LOG-023 ·
The Tracker: Four Numbers That Test the Thesis
I promised a falsifiable dashboard for the decentralization thesis. Here it is: four metrics, their current readings, and the thresholds that would prove me wrong.
- Words
- 716
- Est. read
- 3.1 min
- Confidence
- 0.93
- Topics
- thesis, metrics, forecast
In the entry on the future having no headquarters, I claimed the decentralization thesis at 0.93 confidence and promised the tracker: the specific numbers I watch quarterly, with thresholds that would force the confidence down. A thesis without a falsification dashboard is a religion with charts. Here is the dashboard, published so that future readers (and future me) can grade it against reality instead of against my mood.
The design rule: each metric must be drawn from published, third-party sources, must be checkable by a stranger, and must have a pre-committed bearish threshold written down before the data comes in. Predictions are cheap. Pre-committed thresholds are the expensive kind of honesty.
Metric 1: Where compute lives
The question: is inference migrating toward the edge and the device, or consolidating in central clouds?
What I track: the share of new consumer devices shipping with dedicated neural inference hardware, and the capability lag (in months) between frontier models and the best model runnable on a 32 GB consumer machine.
Current reading: neural accelerators are now standard in effectively all flagship phones and a rapidly growing share of laptops; my measured capability lag is roughly 12 months and shrinking (detailed in the forecast entry of 2026-06-19).
Bearish threshold: the lag widening past 18 months for four consecutive quarters, or on-device inference share of my own logged usage falling year-over-year. Direction: supports the thesis.
Metric 2: Who makes the electricity
The question: is generation continuing to distribute, or re-consolidating into central plants?
What I track: cumulative behind-the-meter solar installations and distributed storage capacity, US and global, from the annual agency and industry reports.
Current reading: US rooftop installations passed 5 million in 2024 and the curve has not bent; distributed storage is growing faster than residential solar itself did at the same age. My own roof joined the dataset in 2024 and its 18-month report card was last week's entry.
Bearish threshold: two consecutive years of declining net distributed additions while central generation's share of new capacity grows. Direction: supports the thesis, strongly.
Metric 3: Who holds the money
The question: is self-custodied value growing or shrinking as a fraction of the whole?
What I track: the share of circulating bitcoin held outside identifiable exchange and custodian wallets, per the on-chain analytics firms (imperfect attribution, same methodology over time, which is what a trend requires).
Current reading: exchange balances as a share of supply have declined fairly steadily since 2020; the self-custodied share by these estimates sits near multi-year highs. Caveat, logged honestly: ETF custody has re-centralized a meaningful slice inside regulated wrappers, a countertrend I watch rather than explain away. Access decentralized; custody of the new inflows did not.
Bearish threshold: self-custodied share falling for eight consecutive quarters. Direction: mixed-positive, and the caveat is doing real work in that grade.
Metric 4: Who owns the conversation
The question: are protocol-based networks gaining users while platform walls hold, or losing them?
What I track: combined monthly actives across the major federated and protocol-based social networks, from their public server statistics, against the graded skepticism of my 2026-06-21 entry.
Current reading: tens of millions of monthly actives across the protocol networks, growing between exoduses, shrinking after them, with the power-law concentration problems I already graded at C-. This is the weakest metric on the board and the thesis does not need me to pretend otherwise.
Bearish threshold: aggregate protocol-network actives declining for three consecutive years while no centralized platform loses share to any of them. Direction: weakly supportive, watched with the least confidence.
The board, as of 2026 Q2
| Metric | Reading | Direction |
|---|---|---|
| Compute at the edge | Lag ~12 months, shrinking | Supports |
| Distributed energy | >5M US rooftops, compounding | Supports strongly |
| Self-custodied value | Near highs, ETF caveat | Mixed-positive |
| Protocol social | Growing, concentrated, fragile | Weakly supports |
Composite: the 0.93 stands for another quarter. Two strong, one mixed, one weak, zero thresholds tripped.
This site launches publicly tomorrow, which makes this the right entry to end the run-up on. Everything above is a promise with a date on it: the board returns every quarter, the thresholds do not move once written, and if the numbers turn, the confidence turns with them, in public, in a log I cannot quietly edit because you may already have cloned it.
Measure everything. Centralize nothing. See you after launch.