Claude Sonnet 5.5 Plays Factorio — 100 Green Science | Medium Effort

Claude Sonnet 5.5 attempts to produce 100 green science packs in Factorio on a fresh natural map, using medium effort. This run succeeds within the one-hour game limit. RESULT 100/100 newly machine-produced logistic science packs. Valid machine-production success. Game time: 35:12.42. Real elapsed time from agent ready to terminal: 35:15.57. Original same-session reflection: 22.19 seconds, timed separately. Recording length: 35:42.52; startup and reflection are not extra scored game time. RUN SETTINGS Model: claude-sonnet-5-5. Effort: medium. Native Claude ACP 0.85.1 with SDK 0.3.286 and bundled Claude 2.1.286. Fresh world and conversation. Natural-map seed: 19770917. Target: 100 newly machine-produced logistic science packs. Budget: 216000 elapsed ticks / 60 game minutes from the agent's ready call; success ends earlier. No shared learning notes, prior-run helpers or live operator coaching. HOW IT WORKS The agent controls the game through the formalize MCP interface and restricted Python programs, not direct mouse or keyboard gameplay. This is the native spectator recording of autonomous play. MODEL COST Final native ACP reported USD metric, after the original reflection: $5.10. This is not an actual subscription invoice or a paid API charge. Cumulative token totals are unavailable; context occupancy is not total token usage. ABOUT THE BENCHMARK One completed attempt at each of low, medium, high and xhigh is exploratory, not causal proof of effort quality or learning. All four scored attempts reached 100/100. TIMESTAMPED RECORDING SECTIONS These timestamps identify visually reviewed samples, not exact unseen transitions. Player chapter availability is not assumed. 00:00 Initial Factorio framing 00:04 Run started sample 04:17 Steam power milestone sample 14:37 Red science milestone sample 21:02 Automation milestone sample 27:21 Near logistic unlock sample (99.8% visible) 27:35 First green production sample (3 visible) 29:10 25 pack milestone sample (26 visible) 32:54 50 pack production sample 33:43 75 pack production sample 35:20 100 packs and success 35:39 Final terminal frame #Factorio #Claude #AIAgents #FactoryArena