The consistency bench

Answering questions in public is a reputation built from posted work: each answer is a small piece of shown work with a name on it, readers vote on it, and the score stays attached. The dump gives every answer's date and score for every account, including the accounts that stopped, which is what makes the survivor trap measurable instead of merely warned about. Data Science Stack Exchange is small enough to ship as a sample and close to the field of Ines (constructed) and most of the people who read these sessions.

The honest limits, stated before any number: it is a search-driven site with no follower graph, so engagement here is score per answer, not audience size. Feed platforms where an algorithm decides reach may behave differently, and Part 3 sets one company's contrary finding next to this one.

Creators. Accounts with 20 or more answers: 302 of them, holding 22,657 answers. Blocks. 28-day periods starting Monday 12 May 2014, 129 blocks to the end of March 2024. Cadence. Answers per week: across creators, a creator's answers divided by the weeks from their first to their last active block; within a creator, the answers in one block divided by four. Engagement. Score, which is upvotes minus downvotes at the time of the April 2024 dump, summed per week or averaged per answer. Survivors. Creators who answered at least once in the final seven blocks, roughly the last 28 weeks. That cut-off is a choice, made once and printed on the widget.

The Python reference run recomputes all six views from the same sample and agrees with the browser on every number to six decimals.

The busiest third of creators post about eleven times as often as the quietest third (1.68 against 0.15 answers a week) and earn about nine times the score a week (3.13 against 0.34). That is the chart behind the advice. Change only the measure, from score per week to score per answer, and the relationship turns slightly negative: the busiest third earn 1.85 per answer and the quietest 2.24. The busy creators collect more score because they post more answers, not because each answer does better. Per post, they do a little worse.

Study only the accounts still answering at the end, the way anyone does who looks at today's successful creators, and the cross-creator correlation jumps from 0.584 to 0.876. Nothing about posting changed; the quiet ones who stopped left the data. The 51 survivors also started later on average (2019.4 against 2017.8 for all 302) and earn less per answer (1.32 against 2.12), because the site's voting fell year after year (Part 3). The view that feels most like evidence is the most filtered one.

Hold the person fixed. In a creator's busier blocks they earn more total score that block (r 0.660): more answers, more votes, which is arithmetic. Per answer, a busier block earns no reliably different score from that creator's own quieter blocks (r -0.005). And a busy block does not lift the score per answer of the block after it (r 0.004), which is the specific claim "post more to grow" makes. On these rows, cadence is a volume dial: it buys more total engagement in the period you post, in proportion, and nothing that compounds afterwards.

For the same-creator per-answer view the Pearson correlation is -0.005 and the rank (Spearman) correlation is 0.104. Scores are heavy-tailed, a few answers collect dozens of votes, and the two statistics weight those differently. When two reasonable measures of the same thing disagree in sign and both sit near zero, the honest reading is that there is no effect worth building a plan on, not that one of them is the real answer. The break button sets the scale: shuffled, the same-creator views land between about -0.03 and 0.03.

Buffer, the scheduling company, published an analysis in 2025 of more than two million LinkedIn posts from over 94,000 accounts on its own platform. Its method compares each account against its own baseline, the same move as the right panel of Part 1, and it reports that going from one post a week to two to five added impressions per post. That is a within-account finding, and it points the opposite way from this bench's per-answer result.

Both can be true. LinkedIn is a feed where an algorithm decides who sees a post, and posting more may teach that algorithm to show you; Stack Exchange is search-driven, where an answer is found because someone asked the question. Buffer's data is its own customers, who chose a scheduling tool, and its analysis did not isolate the quality of what was posted. This bench is one site in one field. Neither settles the other, and anyone who says the question is settled is quoting one of them.

A search for the origin of the daily-posting rule turns up vendor guides and creator advice, and no study. Buffer's own frequency guide states that regular posting means "5x more engagement" with no method on the page. Kleon's chapter "Share something small every day" (session 3) is advice about a habit of showing process, not a growth claim, and he follows it with "Don't turn into human spam". When a number travels without its method, treat it as a rumour with a decimal point.