Claude V3 vs Local IG/RAG V1 初步对比

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。

# Claude V3 vs Local IG/RAG V1 初步对比

## Compared Drafts

- Claude V3:
  `drafts/008_GOOD_COMPANY_BAD_BUY_CLAUDE_100K_V3.md`
- Local IG/RAG V1:
  `drafts/009_GOOD_COMPANY_BAD_BUY_LOCAL_IG_RAG_RULE_STACK_V1.md`

## First-Screen Difference

Claude V3 wins on immediate emotional hook.

- Claude starts with a personal-loss scene: a reader did the hard thing right
  and still got a bad outcome.
- Local V1 starts with a more analytical frame: price余地 and three ledgers.

Implication:

- Claude is easier to enter.
- Local is cleaner as a framework, but may feel less sharp in the first 10
  seconds.

## Structure Difference

Local V1 wins on explicit reusable structure.

- Claude uses a natural essay route: quality, price, story, error, checklist.
- Local uses the "three ledgers" frame: cash ledger, expectation ledger, story
  ledger, error ledger.

Implication:

- Claude reads more like a polished公众号文章.
- Local is easier to turn into a repeatable column format.

## Rule Compliance

Local V1 is cleaner on compliance wording.

- Claude V3 still contains trading-flavored wording such as `买进去/买进`.
- Local V1 has no direct trading-action hits except the standard disclaimer.

Implication:

- The local rule stack did improve final-gate cleanliness.
- Claude's 10 万+ strengthening improved hook but brought back risky wording.

## AI-Flavor Residue

Both are materially cleaner than earlier drafts.

- Claude V3 has stronger sentence rhythm and less visible framework signage.
- Local V1 has no hard AI-flavor shell hits, but the repeated "账" frame makes
  its structure more visible.

Implication:

- Claude feels more human.
- Local feels more systematized.

## Evidence And Thinking

Local V1 better reflects the IG/RAG synthesis.

- It separates cash, expectations, story feedback, and error buffer.
- It incorporates the local synthesis of business machine, narrative loop, and
  payoff shape without naming thinkers.

Claude V3 is less explicit about the source logic but more fluid.

## Current Verdict

Claude V3 is better as a first public read.

Local IG/RAG V1 is better as a reusable analytical template and has cleaner
compliance.

Best next step:

- keep Claude's first-screen compression;
- keep Local's three-ledger structure and compliance cleanliness;
- produce a hybrid V2 only after the user reads both.