SHYPIARD
← METHOD
METHOD · RANKINGS · V3

How these lists
are ordered.

Eligibility and order are different jobs. A game reaches a ranking because taxonomy and observed Steam facts say it belongs. Where it sits on that ranking is a weighted score, then a name. The score is a fit to the list, not a verdict on the game.

WEIGHTS

Five signals, then re-normalise.

Intent fit
35%

Does the game belong on this list at all: taxonomy, mechanics, player-count, fantasy, and the ranking's own constraints. An authored page is not a bonus here.

Review sentiment
25%

The observed Steam review score from the snapshot. Missing scores are skipped, never invented as zero.

Review volume
15%

Log-normalised against the largest review count in that list's pool, so a 200-review title is not crushed by a 200,000-review neighbour.

Genome slice
15%

Only the genome attributes this list actually uses. A public score counts at full weight, an estimate at half, a hidden score is dropped. Coverage below 30% drops the whole slice.

Data confidence
10%

A fresh observation, more than one source, an authored page, and evidence strength. This is where editorial presence lives — never inside intent fit.

MISSING SIGNALS

Unknown is not zero.

If the genome slice has less than 30% usable coverage, that 15% is dropped and the remaining 85% is re-normalised: total = sum(weight × score) / sum(used weights). A hidden genome attribute cannot pull a ranking the way a public one can. Inverted attributes — execution difficulty on a cozy list — use 1 − score, so a high twitch demand lowers fit.

TIES AND DATES

Alphabet, then the snapshot.

Equal totals break alphabetically by title. Every metric on a ranking card carries the date it was observed. Price and demo lists go stale faster than a genre list; when those collections age out, they lose indexability rather than pretending the store still looks like the snapshot.

THE 29 LISTS

What each ranking filters.