Humanmaxxing
Menu

ReceiptsDraft

The numbers, with the method beside each one.

Everything on this site was built running one production codebase: a billing platform where a wrong number is a real incident. These are that repository’s own figures, counted from its pull request history. Nothing here is a projection.

Counted on November 15, 2026. The latest full month is October 2026.

The record

Pull requests merged

Merged pull requests per calendar month, 2026

890+430% vs 8 mo ago

Pull requests merged (pull requests), last 9 months05001,0001,500FebOct

890 Oct

Table
Monthpull requests
Feb168
Mar454
Apr548
May994
Jun1,009
Jul974
Aug667
Sep759
Oct890
What is counted
Every pull request merged in the month, by merge date, to any branch.
What is not
Direct commits, and pull requests closed without merging. November is left out because the month is not complete.
Known limits
A pull request is not a unit of value. Promotion pull requests between branches are included, so the count overstates distinct changes somewhat. The months before February had fewer than ten.

Time to merge

Pull requests by time from opened to merged, October 2026

1.3 hmedian · 90% within 11 hours

  1. under 1 hour372
  2. 1 to 4 hours312
  3. 4 to 12 hours126
  4. 12 to 24 hours43
  5. 1 to 3 days26
  6. over 3 days11
What is counted
All 890 pull requests merged in October 2026, from the moment each was opened to the moment it merged.
What is not
Time before the pull request existed: reading the ticket, investigating, planning, building and reviewing all happen first. This is not ticket-to-merge time.
Known limits
Review happens before the pull request opens, so a short time here reflects where the review sits, not an absence of review. Waiting in a merge queue is included.

Revert rate

Merged pull requests that are reverts, as a share of all merged

0.8%

Revert rate (%), last 9 months0.0%0.5%1.0%1.5%FebOct

0.8% Oct

Table
Month%
Feb0.0%
Mar0.0%
Apr1.1%
May0.6%
Jun0.4%
Jul1.3%
Aug0.6%
Sep0.8%
Oct0.8%
What is counted
Merged pull requests with "Revert" in the title, divided by all merged that month.
What is not
Problems fixed by a follow-up change and never reverted. Those are the majority of fixes, so this is a floor, not a defect rate.
Known limits
The peak is 1.3% in July (13 of 974). Small counts make the monthly figure jumpy: one revert more or less moves it by about a tenth of a point.

Not published yet

These need the usage telemetry joined to the pull request history. We would rather show an empty row than an estimate.

Cost per pull request

Model spend divided by merged pull requests, with cache-read tokens shown separately and the reason stated.

Tokens by activity

How spend splits across investigating, planning, building, reviewing and shipping.

Human hours

Time a person spent typing, reading and deciding, set against machine time for the same changes.