Open fixes and ideas for the bots, in priority order. Updated 2026-10-13. Status: TO DO / IN PROGRESS / DONE. Test results for entry ideas: entry knowledge base.
| # | Fix / idea | Why | Bot | Status | Added |
|---|---|---|---|---|---|
| 1 | Intraday rescans re-score the whole universe with what is known at that time (move from the open, range, volume vs normal for that time, VWAP, trend...), instead of reusing the premarket score | Rescans change only 1-4 of 50 names, so stocks that start moving during the day never enter the list | Star bot ON (trade bot option OFF) | DONE 10/13: top 30 beat both premarket scores at every rescan time, both years (10:00 up-runs 53%/57% vs v3 50%/47% vs old 41%/39%); live for the Star bot from 10/14 | 10/13 |
| 2 | min_atr_pct threshold test (4 / 5 / 6 / 7 / 8%) | 8% blocked all 6 Star-bot buys on 10/13 (5 of 6 then rose); its old-score list is 88% calm stocks. Star bot has it OFF now, trade bot ON | both | TO DO | 10/13 |
| 3 | thin_trading avoid rule: median shares/min over 15 min below a floor | RILY 10/12: one 53k-share print made a thin stock pass the volume and liquidity rules (averages) | both | TO DO (build OFF + 2-year test) | 10/12 |
| 4 | Tighter spread limit (e.g. 0.3%, or scaled to the expected move) | RILY: a 0.4% spread ate most of the loss; the 1% limit is loose for $5 stocks | both | TO DO (record live; no quote history to backtest) | 10/12 |
| 5 | Extension measured in ATRs (above the open / above VWAP in daily ATRs) | TAL 10/12: +1.9% over the open = 0.7 of its whole daily range; the % limits are loose for calm stocks | both | TO DO (v2 has ext_hard_max_atr / max_move_x OFF) | 10/12 |
| 6 | Gap-down: previous close as resistance | TAL stalled at its previous close (gap fill) right after the buy | both | TO DO (build OFF + test) | 10/12 |
| 7 | Report the avoid-rule + news tests | No single rule makes entries positive; min_atr_pct best (-0.007%); all on = 18 trades -0.47%. News at the open = more movers (+5% runs 39-41% vs 26-29%), not better trades | - | DONE 10/13 | 10/13 |
| 7b | Combination test: min_atr_pct + lower_low_by_1030 (+ max_above_open_pct) | On ATR >= 8% stocks lower_low_by_1030 gave +0.087% (262 trades) and max_above_open +0.011%; must be a real replay of the pair | both | TO DO | 10/13 |
| 7c | News flag as a scanner input (premarket news, esp. "why moving", guidance, FDA, earnings) | Picks with news reached +5% 39-41% vs 26-29% without, both years -- a mover signal like ATR, not a direction signal | scanner | TO DO (add to the intraday / v3 score study) | 10/13 |
| 8 | Replay last Mon-Wed with today's settings | Settings changed since (scanner, orders, guards, stop) -- separate settings vs market days | both | TO DO (offered) | 10/12 |
| 9 | Order-book pressure test (bid vs ask size at entries and V turns) | The one signal not testable yet; recording since 10/12 | trade bot | WAIT 2-3 weeks of data | 10/11 |
| 10 | SEC filing tags 10-year backtest | Recording since 10/7 (avoid / catalyst tags); needs history to test | both | TO DO | 10/7 |
| 11 | Star's sells vs the bot's exits | Live 7 days: Star better 28 of 49; 2-year test: Star worse | - | WATCH (weekly review) | 10/11 |
| 12 | Compress old decision logs | ~200-260 MB per bot per day; ~1 month of disk left | both | TO DO (offered) | 10/11 |
| 13 | Security before go-live: Apache serves /var/www/screener incl. .env files (keys downloaded by outside IPs in Sept); block dotfiles/data, rotate Alpaca keys, report password | Deferred by the user until the bot is approved to go live | all | BEFORE GO-LIVE | 9/25 |
| - | Watch list not updating on rescans | Checked 10/13: rescans run every 5 min and apply; the list barely changes because the score is premarket-based -> fix #1 | Star bot | EXPLAINED | 10/13 |
| 14 | A. Market filter: no buy while IWM is below its open (A1) / falling over 30 min (A2) | 10/14 all 6 buys lost in a falling market; 10/15 rising market, most won | both | TESTED 10/16: WORSE (A1 -0.151%, A2 -0.142% vs live -0.133%) -- drop | 10/15 |
| 15 | B. Clustering guard: 3+ stocks signalling within 5 min -> only if IWM rising | 10/14 all 6 buys came 10:40-10:46 on a small market bounce | both | TESTED 10/16: only 10 clustered buys in 2 years (-0.38% each) -- negligible | 10/15 |
| 16 | C. More room to resistance (v2 min 1% -> 5% / 10%) | 10/14 room 2.6-7% vs 10/15 13-28% | both | TESTED 10/16: BEST of A-E: 5% -0.103% (halves -0.17% / -0.03%), 10% -0.105%; not positive. Next: test with min_atr_pct | 10/15 |
| 17 | D. Volatility-scaled stop (0.25 / 0.35 x daily ATR, max 6%) | 10/15 CIFR / BLZE stopped at -2% then closed +11% / +5.6% | both | TESTED 10/16: WORSE (-0.142% / -0.148%) -- drop | 10/15 |
| 18 | E. Sector trend for crypto stocks: buy only while bitcoin is up since 9:30 | 10/15 BTC +2.7%, miners ran; 10/14 BTC -0.9%, CORZ lost | both | TESTED 10/16: no help (-0.134%; crypto buys -0.21% -> -0.30%) -- drop | 10/15 |
| 19 | noise2 ON for the Star bot's sells | Best sell method 2 weeks running; 10/20 HL / TDOC / PGEN sold seconds-minutes after the buy above VWAP, then +1.2..+2.5% | Star bot | DONE 10/20 10:54 (user): ON, Star bot restarted flat | 10/20 |
| 20 | Sticky watch list: keep stocks traded today on the list after a rescan drops them | HL / TDOC dropped right after the sell -> no re-buy possible | both | TO DO (check on past days first) | 10/20 |
| 21 | 15-minute window edge check: don't let a big bar leaving the 15-bar window flip the rules (also check 30-min buy share / trend) | HPK 10/20: a 44k-share sell flush left the window -> buy share 31%->61% in one minute -> buy -> -$47 | both | TO DO (2-year test) | 10/20 |
| 22 | Situation-specific entry rules (user's design): tag S1-S28 on every v2 buy; rules S1 window flip, S2 spike wait-and-confirm, S11 flush cool-down | "Not one rule fits all" -- HPK 10/20; catalog entry_situations.html | both | TESTED 10/21: S1/S2/S11 rules do NOT help (keep OFF); harmful situations = S20 calm, S6 extended, S8 near resistance, S3 block print, S21 thin/wide, S10 late -> deep study one by one | 10/20 |
| 23 | C3: confirm the last 3 full bars at every buy signal (price rising, buy share and volume not falling) | HPK 10:06: no up bar in the last 3, flat price -- bought on a window flip | both | BUILT (OFF); 6-month test 10/21 16:20 (c3_test.py), then 2 years | 10/21 |
| - | New exchange "TXSE" broke the stock list (scans/rescans failed from 12:51 on 10/19, and the 16:35 reference build) | Alpaca listed a Texas Stock Exchange stock our library version rejects | both | FIXED 10/19 evening (raw-list fallback; references rebuilt for 10/20) | 10/19 |
| - | Day-page charts in New York time (axis + trade table were UTC) | User 10/13 | pages | DONE | 10/13 |
| - | Star bot min_atr_pct OFF + restart | User 10/13 12:32 | Star bot | DONE | 10/13 |
| - | Star bot: no buy while VWAP is lower than 30 min ago (finished bars; from the first bar if the day is younger). config entry_v2.vwap_down_block_min 30 | User 11/10 after the Star review (2y replay: loss per trade -0.151/-0.144 -> -0.061/-0.119, 3% stops 474 -> 212; 27 live days: -$629 -> -$115) | Star bot | DONE 11/10, live from 11/11 9:30 (revert: entry_v2.py.bak-2026-11-10-pre-vwap30 + data/config.json.bak-2026-11-10-pre-vwap30) | 11/10 |