Results

Complete estimation results, validation, and population characteristics.

Baseline Study Results — Not Realtime

All figures, CI bounds, and statistics on this page are from our one-time baseline study conducted on 2026-02-17 using 16,000 sampled IDs across 7 strata. The 95% CI reflects sampling uncertainty from that study. For the live estimate with a dynamically updated CI, see the Realtime Monitor.

Point Estimate

214,129,064

95% CI Lower

212,542,476

95% CI Upper

215,716,002

Relative CI Width

1.48%

Per-Stratum Results

StratumID RangeSize (M_h)Sampled (n_h)Valid (k_h)p̂_hContribution
F11 - 10M10,000,00061147277.25%7,725,041
F210M - 50M40,000,0002,4451,95680.00%32,000,000
F350M - 100M50,000,0003,0572,75390.06%45,027,805
F4100M - 150M50,000,0003,0572,56183.77%41,887,471
F5150M - 200M50,000,0003,0572,31275.63%37,814,851
F6200M - 250M50,000,0003,0572,41879.10%39,548,577
F7250M - 261.7M11,712,00071661986.45%10,125,318
Total262,206,00016,00013,09181.82%214,129,064
Unbiasedness Check — Does the answer change with fewer samples?

What this test is asking

A good estimator should give roughly the same answer regardless of how many samples you use — it just gets noisier with fewer samples. To verify this, we split our 16,000-sample budget into smaller groups (partitions) at different sizes and ran the estimator independently on each group. If the average estimate stays near ~214M at every budget level, the estimator is unbiased.

Samples per run

How many API probes each independent run used

Independent runs

How many non-overlapping groups we could form

Average estimate

Mean of all runs — should stay near 214M

Run-to-run spread

Std dev across runs — should shrink as samples grow

Samples per runIndependent runsAverage estimateRun-to-run spreadDrift from final
1,00015214,093,447±2,846,7840.02%
2,0007214,375,432±1,782,1800.12%
4,0003214,666,338±1,488,8020.25%
8,0001214,242,3850.05%
16,000(final study)1214,129,064baseline

Result: unbiased. Even with as few as 1,000 samples per run, the average estimate stays within 0.20% of the full-budget answer. The spread (std dev) decreases as expected when more samples are used, confirming the estimator is both unbiased and statistically consistent.