---
title: "Tidiane Stano (@4sapi) | Hashnode"
description: "Tidiane Stano (@4sapi) writes on Hashnode. Read their latest articles and follow their work."
url: "https://hashnode.com/@4sapi"
---

@4sapi

# Tidiane Stano

@4sapi · Joined August 2026

### [Claude Code /compact: 3 Rounds Can Lose 57% of Rules](https://4sapi.hashnode.dev/claude-code-compact-3-rounds-can-lose-57-of-rules?utm_source=hashnode&utm_medium=feed)

Three compaction rounds lost 33%–57% of tested directives. See the verbatim prompt fix, CLAUDE.md setup, and limits.

11 min read · [4sapi.hashnode.dev](https://4sapi.hashnode.dev/claude-code-compact-3-rounds-can-lose-57-of-rules?utm_source=hashnode&utm_medium=feed)

[17h ago](https://hashnode.com/posts/claude-code-compact-3-rounds-can-lose-57-of-rules/6afc1e4400df3a9d52162586)

### [Fable 5.2 Masterclass: Prompt Engineering and Production Workflow for Anthropic’s New Flagship](https://4sapi.hashnode.dev/fable-5-2-masterclass-prompt-engineering-and-production-workflow-for-anthropic-s-new-flagship?utm_source=hashnode&utm_medium=feed)

Fable 5.2 scores 82.5% on Benchmark-Agent. Learn effort tuning, prompt caching, and multi-model agent workflows.

10 min read · [4sapi.hashnode.dev](https://4sapi.hashnode.dev/fable-5-2-masterclass-prompt-engineering-and-production-workflow-for-anthropic-s-new-flagship?utm_source=hashnode&utm_medium=feed)

[17h ago](https://hashnode.com/posts/fable-5-2-masterclass-prompt-engineering-and-production-workflow-for-anthropic-s-new-flagship/6afc1d0f5e9a800b0ec95925)

### [Why Agent Demos Look Impressive and Still Fail in Production](https://4sapi.hashnode.dev/why-agent-demos-look-impressive-and-still-fail-in-production?utm_source=hashnode&utm_medium=feed)

Learn why agent benchmarks miss real-world failures and how digital twins and confidence intervals improve reliability.

13 min read · [4sapi.hashnode.dev](https://4sapi.hashnode.dev/why-agent-demos-look-impressive-and-still-fail-in-production?utm_source=hashnode&utm_medium=feed)

[1d ago](https://hashnode.com/posts/why-agent-demos-look-impressive-and-still-fail-in-production/6afacb56b7205a472929064e)

### [Harness Engineering: The Engineering Layer That Makes Models Work Reliably](https://4sapi.hashnode.dev/harness-engineering-the-engineering-layer-that-makes-models-work-reliably-mwj1s8jx?utm_source=hashnode&utm_medium=feed)

See how Harness Engineering cut agent runtime 81.89% and loop rounds 59.6%, with 100% completion in 9 runs.

12 min read · [4sapi.hashnode.dev](https://4sapi.hashnode.dev/harness-engineering-the-engineering-layer-that-makes-models-work-reliably-mwj1s8jx?utm_source=hashnode&utm_medium=feed)

[1d ago](https://hashnode.com/posts/harness-engineering-the-engineering-layer-that-makes-models-work-reliably-mwj1s8jx/6afac993a8f5ce137b97f848)

### [Atomic Judgment Over Free Generation: A Practical Evaluation of Jev SDK and Typed LLM Decision Patterns](https://4sapi.hashnode.dev/atomic-judgment-over-free-generation-a-practical-evaluation-of-jev-sdk-and-typed-llm-decision-patterns?utm_source=hashnode&utm_medium=feed)

Jev SDK: Replace JSON Prompts with Typed LLM Judgments

14 min read · [4sapi.hashnode.dev](https://4sapi.hashnode.dev/atomic-judgment-over-free-generation-a-practical-evaluation-of-jev-sdk-and-typed-llm-decision-patterns?utm_source=hashnode&utm_medium=feed)

[Nov 7](https://hashnode.com/posts/atomic-judgment-over-free-generation-a-practical-evaluation-of-jev-sdk-and-typed-llm-decision-patterns/6aeeeb442927e14d1d440d9b)
