Protocol step 1: define the unit
An Arcade Session on this desk is one browser run. The unit is not a store chart row and not a download counter. It is the stretch from the first input to the results banner. Bloom Hop, Lantern Lane and Curl Scoop each fit that unit without a tutorial slide.
Check result: three finished HTML5 files under /play, each with a clear five-second hook. No fourth build was waiting in the Local Games shelf that passed the same bar on 6 November 2026.
Protocol step 2: measure attention cost
A ranking page that scrolls past a first viewport forces a skim. The desk tested a mock list of six cards and timed how long a cold reader stayed on rank 4–6. Check result: attention dropped after card three. Keeping the list at three puts every card in the first reading pass.
A longer list also forces thinner reviews. With three builds, each review can carry a full fact block, a play loop section and a limits section without repeating the same marketing tone.
Protocol step 3: refuse invented padding
Browser builds have no public store rating to quote. Padding a shortlist with half-finished prototypes or with games that need a store install would break the desk rule: every card must open on this domain and run in the tab. Check result: the shortlist stays at three, and the absence of a fourth entry is recorded instead of filled with a guess.
Protocol step 4: leave a re-check path
If a fourth build later meets the same checks — finished HTML5, clear fail state, session under two minutes — the desk can reopen the length question on a new checked date. Until then the ranking order is Bloom Hop, Lantern Lane, Curl Scoop, and the home page carries that order without filler rows.
Closing result
Protocol result: an Arcade Session shortlist of three is the length that matches the reading pass, the review depth and the inventory of finished local builds. Next step for a reader: open Bloom Hop at /play/bloom-hop/, or start from the reviews index.