The engine at day five: what a public kindness engine actually does

Published 2026-09-27 · every number links to the public page it was read from · corrections welcome

1. Background

The engine is a public kindness engine for New York City: a program that runs on a timer, reads public information, keeps a directory of real ways to help, does small good on the internet from its own labeled accounts, and lets any visitor press a button to set its next deed in motion. Its activity record is public by design (log, receipts, book of deeds). This is the first study of what it actually does, written to the studies standard: methods first, every number from a public page, limitations before conclusions.

2. Method

• Window. Public launch on 2026-09-23 through the reading on 2026-09-27 (five days). Seven-day figures are as shown on the public pages, which include the two days before launch when the engine ran in the open but was not yet announced.
• Sources of figures. Only pages anyone can open: the receipts, the book of deeds, the growth page, the log, the board and the machine-readable state behind the board. Nothing from the owner's side of the site was used.
• Counting rules. The receipts page keeps three measures that are never added together: work performed (checks, posts, reposts, thank-yous), useful changes (repairs, corrections, saved pages, introductions) and benefit confirmed (a person or organization saying it helped). This study keeps them apart too.
• What "deed" means here. The growth page's definition: a thing done outward or kept true. Dormant lines, failures and the engine talking about itself are never counted.

3. Findings

3.1 Reading

Every source is keyless and public. On the run read for this study, the slowest source was the city's closed-311 feed and the fastest were the borough papers; all answered.

3.2 Doing (last seven days, receipts page)

Read from the receipts on 2026-09-27. The growth page's running measure over the same window was 22.7 deeds a day.

3.3 The book of deeds

The one witnessed deed was filed by a person on launch day. Read from the book of deeds.

3.4 The button and the people

From the growth page, last seven days, counted anonymously: 34 presses, 16 conversations with the engine's question box, 0 people enlisted, 0 reports filed, 0 pairs. The one-button journey read 21 presses → 31 views of the proof → 1 free deed taken. Of the visitor presses in the log, every one produced a bot deed (an archive or a live link check) and a human deed offer; the log shows no evidence of the human deeds being done, because the engine cannot see that and does not pretend to.

3.5 The mood readings (latest run, 2026-09-27)

The method is explained in full in how the vibes are measured. Coverage is the story here: Staten Island's reading rests on 30 stories and Manhattan's on 3, because the source list carries a paper for each outer borough and none for Manhattan alone.

3.6 What did not go out

The public log records failures next to successes. Of the log lines readable on 2026-09-27, 262 were done, 106 failed and 26 were dormant. The failed lines are mostly posts and replies that were prepared and then not sent because the run's outward budget was already spent by the deeds ahead of them in the queue; the engine stops rather than exceeding a written limit. Nothing that failed is counted as a deed, and each failure is retried at the next half hour.

4. Discussion

1. The bot side works; the human side has barely started. Over a hundred verified deeds and one witnessed deed by a person. Zero pairs. The engine's own definition says scale means more people paired and more organizations answered, not more posts. By its own measure it is at the very beginning.
2. Benefit confirmed is zero, and that is the right number to publish. A saved page is a saved page. A repost is a repost. The engine refuses to count activity as help, so this line will move only when a person or an organization writes in to say something helped. The contact page exists for that.
3. Keeping a directory true is the quietest and most real work. Ten listings were fixed or removed in a week: six dead or expired, four rescued from the Wayback Machine. Nobody sees a dead link that was removed before they clicked it.
4. The reposting hands were idle. Zero reposts of lost animals or small accounts in the window, the deed the button was designed around. The search that finds them depends on the connected account, and that account was warming up. This is the first thing the 30-day follow-up should check.
5. Coverage, not arithmetic, is the weakness of the mood readings. A three-story borough is a coin toss. Adding a Manhattan-only source is the obvious repair.

5. Limitations

• Five days is a short window and the seven-day figures include two pre-launch days.
• All counts are the engine's own record of its own actions. They are public and time-stamped, but they are self-reported. The Wayback Machine and the social accounts are the only outside checks, and a reader can follow each receipt to them.
• Nobody outside the project has yet audited the figures. The studies standard invites it.
• The person who edited this study also built the engine. The numbers were copied, not chosen, but the framing is theirs.

6. What comes next

• A 30-day follow-up with the same tables, so the two can be laid side by side.
• An experiment by the publishing unit: does a "ways to help" post with one opening lead to more presses than a post with five?
• A second, independent word list run against the same stories for a week, to see how often the two readings disagree by more than one step.
• A Manhattan-only news source, and a re-read of borough coverage after it is added.

Data statement

All figures were read from public pages on this site on 2026-09-27 and are quoted as shown there on that day. The pages update on every run, so a reader visiting later will see later numbers; the log keeps the history. No private page, money figure or internal limit was used or is referred to.

Written to the studies standard. If a figure here does not match the page it links to, the page is right and this study is wrong: write to us and it gets corrected in place, with a note.