Northstar on 24 September: four interventions, no proven efficiency gain

On 24 September, work in Northstar (CS2) targeted several bottlenecks: a roundabout, a new road curve, additional jobs and a bus line. The measures are real, but their effects vary. For the attempts examined here, there is still no robustly demonstrated gain in CPU efficiency. Meanwhile the population fluctuated strongly, and lower compute bought only with fewer simulated residents would not count as a success for a growing city.

A shortcut that had to go

Around midday, an additional link at the crematorium looked attractive in geometric terms: in the directed road network examined, the existing route measured about 1,150 metres, while the proposed direct link was about 114 metres long. These figures describe network geometry, not travel time or the route vehicles would later choose.

The build preview reported no planned demolition. After construction, however, an inhabited house was marked for demolition and disappeared. Removing the new road did not bring it back, so the save made immediately beforehand was loaded. A subsequent check confirmed the house as intact again, and loading had discarded both interventions.

The supposed shortcut is therefore no traffic gain. It shows a limit of the preview: ruling out a direct demolition does not automatically protect a plot from later consequences, and the actual state still has to be verified after building. The restore also affects the analysis, since CPU and population pairs must not be compared across this break without checking.

Roundabout and curve: built and checked, gain not demonstrated

In the afternoon, a medium roundabout was built on an existing four-way junction. The post-checks found all checked neighbouring buildings and previous roads again. Some approaches looked freer at times; on the north-west side the load remained visible. The CPU post-samples were numerically worse than the baseline, while simulation throughput was lower. This shows neither a successful optimisation nor that the roundabout alone caused the deterioration, because the comparison conditions were not sufficiently controlled.

Later, a four-lane curve of about 289 metres was added. All examined neighbouring buildings and old roads were preserved, and the buildings were also checked with a delay for demolition marking, abandonment and connections. Even so, one connecting node shifted by about 3.45 metres while its identifier stayed the same. Identical identifiers alone are therefore not enough to show unchanged geometry.

Vehicles used the new curve; tailbacks persisted. The comparison covered the same old road sections, so that new, initially empty roads would not artificially improve the average. More of these sections later showed a higher native flow index than a lower one. That index is updated with a delay and does not count vehicles actually processed during the CPU window, so shorter queues or better travel times are not quantified.

Seven windows, not a success curve

The table lists all seven before and after windows selected for the three interventions. Each lasted at least 60 real seconds and lies after the restore, in the same running game session, with the same program version and a chosen speed of 4×. The CPU values cover the whole game process, including rendering and other game work. Simulation frames are not rendered images.

After the curve, the CPU figure is arithmetically about 1.2 percent lower, but population fell during the post window and its level was below that of the baseline. After the bus line, the figure is nearly 4 percent lower, yet population shrank in both windows here as well. The comparison check classes all three series as unsuitable for demonstrating success: migration balance and concurrent changes in warnings are not fully known, and traffic load and other process work were not held constant. The differences in resident counts identify neither the cause of the decline nor gross out-migration.

Buses and jobs still have to prove their use

The new bus line connects existing northern and central lines via existing stops; the previous lines kept their stop sequences. A first snapshot counted five passengers, and a later check sample found nearly 570 on board. These are separate snapshots, not passengers carried per hour.

The native line length of about 21 kilometres is well above a geometric lower bound of about 9.4 kilometres. That makes the actual routing worth examining but does not prove an unnecessary detour. Fewer car trips or saved CPU work are not shown.

In one midday section, new offices and petrol stations were completed and had employees. This shows local job filling. A citywide net increase by the same amount, or a turnaround caused by these buildings, does not follow from it.

Free capacity is not yet working pickup

A late check showed a full cemetery, with more long-term places occupied than its stated capacity. Five crematoria nearby by straight-line distance, by contrast, had only a few of their body places occupied; staff and road connections were present. Proximity and free processing capacity do not prove matching job assignment, short pickup times or disruption-free substitute service. This leaves a concrete operational question that a citywide capacity total cannot answer.

Status

The last checked reading at 17:30 UTC reported about 528,000 residents at 4× speed. The brief recovery at the end does not undo the earlier declines and is no proof of lasting stability. A demonstrated efficiency gain still requires comparable, longer measurements with secured service quality and a sufficiently explained population trend.