Inverted Pyramid 3D

Readers walk down the story and leave at random depths; reorder the facts and see how much news is still delivered.

Switch to 2D

Story order

Newsroom conditions

Camera

Playback

Quality

Model checks

running...
The science

What is simulated

Each simulated reader scrolls through one news story and leaves at some depth y (0 = headline, 1 = last line). Rather than a fixed reading time, leaving is a hazard that grows as the text gets longer and denser: h(y) = k (1 + f y). The share of readers still reading at depth y is the survival function S(y) = exp(−k (y + f y²/2)). A strong lede lowers the base rate, k = k0 (1 − hook/150); dense prose raises the growth of the hazard, f = complexity/50.

Why the order of facts matters

The story has six facts of different newsworthiness (A–F, 10 down to 2; illustrative scores). Each fact fills one sixth of the story. A reader exiting at depth ye sees the fraction clamp((ye − p/6)·6, 0, 1) of the fact in position p. The delivered information is the newsworthiness-weighted average of what readers saw, I = Σ wp · 6∫p/6(p+1)/6 S(y) dy / Σ w. Because S(y) only decreases, putting the largest weights where S is largest maximises I (rearrangement inequality). That is the journalistic “inverted pyramid”: front-load the most important facts because attrition is unavoidable.

Independent numerical checks

(1) 200,000 readers are drawn by inverse-transform sampling (solving the quadratic k(y + f y²/2) = −ln u) and their mean read depth and completion rate are compared with the closed forms ∫S dy and S(1). (2) The same Monte-Carlo sample is used to compute information delivered and compared with the Simpson-rule integral. (3) All 720 orderings are scored; the maximum must be the newsworthiness-descending order. The on-screen crowd is a smaller sample, so its numbers wobble around the model values (shown in brackets).

Honest simplifications

The hazard form, the constants k0 = 1.8 (adjustable), the hook and complexity scalings and the newsworthiness scores are illustrative, not fitted to any publisher's analytics; real scroll-depth studies show that most readers leave before the end, but the shapes vary a lot by site, device and story. Readers are assumed to read each fact fully until they exit and never skip ahead or re-read, facts are equally long, and the headline and design effects are folded into the single hook parameter. The model says what order loses least to attrition; it does not say that short, shallow stories are always better, nor that all facts must be sorted when a story has a narrative arc.

Scroll position0 %
Avg read depth0 %
Info delivered0 %
...
Loading...
Readers still reading at each depth (model vs this crowd)
Newsworthiness of each position and the part readers actually see
Information delivered vs hook strength, for three orders
very newsworthybackground detailblue reader: still readingamber: left after the first thirdgrey: left early
Made with 🟠 Blender · Three.js