basquetWi + New ticket
mars MARS-157

Analyze client-env screen-size distribution, optimize breakpoints to dominant cohort (verified vs distribution)

Done low unassigned

Sub-tickets

No sub-tickets.
+ Add sub-ticket

Questions

No questions.

Activity

  • wi-cli-venus created · 2026-06-20
  • wi-cli-venus commented · 2026-06-20
    From Elazar telemetry brainstorm bs-mqmscr4k3wh (2026-06-20). Deliverable: ranked screen-size distribution + CONCRETE layout/breakpoint changes, optimization VERIFIED against the actual distribution (not a guessed breakpoint). Blocked on MARS-156 (frozen dataset snapshot). Mars data already directional: 79% mobile / 21% desktop; dominant cohort iPhone-class 384-415px iOS portrait = 32% of ALL sessions alone; top-3 mobile bands 320-447px = ~69%. CAVEAT: only 5 days of data - unambiguous for the mobile-first directional call, but keep fine breakpoint-tuning honest about the short window; do not over-fit a 5-day sample. Analysis runs against the MARS-156 frozen export.
  • wi-cli-venus commented · 2026-06-20
    FINAL per closed decision bs-mqmscr4k3wh. Analysis runs against the MARS-156 k-anon frozen export (k=5, role[] not actorUserId). Deliverable: ranked screen-size distribution + CONCRETE breakpoint/layout changes, VERIFIED against the actual distribution. Mars directional signal: 79% mobile / 21% desktop; dominant iPhone-class 384-415px iOS portrait = 32% of all sessions; top-3 mobile bands 320-447px ~69%. Keep the 5-day over-fit caveat honest in fine breakpoint tuning. Blocked on MARS-156 export.
  • wi-cli-venus unblocked · 2026-06-20
    Frozen dataset ready (MARS-156 DB leg): ~/gdrive/projects/evolutiva/mars/telemetry/client-env-viewport-kanon-k5-2026-06-20.json. Analysis can proceed against this file; coder leg of 156 (stop-pipe) does not gate analysis.
  • wi-cli-venus completed · 2026-06-20
    Analysis complete — Mars breakpoints (640/768/1024) confirmed OPTIMAL, no change warranted (640 captures 100% of phones ≤447px; lowering would regress 416-447 cohort). 79% mobile, 384-415px portrait = 50.5% of sessions = canonical design/QA target (320px hard floor). No code diff is the correct outcome. Captured in memory reference_mars_viewport_distribution. Concrete 384px hardening split to MARS-158 (needs browser-tool agent).
coder
2026-06-20 by wi-cli-venus
6w ago
2026-06-20 20:17