# A1 iterate run 28 (10 Oct 2026): radar_hist cross-venue cut

Data: /opt/sw/radar/radar_hist.jsonl (18,359 crossings, 12 Sep - 9 Oct 06:28, 8 venues; the history I asked for in #154 arrived). Scripts t1-t4 in research-box /opt/crew/work/analyst1.
Universe U: pools born < 300k that touched 300k, not thin. "Real" = reached 1M at least 120 s after the 300k touch and depth >= $20k at 1M. Real is only measurable from 21 Sep (depth field missing before; everything before that shows ~0 real on hooked, do not read it).

| venue | touched 300k (non-thin) | real 1M | real 3M | 5M | 10M | 20M |
|-|-|-|-|-|-|-|
| v3 | 508 (244 since 21 Sep) | 149 (72) | 20 | 17 | 12 | 10 |
| v4 WETH hooked | 288 | 116 (64) | 33 | 17 | 5 | 0 |
| v4 WETH no hook | 381 | 37 (10) | 0 | 0 | 0 | 0 |
| v4 native no hook | 192 | 23 (19) | 0 | 0 | 0 | 0 |
| v4 native hooked | 211 | 9 | 0 | 0 | 0 | 0 |
| v4 USDG no hook | 19 | 0 | 0 | 0 | 0 | 0 |

(counts of 3M+ are since 21 Sep.)

## Iterations
1. Lens: backing ratio (depth / cap at the touch). Assumption: thin depth = cheap to push = more runners. v3 pre-21 Sep 46% vs 12%, but after 21 Sep the sign flips between halves (0.38/0.22 vs 0.20/0.43). DISCARD (depth field artefact before 21 Sep).
2. Lens: age at 300k touch. Hooked WETH: touch within 30 min of birth 80% real (138 pools) vs 4% older (150); every day since 21 Sep, n per day 5-11. v3 age <=2 min 35% vs 21%; native <=12 min 23% vs 4%; NHW flat. INFO: for hooked WETH, a pool that has not touched 300k by minute 30 is dead (6 of 150 later real) = time stop for the H3000 tests. Hooked is already covered by tm-h3k/h3n (13x means), so no new test.
3. Lens: birth cap and birth hour on all venues. Birth cap tracks age effect, hour flat except v3 night 17% vs 37% (known). DISCARD.
4. Lens: regime / crowding (other 300k touches in prior 60 min, Ben's "how many coins run at once"). Flat on every venue (v3 .32/.30/.19/.32; NHW .08-.15, noise). DISCARD (#129, #95 agree).
5. NEW lens: social attention lead-lag. savenow.db xposts (17.8k X posts, 121 tokens, 89 in radar). First post is before the 1M touch for only 17% (0% on native, 12% USDG), before the 300k touch 29%; median first post 3.7 days after. The X scrape is search-after-the-fact, not a lead signal. DISCARD; it would need a live scrape per new ticker, and the 17% says it lags price anyway.
6. **NEW lens: exit ceiling by venue (tier survival).** 29 real climbers on v4 no-hook (10 WETH + 19 native): ZERO reach 3M. v3: 28% of real 1M reach 3M, 24% 5M, 14% 20M. Hooked WETH: 46% 3M, 27% 5M. P(0/29 | v3 rate 28%) = 7e-5. So for unhooked v4 the only runner tier that exists is $1M (maybe 2M); a hunt there is a 3.3x move from 300k, never a 10x ride. Exit rule for NHW-S6 / USDG nohook / native tests: sell at 1M, do not hold for tp5+ (their own sims show tp3 ~ tp5 for NHW-S6 1.86 vs 2.25, so keep tp3-3.3 as primary and treat tp5 gains as jump artefacts to re-check). Also tells the mission: every $3M+/$5M+/$10M/$20M runner that is real is v3 or hooked WETH; the unwatched unhooked pools hold only $1M-tier runners, the rest of their 3M+ rows are first-print phantoms (#145, #176). MATURE: sent as note to testmgr (cap NHW-S6 and native tests at 1M target; check if tp5 arm wins by a few trades).
7. Lens: unknown venue = token-quoted hook (tokq). Radar now labels it: v4:tokq:hook, 4 crossings at 1M+, 3 thin, three PLTR pools show identical cap 2,174,276 (stock-token quote price artefact), SHROOM $5M row is the same family. DISCARD as a runner source (agrees #59, #79).
8. Hooked WETH 300k-touch confirm entry (age <=30 min): P(1M) 80%, tier 3M 46%. Not tested: H3000 tests enter at +60 s with 2.5-13x means, an entry at 300k gives away 10x. PARK (not worth a new arm; only valid as fallback for pools missed at birth).
9. Native hook (v4:native:hook) 1M "runners": 245 touches, 236 are same-minute jumps (median gap 0 s, depth 6 dollars). Pure first-print rows: tell runner-check to drop gap=0 / depth<$20k rows (research #568 already open).
10. Data request (below).

## Data request (ppfeeds / ppdata): cap path denominators
radar_hist only holds pools that touched 300k, so I cannot measure time stops, nor entry below 300k, nor what fraction of young pools die. Need, for every pool born since 12 Sep on all venues (v3, v4 hooked/no hook by quote, not thin-only): one row per pool with cap_first, cap at age 1/5/15/30/60 min (cap path from the pool's own swaps, mark-to-last-print), first time cap >= 50k/100k/200k (th events at these levels in addition to 300k), max cap, depth_usd and n_senders at 5 min. Format: jsonl next to radar_hist, same fields as radar.py plus these. Tests: (a) real time stop minute per venue (hooked suggests 30), (b) entry at 100k-200k touch for v3 and NHW: P(1M)*10 vs fee, (c) shape at 15 min as filter for second-chance entries. Not asked for is X data.