Skip to content

Instantly share code, notes, and snippets.

@sebas5384
Last active April 28, 2026 15:49
Show Gist options
  • Select an option

  • Save sebas5384/ca9f3dec167040ae1d89011851d6762e to your computer and use it in GitHub Desktop.

Select an option

Save sebas5384/ca9f3dec167040ae1d89011851d6762e to your computer and use it in GitHub Desktop.
SocratiCode vs. No SocratiCode

Prompt

what are the impacts if we change the firebase authentication to use a different one. use socraticode to explore the impacts

Scoring Summary

Factor With SocratiCode Without SocratiCode
Accuracy / Completeness 5 4
Depth of Analysis 5 3
Actionability 5 4
Structure / Clarity 5 4
Risk Assessment 5 4
Total 25 / 25 19 / 25

Cost Analysis

Token usage extracted from the actual Claude Code session logs. Both sessions used claude-sonnet-4-6 throughout — no model difference.

Pricing used: $3.00 / MTok input · $15.00 / MTok output · $0.30 / MTok cache read · $3.75 / MTok cache write

Metric With SocratiCode Without SocratiCode
Input tokens 52 33
Output tokens 14,304 9,299
Cache read tokens 476,142 588,389
Cache write tokens 209,591 82,469
Total tokens processed 700,089 680,190
Assistant turns 22 25
Estimated cost ~$1.14 ~$0.63

What drives the cost difference

Total tokens processed are nearly identical (~700K each), but SocratiCode created 2.5× more cache (209K vs 82K tokens written). Cache creation is the most expensive operation ($3.75/MTok vs $0.30/MTok for reads), which accounts for most of the gap.

The reason: SocratiCode's semantic search queries build up new context artifacts at each step. The non-SocratiCode approach relied on 14 Read + 7 Bash calls that hit the existing prompt cache more efficiently — cheaper per token, but returned less targeted information.

SocratiCode also produced 54% more output tokens (14,304 vs 9,299) — the richer analysis is directly reflected in the output cost.

Cost-effectiveness verdict

SocratiCode cost ~ 83% more for this task (~ $0.51 premium). Whether that is cost-effective depends on context:

  • For a one-off exploratory analysis feeding into sprint planning: yes — the deeper insight (surfaced hidden costs, test infra, mobile coordination risk) is worth far more than $0.51 of scope risk avoided.
  • For high-frequency, lightweight queries: no — the cache-creation overhead adds up without proportional gain.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment