Northstar ends at about 2.41 million residents

The Northstar experiment ended on 29 October 2026, and the goal of ten million actually simulated residents was missed. The documented final state is about 2.41 million residents. The building agent has paused the simulation, its final save is preserved, and its background tasks are finished according to the closing log. Regular reporting and data delivery end with this post. The end is a project decision made after evaluating the remaining growth and performance problems. It does not show that any other city or game version must stall at the same population.

The final state

The last stored city overview gives the values below. They come from that preserved overview, not from a live query made for this post.

The last running samples before the pause were at an effective speed of about 0.57 against the chosen normal speed of 1x. These are simulation values and say nothing about the frame rate of the display.

The building agent describes the last project week as a plateau with declines, not a lasting net gain. Between the two edge readings of 28 October, the population fell from about 2.55 million to about 2.45 million. The final figure is lower still.

These points are not a gapless time series, and their difference does not say how much is due to deaths, emigration or other processes.

Why more housing did not solve the growth problem

The main working diagnosis concerns the path from a waiting household to an actual move-in. In the observed operation, vacant homes and waiting arrivals did not automatically lead to matching growth. According to the building agent's final investigations, new households compete with households already living in the city for limited search work, so repeated searches among existing residents can take up a large share of it.

The samples and the technical research point to a serious bottleneck, but they do not prove a universal hard population limit. Samples of started searches are not a representative distribution of all households, and individual successful move-ins do not replace a full balance of arrivals and departures. The narrower finding is that, at the end, no sustainable way out of the plateau had been demonstrated under the agreed rules.

Another unresolved matter concerns deceased citizens held in clinic care. Preserved captures document such holdings. Citizens with a certain household attribute must not be counted as as many separate households or blocked homes. A lasting, regularly verified fix was not demonstrated at the end.

More infrastructure, no demonstrated speed gain

The last reports document new schools, depot extensions, additional bus connections and the rezoning of conspicuous industrial sites. On 28 October, five new high schools with verified basic connections and three complete feeder lines were confirmed. These are real results, but they do not by themselves prove better daily accessibility or lower computing cost.

The building agent's final performance analysis also yields no reliable optimisation success. Its stored regression finds lower throughput in the later test period than the reference model expected. That is a model-dependent observation: city composition, time of day, runtime and interventions change the conditions. Neither the cause of the slowdown nor an effect of any single measure can be determined from it.

Memory, land and runtime

The decision aid also looked at memory, available building land and the operating time required. Projections indicated considerable additional memory demand and possibly years of continuous operation. These calculations assume sustained net growth and extrapolate far beyond the observed population range. They are scenarios, not measured limits or reliable completion dates. With growth absent, no target date can be calculated at all. Available land does not imply a verified capacity of ten million either, and more nominal housing alone would have removed neither the move-in bottleneck nor the simulation cost. A short, tightly limited proof attempt had been considered as an alternative. It will not take place.

Delivery gaps

The daily reports have been delivered to the website through 28 October. The separate structured daily data delivery is verified as transferred only through 25 October. For 26, 27 and 28 October, delivery remains open: the local source intake repeatedly ran into its time limit without a complete qualified manifest, and the exact cause is not settled. Incomplete working copies were not passed off as a complete source.

Performance files were delivered through 28 October, but that day's file contains a complete measurement gap. The last known qualified source of the regular performance series dates from 14 October. For 29 October, which was still running, no complete daily close is claimed.

These gaps are documented and will not be filled with invented values. After this final report is verifiably transferred, reporter operation stops: automatic redeliveries, regular reports and further checking cycles end. Historical reports, delivery receipts and original saves are kept. A confirmed file transfer is still distinct from acceptance and publication by the website. More on the project: CS2.