Local IG/RAG Workflow Run

以下为模型输出的原始 Markdown，未做内容改写。用于审阅规则本身是否成立，以及后续提炼共同规则。

# Local IG/RAG Workflow Run

## Question

Was `drafts/009_GOOD_COMPANY_BAD_BUY_LOCAL_IG_RAG_RULE_STACK_V1.md` written
strictly according to the merged workflow?

## Honest Answer

The first local draft was not a full workflow artifact. It used the merged rule
stack and IG/RAG materials, but it was initially missing explicit topic scoring
and gate-by-gate run evidence.

This file records the completed workflow gates after that gap was identified.

## Step 1: Topic Gate

Topic:

> 好公司为什么也会变成坏买卖

Score using the `10 万+` topic gate:

| Dimension | Score | Reason |
|---|---:|---|
| Public relevance | 2 | Many readers encounter "good company / high price" arguments. |
| Counterintuitive angle | 2 | A good company can still be a poor asset outcome at the wrong price. |
| Reader relevance | 2 | Applies to public-company analysis, career judgment, and market discussions. |
| Time window | 1 | Evergreen topic; can be stronger if tied to a recent earnings/AI-infra debate. |
| Misconception correction | 2 | Separates company quality, expectations, and error buffer. |
| Identity signaling | 2 | Sharing signals sober judgment rather than story-chasing. |
| Save value | 2 | Ends with reusable checks. |

Total: `13/14`

Decision: pass. The topic can enter RAG/IG evidence collection.

## Step 2: Angle Lock

Common reader misconception:

> The company is excellent, so later participants will probably still be fine.

Article judgment:

> Company quality, market expectations, and error buffer are three different
> ledgers. A company can pass the quality ledger while failing the expectation
> or error-buffer ledger.

Share reason:

> The article gives readers a clean language for explaining why "good company"
> is not enough.

Save reason:

> The article leaves a reusable checklist: cash, retained capital return,
> prepaid future, narrative feedback, strongest counterargument, error payer.

## Step 3: RAG/IG Materials

Material pack:

- `runs/ai-flavor-rules-2026-06-19/local-ig-rag-good-company-writing-pack.md`

Key local inputs:

- synthesis card: business machine, narrative loop, payoff shape;
- Fisher quality-vs-price card;
- Marks good-company/bad-buy card;
- Graham margin-of-safety/error-buffer material;
- Soros/Shiller narrative and price-feedback material.

Public draft constraint:

- no master names in the article body;
- no investment advice;
- no direct trading action wording.

## Step 4: Draft

Failed first output:

- `drafts/009_GOOD_COMPANY_BAD_BUY_LOCAL_IG_RAG_RULE_STACK_V1.md`

Failure reason:

- it invented the `三张账` frame;
- the frame sounded artificial and model-made;
- it replaced direct explanation with a coined structure.

Current output:

- `drafts/010_GOOD_COMPANY_BAD_BUY_LOCAL_IG_RAG_RULE_STACK_V2.md`

Structure:

- title: `好公司最容易让人放下戒心`;
- first screen: good companies lower reader caution, making price discipline
  easier to skip;
- middle: company quality, prepaid expectations, story feedback, error room;
- ending utility: six-question checklist.

## Step 5: AI-Flavor Gate

Scan command checked current output:

- mechanical reversal;
- mechanical guide language;
- fake insight shells;
- report connectors;
- abstract consulting words;
- slogan endings.

Result for V2:

- hard AI-flavor hits: `0`.

## Step 6: Compliance Gate

Scan checked current output:

- `买入`;
- `卖出`;
- `持有`;
- `目标价`;
- `上车`;
- `该动手`;
- `什么时候买`;
- `买进去`;
- `买进`.

Result for V2:

- direct trading-action hits in article body: `0`;
- only disclaimer contains `证券买卖建议`.

## Step 7: Length And Preview Readiness

- Chinese characters excluding whitespace after metadata stripping: `3004`.
- The draft is ready to add to the local preview index as a local workflow
  experiment.

## Remaining Weakness

The article is cleaner than Claude V3 on compliance language, but may still be
less emotionally sharp in the first screen. V1 should not be used as a quality
comparison; V2 is the current valid local comparison sample.