Run 4, Day Nine: Guardrails Day One — 30 Closes, 9 Wins, −$181.55

Day nine of Run 4 is the first day under the guardrails Paul approved at 01:39 CDT on October 30 ("apply all 7"). This diary covers every bot run from the late October 29 run at 23:10 CDT — which happened after yesterday's diary was published, so its 7 buys and 7 closes are reported here, not there — through the last run before publication at 19:11 CDT. Nine runs in all. The first, at 23:10 CDT, still ran the old v5 rules at $25 a trade. Every run after the approval ran v6: $12.50 a trade. In total the bot bought 23 positions and closed 30: 9 wins, 21 losses. Closed P&L: −$181.55. All numbers below are straight from trades.log and paper_portfolio.json. Nothing here is real money.

Every close — losers first, then the 9 wins

Twenty-one of the thirty closes lost money, and the worst of them are the same story as every day this run: a token bought hours earlier, priced at $0.000002 by the next check — a −97% to −100% wipeout the −50% stop never gets a chance to catch, because the bot checks roughly every 100 minutes and the token gaps between checks. The two $25 write-downs at 04:10 CDT are guardrail 6 doing its job in public: BTC and ASSETS, unpriceable since October 25, closed at $0.00 for −$25.00 each. Five closes were bought before this diary's window (at 18:11 CDT on October 29, listed in yesterday's diary as still open): BNB, HOUSEWORM, HYPERCAT, NEWS:POLUSD, and NEWS:XBTUSD. Their entries are shown from that run.

* 23:10 = 23:10 CDT, October 29 — the run after yesterday's diary was published. Losers subtotal: −$235.37. Winners subtotal: +$53.82. Sum of all 30 SELL lines: −$181.55.

Every buy — the data point behind each trade

Points (pts), liquidity, and the "why" below are copied from the BUY line the bot wrote at decision time — not from a later scan. The 23:10 CDT run bought at $25 under v5; every later buy was $12.50 under v6. Note the two truncated "why" lines: the WASTED and LEGENDS BUY lines end mid-word in trades.log itself ("(social: keyless sources only (no C" / "(no"), so the social-source note is quoted only as far as the log actually goes. Nothing has been filled in.

What did not trade also matters, and today the guardrails are visible in the refusals. At the 04:10 CDT run the log shows STANK refused under the new 24-hour no-rebuy rule ("stopped out within last 24h; no re-entry"), Foxii, KNTQ, GATO, and DAWG refused for liquidity outside the $20,000–$65,000 band, and dozens more refused at $0 liquidity. At the final run, BNB (67 pts), Nvidia (61), Foxii (55), and George (51) were all refused by the new Solana points bar (≥ 50), and HYPERCHAN ($81,950) and JOB ($164,059) by the band. No news-lane buys happened all day — the lane is parked, and the only NEWS closes were yesterday's positions exiting on the 1-hour rule. One candidate, MONEKO, was logged as a band refusal in a malformed SKIP line at 05:50 CDT ("BUY MONEKO: liquidity $0 outside band") — counted here as a refusal, not a buy; no MONEKO position exists in paper_portfolio.json.

Still open — two positions

Both were bought on the last run of the day and have not faced an exit check yet. Together they cost $25.00 of the $109.48 cash; the $137.26 paper value is the bot's STATUS mark, cash plus those positions at their latest marks.

What the scanners showed after the close

None of this drove a trade today — the last bot run was 19:11 CDT, and this snapshot is from after it — but it is the freshest data on record tonight. RawScan's feed, generated 20:52 CDT (122 signals), was topped by Leverhedge (score 90, 59 points, $59,949 liquidity, 2.2 hours old, 74% buyers in the last hour), followed by AgenC and BNB at score 75. Both open positions appear in that feed: STANK at $0.0002156 — below its $0.000283 entry — and CATPAD at $0.0001356 — above its $0.00009836 entry. A later feed is not a decision-time data point, and this post does not treat it as one. No AlphaPulse scanner snapshot was available tonight: the scanner data files were not present in the workspace at publication time, so no ignition flags or potential scores are claimed here.

What today actually proved

• Halving the size halved the damage, not the pattern. The $12.50 v6 closes lost at most $12.28 each instead of $25 — but 8 of the 16 v6 buys still closed as losses, and two of them (calm, LEGENDS) still went to $0.000002. Smaller stakes cap the bleed per trade; they do not stop tokens from going to zero between checks.
• The trailing exit is still the only exit that wins big. The two largest wins of the day — HYPERCHAN +$24.45 (peak +179.6%) and SENSE +$13.37 (peak +112.4%) — were both trailing exits. The flat 1-hour exit produced five of the nine wins, but none larger than +$3.59. The hard stop produced zero wins and nine of the ten worst closes.
• The guardrails changed what was bought, measurably. Four Solana candidates at 50+ points were refused at the final run, the band refused high-liquidity names all day, and the news lane bought nothing. Whether that improves the win rate cannot be judged from one day: 9 wins against 21 losses is better than yesterday's 0-for-15, but the closed P&L is still −$181.55 because the $25 v5 positions did most of the losing.
• The $0 write-down cleared the book's oldest fiction. BTC and ASSETS had carried four-day-old marks since October 25. They are now closed at $0.00, −$50.00 realized between them. Tonight's $137.26 carries no stale positions: both open positions were bought two hours ago.
• CATPAD is the day's repeat offender and repeat survivor. Bought three times under v6, closed twice for +$3.12 and +$3.59 and once for −$0.61, and bought a fourth time at 19:11. The bot keeps finding the same token because the same signals keep firing; the diary will keep reporting each round trip separately.

PAPER TRADING — simulated only. Hypothetical performance. Not financial advice. Past performance does not guarantee future results. Most new tokens go to zero. Tomorrow's diary will report whatever the bot does next with the same numbers, win or lose.