If you could predict reception from significance, there would be no reason to measure reception. You would score the move, infer the reaction, and save yourself the hard half of the problem. So we checked whether you can.
Across 6,569 deduplicated moves from 648 companies, the Pearson correlation between significance score and reaction count is +0.233. Positive, real, and weak — it accounts for roughly 5% of the variance in reception. The other 95% is something else.
Does the size of a launch predict whether it lands?
Only weakly. At r = 0.233, knowing how significant a move is tells you very little about how much public reaction it will earn. Big moves do land somewhat more often than small ones, but the relationship is nowhere near strong enough to substitute one measurement for the other. In practice this means a company can ship something genuinely major to near-silence, and ship something minor into a large reaction, and neither outcome is anomalous.
That is the empirical case for treating 'what shipped' and 'did it land' as two separate axes rather than one. They are not measuring the same underlying thing.
Why is the correlation so low?
Because significance is a property of the move and reception is a property of the audience. A significance score reads the artefact: how much changed, what kind of change, how it compares to that company's baseline. None of that knows whether the people who would care were paying attention that week, whether a louder story buried it, whether the framing landed, or whether the market had already moved on.
Those audience-side factors dominate. Which is inconvenient if you want a single number, and useful if you want to know where the openings are.
Why is the correlation weaker among moves that already landed?
Restricted to moves with three or more reactions, the same correlation drops to r = 0.095 across 280 moves. That is not a contradiction — it is restriction of range, a standard statistical artefact. Filtering to moves that already landed truncates both variables, and truncating the range of a correlation attenuates it toward zero.
It is worth naming because it is an easy way to publish a wrong number. Analysing only the top-ranked moves in each sector is a natural thing to do — they are the ones already surfaced in a report — and it would have produced a near-zero correlation and a much more dramatic headline. The full-population figure of 0.233 is the honest one, and it is the smaller claim.
What should you do with a weak correlation?
Stop inferring impact from size, and start looking at the residual. The interesting moves are the ones where the two axes disagree: high significance with no reception is an opening, and low significance with heavy reception tells you something about what that market currently cares about.
If the correlation were 0.9, a competitive intelligence tool could just rank by significance and be done. At 0.25 it cannot, and any product that ranks on a single blended score is quietly averaging away the exact signal you needed.
How was this measured?
Pearson correlation between the deterministic significance score and the count of coupled public reactions, computed across all 6,569 deduplicated moves — no filtering, no top-N truncation, no outlier removal. Reactions come from Hacker News, Bluesky, GitHub and news, bound to a specific dated move rather than counted loosely against a company name. Mean score 0.331, mean reactions 0.40.
Caveats: reaction counts are heavily zero-inflated (most moves draw nothing), so Pearson is a rough instrument here and the true relationship is unlikely to be linear. The direction and the weakness are robust; the exact coefficient should be read as an approximation. And as always this is public reaction on the surfaces we read, which under-represents markets that discuss themselves privately.
Figures are deduplicated. Our own tracking table held several records for a few companies (Linear had seven), and each record independently re-detected the same changelog entry and re-coupled the same public reactions to its own copy — inflating one company's reaction count by 268% before we caught it. A move is now identified by title, category and date, and reactions are counted by distinct source URL, so the same event and the same reaction each count once.
Revised after a coupling audit. Content matching bound reactions to launches without a causality check, so an old forum thread naming a company could attach to a recent, unrelated release — in the worst case a 2007 Hacker News post counted as reaction to a 2026 changelog entry. 1,099 of 2,885 couplings violated causality (Hacker News worst at 80%, because searching a company name surfaces its whole history) and have been removed; the coupler now rejects them. Every finding above survived the correction and two got stronger — but the landed rates here are lower than first published, and that is the honest direction of the error.