Step 2: Data validation

A strict silent-storage mode. Claude reads sc-step1-salary-research.md written by Step 1 and processes each country’s data automatically, no pasting required. If the file is missing, Claude stops and reports it rather than waiting for manual input.

Flow 

flowchart TD
  Start([Step 2 begins]) --> FileCheck{sc-step1-salary-research.md\nexists?}
  FileCheck -->|no| Error[Stop, report missing file\nAsk user to rerun Step 1]
  FileCheck -->|yes| ReadFile[Read file, process\neach country block in order]
  ReadFile --> MultiCheck{Multiple countries\nor none in block?}
  MultiCheck -->|yes| Err1[Skip, record issue\nNothing stored]
  Err1 --> ReadFile
  MultiCheck -->|no| FieldCheck{Both tiers and\nsources present?}
  FieldCheck -->|no| Err2[Skip, record missing fields\nNothing stored]
  Err2 --> ReadFile
  FieldCheck -->|yes| DupCheck{Country already\nstored?}
  DupCheck -->|yes| Skip[Skip, keep original\nRecord as duplicate]
  DupCheck -->|no| Store[Store verbatim]
  Skip --> ReadFile
  Store --> ReadFile
  ReadFile --> Done{File\nexhausted?}
  Done -->|no| ReadFile
  Done -->|yes| Report[Reply with one consolidated report:\nstored count + all skips by reason]
  Report --> End([Continue automatically\ninto Step 3])

What it reads

  • sc-step1-salary-research.md: written by Step 1 sub-agents

Rules

This runs fully automated, with no pause between countries; bad items are skipped and recorded, never halted on.

Situation Claude’s response
File not found Stops, reports missing file, asks user to ensure Step 1 completed
Block contains multiple countries Skips it and records the issue, nothing stored
Block contains no recognizable country Skips it and records the issue, nothing stored
Missing tier or sources Skips it and records what’s missing, nothing stored
Country already stored Skips it, keeps original, records it as a duplicate
Valid, complete data for one country Stored silently

Claude preserves all values, wording, and formatting exactly as provided. It does not correct, improve, or reinterpret the data.

Required data per country

Each block must include both tiers:

  • Mid-size / Mainstream Local-Market tier: Low, Realistic, Strong figures, city used if applicable, sources with dates
  • Premium / International / Remote-first tier: Low, Realistic, Strong figures, city used if applicable, sources with dates

If either tier or its sources are missing, Claude skips that country and records what is missing; nothing is stored, and the run continues to the next country.

Output

Stored salary data is written to sc-step2-salary-data.md in the workspace after all blocks are processed. Step 3 and Step 4 read from this file directly.

Stop condition

There is no interactive checkpoint in this step. Once all countries from sc-step1-salary-research.md have been processed, Claude writes all successfully stored countries to sc-step2-salary-data.md, replies with one consolidated report (stored count, plus every skip grouped by reason: malformed, missing tier or sources, duplicate), then continues automatically into Step 3 within the same response, without waiting for a new message.

Claude Compass — MIT License