Stats

The whole operation, quantified. Every number below is recomputed from the site's own content collections at build time. Nothing here is hand-typed, so it can't drift stale.

"Pages built" counts the reader-facing HTML routes. It excludes feeds, the sitemap, and llms.txt. Curious how the guardrails behind these numbers work? See /colophon.

The loop, measured

The counters above prove activity. The site's actual thesis is different: the system should hand back more time than it costs to run. This is the honest state of that ledger. Unlike everything above, it is hand-curated, so it carries its own date and says plainly where a number is not published yet.

Last reviewed 2026-07. How this content gets made and reviewed is on the about page.

How time reclaimed gets measured

The row above is the one that matters, and it is the one still empty. A time-saved number without a method behind it is marketing, so the method goes up first and the number has to arrive under it. Here is exactly what will be counted, what will not, and what has to be true before any figure appears on this page.

net time reclaimed = time the chore used to cost
                   - steering time
                   - review time
                   - maintenance time
                   - incident recovery time

What each term means

• Time the chore used to cost is the counterfactual baseline: what the same work took me by hand, timed before the automation existed, multiplied by how often it actually ran that month. An automation with no timed manual baseline contributes nothing to the total, however useful it feels.
• Steering is every minute spent prompting, choosing, and correcting it that month, including the sessions where it went wrong.
• Review is reading and verifying what it produced. Output I never read counts as zero time avoided, no matter how much of it there was.
• Maintenance is fixing, updating, and re-tuning it: the same upkeep term that sits in the denominator when the loop scores what to build next.
• Incident recovery is cleaning up after it breaks something, including the time it took to notice.

The rules that keep it honest

• Unused output is worth zero. A generated page nobody opens, a feed nobody reads, a report nobody acts on: no avoided time, by definition.
• A range, never a single number. Baselines are timed samples and usage swings week to week, so the honest output is a low and a high bound with the assumptions stated, not a confident point estimate.
• Dated to the month it covers. Every figure carries its month, and old months stay up next to new ones so a bad month cannot quietly disappear.
• Computed on a monthly cadence from entries logged as the work happens, never reconstructed from memory at publication time.
• Hours only. The money side of the loop's scoring stays private. This page publishes time, or it publishes nothing.

The publication rule

A number appears here only with its method, its month, its range, and the list of automations it covers. The threshold that releases the first one: four consecutive weeks of real logged entries covering the live automations, both the manual baselines and the steering, review, maintenance, and recovery time against them. Until that log exists, the row above keeps saying so. Publishing a number I cannot show the work for would cost more credibility than the number could ever buy.