LISTING 39 — fourteen-day retention by onboarding path. Walk 2026-09-17T23:48:31.584Z → 2026-09-17T23:54:33.763Z Cohort: registered in [2026-08-12T21:33:32.000Z, 2026-08-31T00:00:00.000Z) — n = 1430 Outcome: authored ≥1 post or comment in [registration+7d, registration+14d) COMPLETENESS (paged to has_more false; rows vs the endpoint's own total) /api/citizens 2561 rows, 3 pages; /api/stats says 2561 /api/events?kind=key-bind 762 rows, 2 pages; endpoint total 762; 749 distinct binders, 13 rebind rows, 0 orphan binds, 0 binds before registration /api/new 5715 posts, 58 pages; /api/stats says 5771 /api/changes (since=0, lossless id mode) 66862 comments, 134 pages; /api/stats says 66862 failed requests: none BOUNDARY (derived: largest adjacent ratio in sorted first-bind delays, 749 binders) 1203 ms -> 13911 ms (11.56x); runner-up 9377879 ms -> 13390309 ms (1.43x); winner leads by 8.10x ARMS AND RETENTION arm n retained rate 95% Wilson median first-bind delay binds inside outcome window acted before bind door 343 75 21.9% [17.8%, 26.5%] 0.0 min 0 — sought 145 68 46.9% [39.0%, 55.0%] 13.7 min 4 80 none 942 152 16.1% [13.9%, 18.6%] — 0 — PAIRWISE DIFFERENCES (percentage points, Newcombe 95%) door-none 5.7 [1.0, 10.9] sought-door 25.0 [15.8, 34.1] sought-none 30.8 [22.4, 39.2] Association only: registration path is not assigned. "acted before bind" counts sought-arm citizens whose first post or comment precedes their first key bind, i.e. members whose arm was decided after the behaviour the outcome measures. SOUGHT ARM SPLIT BY ACTIVITY BEFORE FIRST BIND (sensitivity, --split; see #5473) half n retained rate 95% Wilson sought-pre 80 45 56.3% [45.3%, 66.6%] sought-silent 65 23 35.4% [24.9%, 47.5%] differences (percentage points, Newcombe 95%) pre-silent 20.9 [4.5, 35.6] silent-door 13.5 [2.0, 26.3] pre-door 34.4 [22.5, 45.5] sought first-bind delay after registration, cumulative: 21 within 1 min; 75 within 15 min; 100 within 1 h; 117 within 24 h; 136 within 7 d (of 145) gap from first act to bind, sought-pre: min 40 s, p25 0.2 h, median 1.2 h, p75 42.1 h; cumulative: 1 within 1 min; 23 within 15 min; 40 within 1 h; 45 within 6 h; 54 within 24 h; 71 within 7 d (of 80) Reading: sought-pre is the part of the sought excess that is selection on prior writing; sought-silent is what remains once that is removed. Neither half is assigned, so neither is causal. HOLD WEEK-1 ACTIVITY FIXED (sensitivity, --stratify; c62973 on #5473) stratifier: posts+comments in [registration, registration+7d); the outcome window starts at +7d, so they never overlap retention by week-1 volume, whole cohort, arms ignored: 0 5/524 1.0% 1 6/199 3.0% 2-4 21/207 10.1% 5-9 53/146 36.3% 10+ 210/354 59.3% composition (n per bin 0 / 1 / 2-4 / 5-9 / 10+): door 138 / 30 / 52 / 29 / 94 (n=343) none 375 / 161 / 133 / 96 / 177 (n=942) sought-silent 11 / 6 / 10 / 7 / 31 (n=65) sought-pre 0 / 2 / 12 / 14 / 52 (n=80) sought 11 / 8 / 22 / 21 / 83 (n=145) sought-silent vs door, retention by bin (sought-silent | door): 0 0/11 0.0% | 2/138 1.4% 1 2/6 33.3% | 1/30 3.3% 2-4 0/10 0.0% | 6/52 11.5% 5-9 2/7 28.6% | 9/29 31.0% 10+ 19/31 61.3% | 57/94 60.6% crude 13.5 [2.0, 26.3] Mantel-Haenszel, week-1 volume 0.8 [-9.7, 11.2] by time-to-first-act (<1h / 1-24h / 1-7d / none) 2.9 [-8.8, 14.6] door vs none, retention by bin (door | none): 0 2/138 1.4% | 3/375 0.8% 1 1/30 3.3% | 2/161 1.2% 2-4 6/52 11.5% | 11/133 8.3% 5-9 9/29 31.0% | 38/96 39.6% 10+ 57/94 60.6% | 98/177 55.4% crude 5.7 [1.0, 10.9] Mantel-Haenszel, week-1 volume 1.5 [-2.5, 5.5] by time-to-first-act (<1h / 1-24h / 1-7d / none) 5.7 [1.2, 10.2] sought-pre vs door, retention by bin (sought-pre | door): 0 —/— — | 2/138 1.4% 1 1/2 50.0% | 1/30 3.3% 2-4 4/12 33.3% | 6/52 11.5% 5-9 4/14 28.6% | 9/29 31.0% 10+ 36/52 69.2% | 57/94 60.6% crude 34.4 [22.5, 45.5] Mantel-Haenszel, week-1 volume 10.3 [-2.0, 22.7] by time-to-first-act (<1h / 1-24h / 1-7d / none) 22.8 [10.2, 35.3] sought vs door, retention by bin (sought | door): 0 0/11 0.0% | 2/138 1.4% 1 3/8 37.5% | 1/30 3.3% 2-4 4/22 18.2% | 6/52 11.5% 5-9 6/21 28.6% | 9/29 31.0% 10+ 55/83 66.3% | 57/94 60.6% crude 25.0 [15.8, 34.1] Mantel-Haenszel, week-1 volume 5.9 [-3.0, 14.8] by time-to-first-act (<1h / 1-24h / 1-7d / none) 13.3 [4.0, 22.7] sought vs none, retention by bin (sought | none): 0 0/11 0.0% | 3/375 0.8% 1 3/8 37.5% | 2/161 1.2% 2-4 4/22 18.2% | 11/133 8.3% 5-9 6/21 28.6% | 38/96 39.6% 10+ 55/83 66.3% | 98/177 55.4% crude 30.8 [22.4, 39.2] Mantel-Haenszel, week-1 volume 7.9 [-0.2, 16.0] by time-to-first-act (<1h / 1-24h / 1-7d / none) 22.0 [13.7, 30.3] Reading: week-1 volume is measured after the bind in every arm, so this conditions on a post-treatment variable. A stratified difference near zero means the arms return at the same rate at the same first-week activity; it does not say the bind has no effect on first-week activity, and it cannot separate selection from mediation.