From 2c455d68674d0f76c1a657272e63eec700d79b21 Mon Sep 17 00:00:00 2001
From: Florian BRUNIAUX <florian@bruniaux.com>
Date: Tue, 17 Mar 2026 10:22:25 +0100
Subject: [PATCH] =?UTF-8?q?fix(whitepapers):=20v2=20corrections=20relecteu?=
 =?UTF-8?q?rs=20=E2=80=94=20WP00/02/05/08/10=20FR+EN?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

8 phases of corrections based on reviewer feedback (Edouard, Mat,
Nicolas, Marc, Anthony, Emmanuel) across 10 whitepapers (WP00-WP10).

Key factual fixes:
- npm "deprecated" → "not recommended", native installer added (WP00)
- Context window 200K vs 1M clarified with agent teams distinction (WP00)
- Claude Max cost ratio corrected to range 1.5-3% FR / 1-1.6% EN (WP10)
- Sub-agents vs agent teams: difference = P2P messaging + coordination
  tools, NOT context isolation (WP08 EN)

Content improvements:
- ~200+ accent restorations in WP08 FR (977 lines)
- Anglicisms corrected in WP02 FR (on-demand, per-project, Memories)
- WP10 executive summary added with 5 key metrics
- WP10 learning curve duplication removed, BCG source improved
- WP10 "Beyond engineering teams" expanded with 3 concrete use cases
- WP08 tmux noted as external tool; meta-sentence removed; solo agent
  clarified as full autonomous agent, not chatbot
- WP05 Turnkey Quickstart explained; "Multiple Claude instances"
  rewritten as complete sentence

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 CHANGELOG.md | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9f71db7..01cca6c 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,6 +6,16 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 
 ## [Unreleased]
 
+- **Whitepapers v2 — Reviewer corrections** (6 relecteurs: Edouard, Mat, Nicolas, Marc, Anthony, Emmanuel): 8-phase correction plan applied across 10 whitepapers (WP00–WP10) FR+EN.
+  - **Phase 1 — Factual errors**: npm "deprecated" → "not recommended" (WP00), native installer added as primary install method (WP00 EN), context window 200K vs 1M clarified (WP00 FR+EN), Claude Max ratio corrected to range 1.5-3% FR / 1-1.6% EN (WP10), sub-agents vs agent teams difference correctly stated as P2P messaging + coordination tools, not context isolation (WP08 EN)
+  - **Phase 2 — Accents + articles + typos**: ~200+ accent corrections across WP08 FR (977 lines); anglicisms corrected in WP02 FR (on-demand, per-project, Memories, Feature opt-in); table row reformulated in WP00 FR; accents restored in WP02 FR (~10 passages)
+  - **Phase 3 — Malformed sentences**: "chez dans une fintech" fixed, meta-sentence "This is the single definition..." removed (WP08 FR+EN); "Multiple Claude instances..." rewritten as complete sentence (WP05 FR+EN); Turnkey Quickstart explained (WP05 FR+EN)
+  - **Phase 4 — Anglicisms + clarifications**: tmux noted as external tool, not built-in (WP08 FR+EN); "OUI →" residue fixed in WP05 EN; user-level/project-level translated (WP05 FR)
+  - **Phase 6 — WP08 structural corrections**: "~30K lines" clarified as "~30K lines ≈ 1M tokens"; continuous merge definition improved; note added that solo Claude Code agent is a full autonomous agent, not a chatbot
+  - **Phase 7 — WP10 restructuration**: executive summary added (5 key metrics); learning curve duplication removed; BCG 2025 source improved; pricing note "March 2026" added; "5-10 people in 2020" reformulated without unsourced claim; "Beyond engineering teams" section expanded with 3 concrete use cases; CFO/DAF and change management placeholder callouts added
+
+- **Resource Evaluation #078 + integration** (`guide/ultimate-guide.md`): HumanLayer — "Skill Issue: Harness Engineering for Coding Agents" (March 12, 2026). Score 3/5. "Harness Engineer" concept already covered in `guide/roles/ai-roles.md §10` (Martin Fowler source); ETH Zürich study already integrated. Two genuine gaps filled: (1) **Silent Success hook pattern** — asymmetric hook behavior (exit 0 = silence, exit 2 = re-engage agent) named and documented with example before §7.3; (2) **CLI vs MCP training-data argument** — familiar CLIs (git/grep/jq) are already in model weights, CLAUDE.md usage examples often beat a verbose MCP server for standard tools; added after MCP Tool Search section.
+
 ## [3.36.0] - 2026-03-17
 
 ### Documentation