X — @vicholtreman

loading…
Articles, quote tweets and standalone posts are counted separately and never averaged. The open test reads only new-format posts: posts written to the full corrected format (2026-08-14 evening onward). Impressions, profile visits and likes are deliberately not here: profile click is weighted 0.0 and a like is 0.5.

The open test — new format against old format

What is a new-format post vs an old-format (baseline) post?

Old format — the baseline / control

The 12 original standalone posts published before the evening of 2026-08-14, back when nothing was written for X on purpose. None of them used any of the gate changes: no forwardable specific picked out, no reply-inviting closer, no authority open, no credential in the post (only the pinned one carries that).

Frozen at median 118 views, 0–4 replies each. This is a control group, not a verdict — the number the new posts are measured against.

New format — the treatment

Every original standalone post from 2026-08-14 evening onward, written to the full X post gate. Four deliberate changes on top of an old post:

A forwardable specific — one copyable named number or object in the post (e.g. 220 to 380 on the seated leg press), aimed at the copy-link share signal (weight 20.0).
An authority open — the post opens on the concrete credentialed thing, not a windup.
A reply-inviting closer — at most one per batch.
The negative-signal screen — nothing that risks a mute, report, or block.

The four changes ship together and are measured together against 118 — at this sample size, no single one can be credited on its own. Read the batch median as direction only: near 250 (~2x) means something in the treatment is working; near 120 means it is flat and the constraint is the surface itself, not the writing. Source: rsh-editorial / notes-batch / x-post-gate.md, the open test locked 2026-08-14, baseline corrected 2026-08-17.

reading…

Settled new-format posts only. A post settles at 48 hours. Measured against the analytics export on 2026-08-17: a post captured on day 0 grew 533% afterward and one captured on day 1 grew 9.8%, but from day 2 on the median growth is 0.0%. Views stop moving at two days on this account. Baseline: 12 posts published before any of the new format checks existed (no selected forwardable specific, no reply-inviting closer, no authority open). Median 118. Baseline corrected 2026-08-17: 147 → 118. The old figure carried four hand-entered seed-post-* rows with no real X post id, sitting at 466, 430, 784 and 610 — far above every verifiable post, and two of them link-shares rather than standalone posts. They are now cohort unverified_seed and out of the test. The earlier note here claimed 147 was biased low because the posts were still accruing. That was wrong: across all 12 real baseline posts, total growth from capture to the 2026-08-17 export was +15 views.

New-format posts — one row each, newest first

loading…

Every post written to the full format. This is where the median comes from — and where the next batch learns which specifics earn views and replies.

Articles and their captions

loading…

The caption is the post a stranger reads in-feed. Reading views against captions is how the caption standard gets sharper. The daily-article run is a 30-day experiment (2026-07-26 to 2026-08-24), then occasional articles.

Account signals that carry weight — shares, replies, new follows

From the weekly analytics task, source x, 4-week window. Copy-link share is 20.0, reply 5.0, follow 4.0. X shows no "new follows" scalar, so follower growth is derived from the followers_total series.

Median views by type, per week

Settled items only. The pinned credential thread is excluded from the post line.

Type comparison — all settled items

loading…

Never quote a single account-level average across these rows. That is the error this page exists to prevent.