Median views per post this week: 130, up 15% on last week (113).
5 series and platforms are inside their usual range: keep the calendar as it is for those.
Each post against the usual for its own platform (its median over the last 20 posts at least two days old). 1x is a normal post.
Whether people stayed (watched more than your usual) against whether the platform spread it (more views than your usual).
Written from the 23 Oct 2026 snapshot (the first one), 23 Sep to 23 Oct on the RomanLogic account: 27 Instagram posts, 25 YouTube uploads, 20 TikToks, 25 Facebook reels, 20 Threads posts. Every figure below is computed in the sections above this one; this section is the reading of them. Numbers here are ratios to the account's own median on that platform, which is the only fair yardstick at this size. Where the evidence is one or two posts, it says so.
Three corrections to the first draft, now built into the grading above: the launch posts are promoted by the platforms whatever they contain, so they are shown but not graded; a series sitting inside the normal band has not had enough time, so it reads "Keep", not "Change"; and the thing to work on across every series is the hook, which the Hooks section now measures from Instagram (percent of the cut watched) and YouTube (engaged views as a percent of views, since Roman enabled the Analytics API on 24 Oct), with TikTok joining once Studio screenshots cover most of a rolling window.
The same piece lives three different lives. Instagram rewards Roman in character; YouTube rewards Rachel's Telegram builds and anything under a minute; TikTok rewards nothing yet. That is the single most useful thing the numbers say: stop judging a piece by one platform's result, and stop sending identical cuts everywhere expecting the same outcome.
Voice Agent Auditions, on Instagram. Four posts, median 2.7x the account median, and they also carry the account's best share rates (the Pirate and the 30-second explainer each got three sends against a reach under 300). The humor is doing what the research says humor does for B2B: a recurring character and a shared pain, not a joke for its own sake. On YouTube the same auditions are average (0.8x), except the Caveman, which broke out on both (3.3x Instagram, 11.6x YouTube). Hypothesis: the Caveman is the shortest setup of the four and the hook is in the first line of audio; worth testing against the Auctioneer, which is the newest and the weakest.
Short voice-agent explainers. One post so far ("Everything behind a voice agent in 30 seconds", 25 s): 3.1x on Instagram and 13.5x on YouTube, with the account's fastest view rate per day on Instagram. One post is an anecdote, not a pattern. The second explainer ("How to pick a good voice", 82 s) is under a day old and tracking around average on Instagram and below it on YouTube. If a third 30-second explainer lands near the first, that is a series to scale.
My Telegram Jarvis, on YouTube only. Four posts, median 4.6x the YouTube median, including the account's best upload (the Design Bot, 47 s, 25x) and the long tutorial (156 s, 8x). On Instagram the same four posts sit at 0.6x. The audience for "how I built my own bot" is on YouTube and arrives by search and the Shorts shelf, not by following. This is the builders and creators audience the strategy names as secondary, and YouTube is where it is.
Prompt Ranking: Replying to Emails (10x on YouTube, average on Instagram) is the one ranking that worked, and it is the one whose subject is a task the viewer does every day.
Facebook is a zero. 25 reels, 0 to 4 views each, 53 followers. The crosspost costs nothing, so leave the route on, but nothing should be made or adapted for Facebook until there is a reason. The research is clear that the tradespeople are there, but through Groups and recommendation threads as a person, not through a Page's reels.
TikTok is flat. Every post lands between about 90 and 220 views and stops; the only outlier is the 21 Sep intro. Likes are 0 to 6. That is the shape of a test pool that never clears: the first few hundred viewers are not watching long enough for the next batch. We cannot see percent watched or the For You split without a Studio export, so the diagnosis is a guess until that arrives. What we can see: captions are going on, keywords are in some of them, and the cuts are the same cuts as Instagram. The platform's own guidance and every trade creator with reach say unpolished, phone-shot, trade-specific; our TikToks are the polished cut.
Long teaching shorts about our own process. "How we run discovery calls", "Disco calls", "The courses I took to learn AI", "What should and shouldn't be automated": 0.2x to 0.7x on YouTube, under 1x on Instagram, the lowest group on both. Two things they share: they are long (89 s to 223 s) and they are about us selling or us learning, not about the viewer's business. The teaching pieces that did fine ("How to rank on Google", "Three questions", "Targeted leads list") are about the viewer's phone, the viewer's Google listing, the viewer's leads.
Rankings in Rachel's voice on Instagram. "Local SEO moves" and "Contractor websites" are the two worst Instagram posts (0.3x each) and average on YouTube and TikTok. The format is not the problem (the email prompt ranking worked); the subjects are vocabulary the owner does not search for. Keep the format, pick subjects the owner does with their own hands this week.
The full brief is the hooks research page; the short version, in priority order, proposals for Rachel and Roman since scripts are theirs:
What this means for the calendar: every new row carries its opening type, first line, burned line and first frame from the day it is written, so that in a quarter the types can be judged against each other within a length band. That tag schema is built (28 Oct): the How it opens section below groups hold by it, and the Monday hooks review in Telegram reads each post at 72 hours.
Threads, Facebook and YouTube are now full (24 Oct): Threads reads per-post insights, Facebook reads reel insights off all three re-minted Page tokens, and YouTube reads average percent viewed, engaged views, subscribers gained and watch minutes off the Analytics API.
Snapshot Thu 29 Oct 2026, 06:00 UK. 147 posts read, 144 matched to 70 calendar rows (live calendar). Built by tools/build_performance_dashboard.py. The long version, every chart and table: the performance report.