back

Context Preservation Systems

The Problem

Every context switch without preservation has a cost. Estimates vary, but commonly cited:

For monotropic individuals, this cost may be higher because the cognitive context is deeper.


What is "Context"?

Context is everything you're holding in working memory during a task:

Technical Context

Cognitive Context

Emotional Context

Technical context can be saved. Cognitive context is harder. Emotional context may not be recoverable.


Preservation Strategies

1. Breadcrumb Notes

Before any interruption, write one sentence:

"Debugging auth flow - token expires before refresh, checking middleware next"

Takes 10 seconds. Saves 30 minutes on return.

2. JSON State Dumps

Structured format for complex sessions:

{
  "task": "Fix payment webhook",
  "hypothesis": "Signature validation failing due to raw body parsing",
  "tried": ["body-parser", "express.raw()", "manual Buffer"],
  "next": "Check if Stripe sends different content-type in test mode",
  "files": ["webhook.ts:142", "payment.service.ts:89"],
  "blockers": ["Need Stripe CLI to test locally"],
  "confidence": "medium"
}

3. Session Logs

Automatic capture of session activity:

Enables reconstruction even without explicit notes.

4. Voice Memos

"I'm in the middle of refactoring the auth system. I just finished the login flow and I'm about to start on password reset. The tricky part will be the email verification because..."

Lower friction than typing. Can be transcribed later.


Tool Support

What Helps

ToolHow it helps
IDE session restoreReopens files, tabs, cursor positions
Git stashSaves work-in-progress without commit
tmux/screenTerminal sessions persist across disconnection
Obsidian/notesQuick capture with low friction
AI with historyCan reference earlier in conversation

What Doesn't Help

Anti-patternWhy it fails
"I'll remember"You won't
Browser tab explosionToo much context, can't find anything
Undifferentiated notesCapture without structure = hard to use
Starting freshLoses all momentum

Intervention Results

Before Context Preservation System

After Implementing Preservation

Caveat: Single case study. Individual results vary.


Implementation Recommendations

Minimum Viable Context Preservation

  1. One note per active task (even just a file name)
  2. Breadcrumb before interruption (10 seconds)
  3. Reopen yesterday's session (IDE feature)

Full Context Preservation System

  1. Structured state dumps (JSON format)
  2. Automatic session logging (commands, files, AI)
  3. Regular checkpoint commits (git, even WIP)
  4. Voice memo fallback (when typing is too much friction)
  5. Review ritual (start of day: where was I?)

Open Questions

  1. Can context preservation fully eliminate switching cost?
  2. What's the minimum viable preservation that's actually used?
  3. How does AI with long context windows change this?
  4. Is there a "context debt" that accumulates across days?

Contribute your own preservation systems or data on what works.