Solid-colour initial placeholders collide on same name or same colour

What it is

A letter on a flat colour looks as if it solved "no photo": Zhang on blue, Li on green. Collision has two doors — same name (two Zhangs, two Alexes) and same colour (eight swatches, the ninth person wraps). The doors can open together: two identical blue A's in one list. The placeholder then identifies a class of letters, not a person.

Why it happens

The code's capacity is size of the letter set times size of the colour set. Chinese family names have low entropy — a handful of surnames take a large share; Latin initials also clump. Short palettes fill the Cartesian product as soon as a real directory lands. Recognition also depends on "that blue A from last time", not on a hash that promises global uniqueness. Hashing a name to a colour only postpones collision: the moment two names map to the same (letter, colour) pair, the graphic channel collapses. If letter-on-field contrast is weak, the letter itself vanishes and only the swatch remains, entropy lower still. Same-name and same-colour are not edge cases. They are this encoding's ceiling.

Studying it

Generate lists from real name distributions (not uniform random letters), count collisions of (letter, colour) pairs, then run a "find this person" task. Independents: family name versus full name, palette length, second letter present or not. Dependents: colliding pairs, search time, false identification. Report where the names came from — a uniform 26-letter draw hides the problem. Also measure how many palette pairs remain discriminable at small size and on a dark field; an indiscriminable colour is a missing dimension.

Where it stops holding

• A tiny team with very different names can live with the collisions.
• A second letter, or a letter plus a stable pattern, raises capacity. It does not turn the encoding into a different kind of thing.
• The claim does not start when everyone has a unique photo.
• Distinguishable defaults are the general demand. This isolates this encoding collapsing on same name or same colour.

Applying it

• Do not ship a single letter plus a short palette. Add a second character, or lengthen the palette to pairs still discriminable at small size, and handle hash collisions (change the pattern, do not silently reuse).
• Letter and field must each remain readable at small size and on a dark field. A colour that only works in a large disc on the mock does not count.
• How to check: run placeholder generation on real user names, list identical (letter, colour) pairs. Each pair is a mis-tap candidate; seat those two people as neighbours and see whether the wrong one is hit.

Related

• Same group: F6.09.1 Avatars identify people; defaults need distinguishable placeholders · F6.09.3 Inconsistent avatar shapes slow list scanning · F6.09.4 Photo crops have to keep the face and other recognition-critical regions
• Nearby: F5.09 Perceptual requirements of a palette · F4.08 Legibility and discriminability
• Search terms: initial avatar · avatar collision · letter avatar