Episode 16: Q&A β€” Session, Context & Token Basics

⏱ Est. reading time: 4 min Updated on 5/7/2026

This Episode: The 7 most common questions developers ask after first using Claude HUD.


Q1: What's the difference between a Session and a conversation?

No difference. One session = one Claude Code conversation = from launch to exit (or /clear). The Session ID is the UUID in the .jsonl filename.

/clear resets the session (clears conversation history, starts new dialogue), but the transcript file and Session ID remain. Exiting and relaunching Claude Code creates a truly new session.


Q2: What does 200K context window mean?

200K = 200,000 tokens. Tokens and characters aren't 1:1:

Language 1 Token β‰ˆ
English 0.75 words
Chinese 1-2 characters
Code Irregular, const x = 1; β‰ˆ 5-7 tokens

200K tokens β‰ˆ 150,000 English words, or 300-400K Chinese characters.


Q3: Why did context jump 40% after reading files?

The Read tool puts entire file contents into context as tool_result. 6 files (200-400 lines each) β‰ˆ 40K-60K tokens, consuming 20-30% of the 200K window.

Avoidance: Use Grep for relevant lines only, or Read file.ts:100:50 to read just 50 lines.


Q4: Which costs more β€” Input or Output tokens?

Output is 5Γ— more expensive (Opus model):

Token Type Opus Price Sonnet Price
Input $3/MTok $0.80/MTok
Output $15/MTok $4/MTok
Cache read $0.30/MTok $0.08/MTok

Saving output matters more. Caveman mode, specific instructions, "only change code, don't explain" all save output.


Q5: Is in: 318 in HUD cumulative for the whole session?

No. The identity line's (in: 318, cache: 66k) is for the current single API request.

The Session Tokens line (tokens 1.3M) is session cumulative. They complement each other.


Q6: What's the difference between Session Tokens and Identity line?

Identity Line Session Tokens Line
Scope Single API request Entire session cumulative
Data source stdin current_usage transcript sessionTokens
Purpose See how much this turn cost See total session consumption

Q7: What does 95% auto-compression lose?

All details from early conversations are lost, only summaries remain:

  • Lost: File contents, tool return results, specific code snippets, exact error messages
  • Kept: Semantic summary of conversations

Compression is irreversible. Proactively manage at 85% rather than waiting for 95%.


Next Episode: Episode 17 continues Q&A β€” advanced questions about Cache mechanisms and token optimization.