Plan follow-through, Claude Opus 4.6
The essay reports: the proportion of concrete next-moves Claude Opus 4.6 wrote down for itself that it actually carried out within ten turns.
FindingSources & confidence: Limited evidence
Checked 2 October 2026 · AI-assisted review
Compared the reported Claude Opus 4.6 plan follow-through of 48.2% within ten turns with the local essay and available analysis/code; no reproduction.
- Local essay: i-gave-an-ai-a-civilization
Local content/posts/i-gave-an-ai-a-civilization.mdx around line 239: Claude Opus 4.6 plan follow-through of 48.2% within ten turns. The canonical URL identifies the essay; the checked artifact was the local file.
- civ6-mcp: cross-game analysis
Local file checked: qualitative analysis of early games; does not reproduce the seven published aggregate measurements.
What this check doesn’t establish
- The numeric result is a transcription of the author’s report, not independently verified experimental evidence.
- The number of eligible Claude next-moves, game allocation, exclusions and scored matches are missing.
- No games or aggregate analysis were rerun.
Confidence applies to the scope above, not every claim made by the underlying source. How confidence works
48.2%
- Measured
- plan follow through
- Of
- Claude Opus 4.6
- Sample
- not recorded
- Observed
- 2026-06-22
The essay reports 48.2% for Claude Opus 4.6. The eligible unit is a recorded next-move, not a whole game. The overall 23-game pilot does not give the Claude denominator. The available strategic-coherence scanner uses ratings and does not reproduce this percentage; an action-level scoring export is still needed.
The publication date is retained in observed because no experiment date is documented here. Confidence concerns this reported result; the underlying experiment has not been independently verified.
In the essays
- I Gave an AI a Civilization to Run. It Built a Nuke.
Read the essay
Projects
- CivBench
Produced by
Ideas
- The knowing-doing gap
Subject
Record history
Catalogue source: knowledge/findings/plan-follow-through-claude-opus-4-6.md
Generated by claude-code/claude-opus-5 on 2026-07-25