [event] even though [asset] is crowded long/short?"
"Run a news-failure check on [symbol]"
"Is [symbol] confirmed for a Shapiro-style contrarian setup?"
当
```
cot-contrarian-detector
```
标记某市场为CROWDED_LONG/CROWDED_SHORT，且用户希望进入第二步流程时

日语场景：

「この市場は好材料に反応しなかった？」
「COTで偏っているこの銘柄のニュース失敗を確認して」

不适用场景：

市场未处于拥挤状态（NEUTRAL分类）——若无明确
```
--direction
```
覆盖参数，本工具将拒绝生成判定结果
尚未生成整理后的事件JSON文件——需先运行WebSearch（下文第二阶段）；绝不允许伪造事件或URL来生成判定结果

Prerequisites

前置条件

FMP API Key: Required. Set
```
FMP_API_KEY
```
or pass
```
--api-key
```
. Used for price data only (
```
stable/historical-price-eod/light
```
) — coverage varies by symbol; see
```
references/price-source-map.md
```
.
Python 3.9+ with
```
requests
```
installed.
WebSearch access to curate the events JSON (Phase 2). Skill degrades gracefully without it (states the limitation; never fabricates events).
Optional: a
```
cot-contrarian-detector
```
JSON report (
```
--detector-json
```
) to auto-resolve symbol + direction, or supply
```
--direction
```
explicitly.

**FMP API密钥：**必填。设置环境变量
```
FMP_API_KEY
```
或通过
```
--api-key
```
参数传入。仅用于获取价格数据（
```
stable/historical-price-eod/light
```
）——覆盖范围因标的而异；详见
```
references/price-source-map.md
```
。
Python 3.9+ 环境，并已安装
```
requests
```
库。
WebSearch访问权限：用于整理事件JSON文件（第二阶段）。若无该权限，工具会优雅降级（说明限制；绝不伪造事件）。
可选：
```
cot-contrarian-detector
```
生成的JSON报告（
```
--detector-json
```
参数）：用于自动解析标的+方向，也可直接通过
```
--direction
```
参数明确指定方向。

Workflow

工作流程

Phase 1: Obtain symbol + direction

第一阶段：获取标的+方向

From a

cot-contrarian-detector

report (

--detector-json

, symbol looked up in

markets[]

) or directly from the user (

--symbol

--direction

). A

NEUTRAL

classification, a symbol missing from the report, or a report older than

--max-detector-age-days

(default 10) all refuse fail-closed with a specific reason — only an explicit

--direction

overrides.

通过

cot-contrarian-detector

报告（

--detector-json

参数，从

markets[]

中读取标的）或直接从用户输入获取（

--symbol

--direction

参数）。若报告中为NEUTRAL分类、标的未在报告中存在、或报告早于

--max-detector-age-days

（默认10天），工具将明确拒绝生成判定结果——仅当传入明确的

--direction

覆盖参数时例外。

Phase 2: Curate the events JSON via WebSearch

第二阶段：通过WebSearch整理事件JSON文件

Search news in the evaluation window (

--window-days

, default 10) using the 4-tier source hierarchy (issuer/primary → SEC/official stats → wire → portal — see

references/news-failure-patterns.md

). Write findings into an events JSON from

references/news-failure-patterns.md

's template —

event

event_time

(ISO8601 with explicit UTC offset),

source_url

source_tier

expected_impact

(BULLISH/BEARISH) per event.

Never fabricate events or URLs. WebSearch unavailable → state it explicitly; proceed without an events JSON only if the user accepts an

INSUFFICIENT_EVIDENCE

result (reason

no_events_provided

) — the CLI never raises an exception for a missing events file, it always exits 0 with a documented reason.

在评估窗口内（

--window-days

，默认10天）搜索新闻，采用4层级来源优先级（发行方/官方来源→SEC/官方统计→通讯社→门户平台——详见

references/news-failure-patterns.md

）。将搜索结果写入

references/news-failure-patterns.md

模板中的事件JSON文件——每个事件需包含

event

、

event_time

（带明确UTC偏移的ISO8601格式）、

source_url

、

source_tier

、

expected_impact

（BULLISH/BEARISH）字段。

绝不允许伪造事件或URL。若无WebSearch访问权限→明确说明；仅当用户接受

INSUFFICIENT_EVIDENCE

结果（原因

no_events_provided

）时，才可在无事件JSON文件的情况下继续——CLI绝不会因缺少事件文件抛出异常，始终会返回0退出码并说明原因。

Phase 3: Run the CLI

第三阶段：运行CLI

bash

python3 skills/news-reaction-failure-analyzer/scripts/analyze_news_reaction.py \
  --symbol B6 --detector-json reports/cot_crowding_2026-07-12.json \
  --events-json reports/nrf_events_B6_2026-07-12.json \
  --output-dir reports/

The script fetches the price series (documented fallback chain — futures symbol first, ETF proxy if 402/restricted or

rows == 0

; see

references/price-source-map.md

), computes effective dates / returns / z-scores per event, clusters events whose 3-trading-day windows overlap (independence guard), and synthesizes the verdict.

bash

python3 skills/news-reaction-failure-analyzer/scripts/analyze_news_reaction.py \
  --symbol B6 --detector-json reports/cot_crowding_2026-07-12.json \
  --events-json reports/nrf_events_B6_2026-07-12.json \
  --output-dir reports/

脚本会获取价格序列（有文档记录的回退链——优先使用期货标的，若返回402/受限或

rows == 0

则使用ETF替代；详见

references/price-source-map.md

），计算每个事件的有效日期/收益率/z值，对3个交易日窗口重叠的事件进行聚类（独立性校验），并生成最终判定结果。

Phase 4: Present verdict + handoff

第四阶段：展示判定结果+流程交接

Present the verdict, aggregate stats (

drift_stat

responded_ratio

), and the evidence table (per-event returns/z-scores/reaction labels, with any

dropped_events

reasons shown — never silently hidden). If a proxy (

run_context.proxy_used

) was used, note the tracking-error caveat.

Emit a handoff block for

contrarian-setup-gate

(#241, not yet built):

json

{"news_failure": {"verdict": "CONFIRMED", "confidence": "HIGH", "report_path": "reports/nrf_B6_2026-07-12.json"}}

展示判定结果、汇总统计数据（

drift_stat

、

responded_ratio

）以及证据表格（每个事件的收益率/z值/反应标签，同时展示

dropped_events

的原因——绝不隐藏）。若使用了替代标的（

run_context.proxy_used

），需注明跟踪误差提示。

输出用于

contrarian-setup-gate

（#241，尚未开发）的交接块：

json

{"news_failure": {"verdict": "CONFIRMED", "confidence": "HIGH", "report_path": "reports/nrf_B6_2026-07-12.json"}}

Output

输出结果

JSON:

reports/nrf_<symbol>_<as-of-date>.json

—

schema_version

symbol

direction

expected_direction

actual_reaction

(

FAILED_TO_RALLY

FAILED_TO_SELL_OFF

RALLIED

SOLD_OFF

MIXED_REACTION

NO_DATA

verdict

confidence

relevant_events_used

aggregate

(mean_z3/drift_stat/responded_ratio),

evidence[]

dropped_events[]

run_context

Markdown:
```
reports/nrf_<symbol>_<as-of-date>.md
```
— human-readable verdict, aggregate stats, evidence table, dropped-events table, proxy caveat (if used), and methodology footnote.

JSON文件：

reports/nrf_<symbol>_<as-of-date>.json

——包含

schema_version

、

symbol

、

direction

、

expected_direction

、

actual_reaction

（

FAILED_TO_RALLY

FAILED_TO_SELL_OFF

RALLIED

SOLD_OFF

MIXED_REACTION

NO_DATA

）、

verdict

、

confidence

、

relevant_events_used

、

aggregate

（mean_z3/drift_stat/responded_ratio）、

evidence[]

、

dropped_events[]

、

run_context

字段。

Markdown文件：
```
reports/nrf_<symbol>_<as-of-date>.md
```
——人类可读的判定结果、汇总统计数据、证据表格、事件剔除表格、替代标的提示（若使用）以及方法论脚注。

Guardrails

防护规则

CONFIRMED is not a trade signal. It confirms step 2 of 5 — price- action confirmation (step 3), entry (step 4), and exit (step 5) are still manual and still required before any position.
INSUFFICIENT_EVIDENCE never advances the pipeline. Fewer than
```
--min-events
```
(default 3) usable relevant event clusters, a missing detector report, or a detector vintage (
```
data_date
```
) that's missing, unparsable, dated after
```
--as-of
```
, or older than
```
--max-detector-age-days
```
(stale), a
```
NEUTRAL
```
classification without an explicit override, or no working price source all produce this verdict — never a crash, never a forced call on inadequate data.
COT publication lag. COT data is 3-9 days old by the time it's read (see
```
cot-contrarian-detector
```
); news-failure evidence should be read in that context, not as same-day confirmation.
Counter-direction events are context only — shown in the evidence table but excluded from the verdict (only events whose
```
expected_impact
```
matches the crowd's
```
expected_direction
```
count).
Proxy-based prices are noted, not hidden. When an ETF proxy was used (
```
run_context.proxy_used
```
), the report says so — tracking error, expense drag, and roll-timing differences make the reaction-direction read approximate, not exact.
Residual statistical risk under extreme correlation. The verdict's null false-CONFIRMED rate is hard-verified under i.i.d. noise (<8%) and under a realistic residual-correlation stress (AR(1) ρ=0.1, <10%). Under an intentionally extreme correlation stress (lag-1 ρ=0.3 across non-clustered event windows — roughly 10x liquid-futures empirical autocorrelation), the measured null rate rises to ~11-13%. This is a documented v1 limitation, not a silent gap — see
```
references/news-failure-patterns.md
```
for the full numbers. Users who want the stricter <10% margin even under that stress can pass
```
--drift-z 1.75
```
(at the cost of missing some genuine news-failure signals, not just noise).
Not investment advice. Research/educational purposes only.

CONFIRMED并非交易信号：它仅确认5步流程中的第2步——价格行为确认（第3步）、入场（第4步）、出场（第5步）仍需人工操作，且是建立头寸前的必要步骤。
INSUFFICIENT_EVIDENCE绝不会推进流程：可用相关事件聚类数少于
```
--min-events
```
（默认3个）、缺少detector报告、detector报告的时间戳（
```
data_date
```
）缺失/无法解析/晚于
```
--as-of
```
/早于
```
--max-detector-age-days
```
（过期）、无明确覆盖参数的NEUTRAL分类、无可用价格来源等情况，都会生成该判定结果——绝不会崩溃，绝不会基于不足数据强行生成判定。
COT数据发布延迟：COT数据在读取时已滞后3-9天（详见
```
cot-contrarian-detector
```
）；新闻反应失败的证据需结合该背景解读，不可视为当日确认信号。
反向事件仅作为参考——会在证据表格中展示，但不纳入判定结果（仅
```
expected_impact
```
与拥挤群体
```
expected_direction
```
匹配的事件才会被统计）。
替代标的价格会明确标注：当使用ETF替代标的时（
```
run_context.proxy_used
```
），报告中会明确说明——跟踪误差、费用损耗和滚动时间差异会导致反应方向的判断为近似值，而非精确值。
极端相关性下的残留统计风险：在独立同分布噪声下，判定结果的零假设假CONFIRMED率经严格验证低于8%；在现实残留相关性压力下（AR(1) ρ=0.1）低于10%。在刻意极端的相关性压力下（非聚类事件窗口的滞后1阶ρ=0.3——约为流动性期货实际自相关性的10倍），测得的零假设率会升至约11-13%。这是v1版本的已记录限制，而非隐性漏洞——详细数据请见
```
references/news-failure-patterns.md
```
。若用户希望在该压力下仍保持<10%的假阳性率，可传入
```
--drift-z 1.75
```
参数（代价是会错过部分真实的新闻反应失败信号，而非仅过滤噪声）。
不构成投资建议：仅用于研究/教育目的。

Resources

参考资源

references/news-failure-patterns.md

references/news-failure-patterns.md

Full methodology: what qualifies as a relevant event, the 4-tier source hierarchy, worked examples, the events-JSON curation guide + template, and the verdict-threshold rationale (why drift-significance, not a naive ratio; the Monte-Carlo-verified null bounds).

完整方法论：相关事件的定义、4层级来源优先级、示例、事件JSON整理指南+模板、判定阈值原理（为何采用漂移显著性而非简单比率；蒙特卡洛验证的零假设边界）。

references/price-source-map.md

references/price-source-map.md

Per-market price-source fallback chain, verified/402/0-rows status (live- probed at implementation time), ETF-proxy caveats, and markets with no viable source (documented

no_price_source

cases: VX, ZQ, HO, all agri on this key).

各市场的价格来源回退链、验证/402/0行状态（实现时实时探测）、ETF替代标的提示、以及无可用来源的市场（已记录的

no_price_source

案例：VX、ZQ、HO、所有农产品标的）。

When to Load References

何时加载参考资源

First use / explaining the methodology: Load
```
references/news-failure-patterns.md
```
Explaining why a market has no verdict (no_price_source): Load
```
references/price-source-map.md
```
Regular execution: References not needed for the CLI itself — needed for Phase 2 (events curation) and for explaining results to the user

**首次使用/解释方法论：**加载
```
references/news-failure-patterns.md
```
**解释为何某市场无判定结果（no_price_source）：**加载
```
references/price-source-map.md
```
**常规执行：**CLI本身无需参考资源——仅在第二阶段（事件整理）和向用户解释结果时需要