Can we discover Bitcoin's power law origin from the data itself? We tested this by fitting P = A·(t - t₀)^n where all three parameters are optimized simultaneously. The question: does the data prefer Genesis Block (t₀=0) or some other origin?

The R² landscape reveals the answer. When we scan across different values of t₀, the function is remarkably flat around Genesis Block. Plus notice the enormous confidence band.
The unconstrained optimizer finds a mathematical optimum at t₀ = +202 days (July 2009), improving R² from 0.9613 to 0.9628 - a gain of just 0.15% (well within the error bands).
But there is an enormous cost for this insignificant improvement in R^2.
That cost is substantial. At t₀=0, the exponent is n=5.694, within 3.5% of our SSA result (β=5.9). At the unconstrained optimum t₀=+202, the exponent drops to n=5.087, now 13.8% away from SSA. We gain 0.15% in R² but lose agreement with the model-free benchmark.
This is the signature of overfitting: optimizing a narrow metric (R² alone) while degrading broader model quality.
Occam's razor applies with force here. The R² landscape shows that any t₀ within roughly ±200 days of Genesis Block produces nearly identical fits.
With no compelling preference from the data, we choose the simplest model: no offset, no extra parameter. The principle of parsimony favors t₀=0 when adding complexity buys nothing meaningful and damages agreement with independent methods.
When we constrain the optimizer to physically reasonable values (t₀ must precede price data), it converges to t₀ = -0.01 days - essentially Genesis Block itself. The exponent remains at n=5.694, matching SSA within 3.5%. Three independent methods now agree: SSA eigenmode decomposition gives β=5.9, standard power law regression gives β=5.69, and direct nonlinear fitting with optimized origin gives n=5.69 at t₀≈0.
The data offers no meaningful evidence for a shifted time origin. The simplest model consistent with all evidence is the pure power law P ∝ t^5.69 from Genesis Block, January 3, 2009. The network effects started with the network itself.