Sweeper — RSI(2) outcome distribution

Legacy above-SMA rank — classic exit diagnostic

Per-day above-SMA rank buckets resolved with the classic exit: first close > 5-day SMA, else −7% catastrophe stop, sold at the exit-signal close. 963 resolved trades. This is retained to compare old queue behavior under the current exit.

Ranking-feature stability — does the direction hold, week after week?

Per ISO-week Spearman between each candidate ranking feature and the classic-exit return (963 resolved trades). Pooled tables above can be driven by one hot week; a real prioritizer keeps its sign across weeks. ✓stable = sign held across ≥4 weeks (one dissent allowed; near-zero weeks abstain, shown ·). Positive rho = higher value → better return. Same construction as the 1-min eval.

Pre-close buy approximation

Paper stock RSI entries since 2026-07-07: the 15:50 ET pre-close buy pass compared with the same-day settled 17:00 RSI signal. A miss means the live pre-close buy did not appear in the final EOD signal list for that symbol/date.

Classic exit timing: close vs next open

Compares the current eval assumption, selling at the SMA5/time-stop bar close, with the executable alternative if the decision waits for the confirmed close: sell at the next session open. Only rows with a next open logged are included.

Deterministic setup assessment

Advisory risk labels built from RSI14, SPY40, EMA5, candle shape, recent-high failure, and support-break proximity. These labels do not change the current rank or stock execution.

Setup quality

Options quality

Assessment flags

Connors-style filter diagnostics

These are visibility-only filters for the RSI list: RSI14 leader strength, close below EMA5, and SPY above SMA40. They do not change the existing candidate rank. Rows resolve with the current classic/SMA5 exit. Resolved rows with any of these fields: 0.

RSI14 leader strength

Higher RSI14 means the RSI2 dip happened inside a stronger intermediate trend.

EMA5 pullback state

Connors-style pullback confirmation: the signal close is below its 5-day EMA.

SPY SMA40 regime

Broad-market filter from the Reddit test: only buy dips when SPY is above its 40-day SMA.

Combined filters

These combinations show whether the filters add value together; they still do not change the live rank.

Track record — classic exits

1018 alerts (2026-05-12 → 2026-07-14), each entered at the NEXT session's open and resolved with the current classic/SMA exits. classic = exit on first close above the 5-day SMA, sold at that close (−7% catastrophe stop); a few run past the logged window (46 unresolved). classic+5d = same, but forced out at 5 days if the SMA5 signal hasn't fired (a time cap on the open-ended hold). pure time-stop 6d = SMA5 exit OR a hard exit at 6 days, with NO price stop — avoids the intraday-low whipsaw (a −7% low-stop can eject you on a wick the name recovers from) but has no catastrophe cap, so the worst-trade tail is much wider. MFE = the best gain reachable BEFORE the −7% stop (a ceiling, not a capturable return). Losers are counted, never hidden. "avg/trade" is per-trade, flat stake — NOT a compounded account return.

Realistic account — finite slots (the honest total)

A real $25,000 account with 30 small slots (~$1,667/position, 7% of equity each → 2.0× buying power, fundable under Reg-T), taking alerts in date order with the live queue, holding each to its classic exit. This is the honest total — NOT the "sum of per-trade %", which pretends you can hold every alert at once. Net of ~0.10% round-trip cost (spread + slippage; commission-free broker). 2026-05-12 → 2026-07-14 (~7 weeks).

Small positions (not leverage) are how you fund more slots: 100 slots on $25k would need 20× buying power (impossible) — but 30 slots of ~$1.6k each is 2×, fundable, and captures far more of the signal flow than a few fat positions. Diminishing past ~50 slots, and below ~$1k/position costs start to dominate.

The classic sim deploys ~2× / $1,667 per position. Note: your own account is slot-bound — most alerts get skipped when 30 slots fill. An alert service isn't: each subscriber takes the alerts they want with their own capital, so the per-trade record above is what a subscriber experiences, not this slot-capped account total.

Candidate score diagnostics — live puretime exit

Diagnostic only. The live trader does not use this score for priority. These use the same puretime/classic return shown on the list page, so score drift and inverted buckets are visible before anything is promoted to live selection.

Score bands — lower score is not necessarily worse

All signals — hold-day distribution

By RSI(2) depth at entry

Same distribution, split by how oversold the signal was at entry. Compare the median row across buckets: if deeper-oversold signals carry a higher median close return, the depth is doing work; if the medians overlap, it isn't.

RSI(2) 0–2 — 131 signal(s)

RSI(2) 2–4 — 213 signal(s)

RSI(2) 4–6 — 229 signal(s)

RSI(2) 6–8 — 239 signal(s)