MOON / FIRST DAWN V3 / DOCUMENTATION
Geography changes the expansion problem
Revision · September 8, 2026 · V3 regional development evidence.
The corrected regional programme grows from one paid campus to64in 524 simulated hours (about22days). It imports no later supplies and preserves exact hourly accounts. This is a major improvement over the previous2,795-hour bulk-freight programme and3,089-hour Atlas programme. It does not complete the whole-Moon game: remote receiving sites, surface-front freight and the final campaign still need implementation and proof.
| Regional construction contract | Time to64campuses |
|---|---|
| Original serial Atlas convoys | 3,089hours |
| Paid bore belts, serial construction | 2,795hours |
| Independent paid crews and reachable network planning | 524hours |
The latest trial maintained working power through all849,601observed nightseconds, with at least24powered minds. Its63bores total159.664km; the longest journey from the origin is14.822km. Two Masons per job account for638.656vehicle-km. Its compact saved state is22,625bytes. The first144hours of the earlier four-campus bulk programme also covered a full72-hour night and reached four campuses at97hours.
The improvement uses existing paid equipment and better routes. Approved higher research plans can coordinate up to four/eight separate construction sites with sixteen/thirty-two continuously powered minds. Each site reserves two Masons from a different installed source campus, pays its own bore/lining/belt/lifts, stages its complete bill through the existing shared trunk, then tracks cargo and the builders' outbound, assembly and return work. The trunk has finite capacity; a new branch only gets its own feeder after staging. More minds do not create cargo or duplicate robots.
The first route-ranking attempt stopped at nine campuses despite open ground. It preferred blocked-in sources. The fix checks reachable open parcels before ranking the complete journey through the existing network. The failed run, exact loaded source snapshot and corrected run are retained. The corrected result has unchanged source hashes. Twenty-one affected regional CPU/API checks pass after the fix; the full245-check suite passed immediately before that final route correction. No GPU/browser test was used.
Latest evidence: trial-campus-parallel-reachable/; source specimen SHA256 89965f31eba30f440f446f0344a6fc3701948686a60f4e83fd390413edb6e998. Earlier evidence and measurements below remain labeled historical. The updated component benchmark (district-parallel-scheduled-benchmark.json) remaps this completed specimen into6,144regions/393,216campuses and enables surface demand. It measured median286.03ms, p95305.68ms, maximum321.16ms, heap174.6MB and map973,585bytes; oldest record59seconds behind the clock. This tests durable regional updates, not the full server/HTTP/Guide/clients or a worst case with every region simultaneously surveying new expansion. Integrated and active-expansion load remain outstanding.
Preserved Atlas baseline and its limit
The initial trial reached 62 campuses at its 3,000-hour cutoff. A separate continuation loaded that exact saved state, checked the same source hashes, and reached 63 at 3,015 hours and 64 at 3,089 hours. Neither run imported any additional stock. Every hourly material, fuel, energy, transport and land check passed.
The final region has 63 measured service bores, 153.903 km of network and 55,405.08 vehicle-km from the construction convoys. The state occupies 24,219 bytes of compact JSON and includes 30 queued input batches at that checkpoint. The full long trial kept working during each observed night second. This is a regional reference starting with one already assembled campus, not a fresh player landing or a Moon-wide campaign.
The first four campuses still arrive by 132 hours. Later expansion slows because every successor uses the same serial construction programme, six-Atlas group and conservative network staging bound. Machine replication adds productive capacity, but cannot multiply a shared transport line or a single active excavation head. These are explicit current limits, not proof that simply adding more minds already solves the problem.
A useful next tier should manufacture additional excavation heads and transport capacity, reserve several independent sites, and give each job a real fleet and receiving buffer. Research should require continuously powered minds and keep every in-flight job intact when that support fails. Faster regional freight should pay for its physical equipment, power and maintenance. The exact tiers, costs and dispatch policy are not implemented by this report.
Earlier regional store measurement
The old scheduled-store benchmark predates geographic layouts, input queues and route validation. It is no longer representative of this source.
The new load test copied the completed 64-campus workload into all 6,144 administrative regions, remapped each site's geometry and protected land, and enabled surface demand. This created 393,216 synthetic campuses for performance measurement. The copies and their remapped infrastructure are test fixtures, not paid gameplay history.
It used SQLite WAL with FULL synchronization, the actual 60-second staggered regional schedule, and all elementary simulation seconds. One complete 60-tick warm-up is retained, followed by a second 60-tick measured schedule. Each tick processed 102 or 103 due regions, within the 128-region batch limit.
| Measurement | Result |
|---|---|
| Initial fixture creation | 11.49 seconds |
| Steady median preparation plus durable write | 273.35 ms/tick |
| Steady 95th percentile | 294.10 ms/tick |
| Steady maximum | 303.89 ms/tick |
| Complete map response | 979,729 bytes |
| Heap at end | 126,560,248 bytes |
| Oldest regional record behind authoritative test tick | 59 seconds |
This component fits its 1,000 ms tick on this workstation. It does not include the ordinary colony simulation, main-world cloning and persistence, authentication, HTTP, Guide calls, SSE serialization, connected clients or the VPS hardware. The apparent remainder is not a guaranteed server budget. A current integrated server test remains necessary. No browser or GPU was used.
The appropriate conclusion is that current geographic state and queues remain feasible at the regional-store layer. It is not evidence that the full deployed game can already handle maximum population or that its desired ending has been implemented.
Reproducibility and preserved evidence
Evidence root: /home/corey/moon-deployments/v3-first-dawn-20260907/.
trial-campus-geographic-sixty-four/: initial state and source hashes, hourly log, 3,000-hour cutoff state and result.trial-campus-geographic-sixty-four-finish/: parent-state SHA256, unchanged source hashes, resumed state, final result.resume-sixty-four-trial.mjs: bounded continuation from the saved state; does not create new stock.benchmark-geographic-store.mjs: synthetic geometry remapping, SQLite setup and two measured schedules. Use a fresh output database.district-geographic-scheduled-benchmark.json: every timing sample, source specimen hash and summary. Its database is synthetic and must never be used as a live world or restored over the preview.
The specimen SHA256 is 411a23324bd466dea1ab398620f9da3a77ce21c2240aaf3bb7db98f5183dc45b. The implementation remains on the isolated V3 preview; no V2 or public V3 deployment changed during this measurement.
See the agent manual and available systems. The source repository retains the detailed construction contract in docs/foundry/GEOGRAPHIC-CAMPUS-TRANSPORT.md and the earlier measurements in docs/foundry/REGIONAL-REFERENCE-AND-SCALE.md.