Tips & Tricks

Tips & Tricks

Plain-English trading concepts, learned hands-on. This is the human companion to the machine docs (Trading Bot Systems and Strategy Findings) — written the day I first day-traded manually (2026-07-15) instead of only running the bots. Everything here is the same MACD + momentum logic the bots use, explained for a person clicking Buy and Short by hand.

1. Long vs Short, Call vs Put

These are two different questions: which way am I betting? (direction) and what am I actually holding? (instrument).

• Long = own the shares → profit when price rises.
• Short = borrow-and-sell shares → profit when price falls.
• Call = option that profits when price rises (the options cousin of long).
• Put = option that profits when price falls (the options cousin of short).

So a put is the cousin of a short, not of a long. “Long” and “put” are near opposites.

Max-loss is not symmetric — this is the part that matters with real money:

2. How a short works

You sell first, buy back later, and keep the difference if the price falls.

1. Broker lends you shares and sells them → you get a credit now.
2. You now owe those shares back.
3. To close, you “buy to cover” — buy the shares back and return them. Whatever’s left over is your profit (or loss).

Worked example (real, 2026-07-15): short 12 SOXL @ $164.92 → credit $1,979.04. Covered by buying 12 @ $158.95 → cost $1,907.40. Profit = $71.64 in ~8 minutes.

Things to know:

• Borrow fee — you pay a small daily “rent” on the borrowed shares for as long as the short is open (SOXL was ~$0.35/day). Negligible for minutes, adds up over days.
• Margin — a short ties up buying power as collateral.
• Equity floor — you must keep the account above the $2,000 margin minimum (see §10).

3. Stops on a short (and why “break-even” isn’t break-even)

For a short, the protective stop is a buy-stop ABOVE your entry (you cover if it rises) — the mirror image of a long, where the stop sits below.

A stop set at your entry price does not guarantee a zero loss. What you actually pay:

1. The spread. You short near the bid and cover near the ask. Even a flat round-trip costs you the spread (SOXL bid/ask ~$0.21 apart → ~$2 on 10 shares). You sell low and buy back high — it works against you on both ends.
2. The borrow fee for the time held.
3. Slippage / gaps. A stop becomes a market order when it triggers, filling at whatever’s available. Intraday on a small order that’s usually pennies — but if the price gaps (news, overnight), your stop can’t fire at your number. A stop at $163.66 that gaps to $167 fills near $167 → a real loss, not zero.

Practical: a stop set right at entry usually triggers immediately on normal noise and churns you out for the spread. Give it a little breathing room.

4. MACD — reading the three numbers

MACD is a momentum gauge built from moving averages. On the chart it shows three numbers, e.g. MACD (12, 9, 26) −0.21 −0.65 0.44:

• MACD line above signal → histogram positive/green → bullish.
• MACD line below signal → histogram negative/red → bearish.
• Both lines below zero = the broader trend is still down (and vice versa).

The bots buy when the MACD line crosses above the signal (histogram flips green → call) and buy a put when it crosses below (histogram flips red).

5. The MACD histogram life cycle (the key picture)

The histogram is smallest at the crossover and grows as the move plays out. So a big, tall histogram usually means you’re LATE, not early — the move is mostly done.

What matters isn’t the size of the bars — it’s the direction they’re heading:

Takeaway: “is it green?” matters less than “did it just turn green, and are the bars still growing?”

The red-flip trap (added 2026-07-18, caught in trainer reps): “flips red” is an exit signal for longs everywhere — but it is a short entry only when the cross is fresh: momentum just crossing zero (not already deep), price just breaking off the top (not $5 into the slide), and not fighting an uptrend day into support. Check momentum’s depth at the moment of the flip: ≈0 and crossing = fresh, tradeable (the 2026-07-17 live short: MACD −0.48, momentum −0.06 → +$21). Already −5 = stale — the laggard confirming a finished move; in an uptrend that’s the bottom of a dip, where the bounce launches (cost two trainer reps + the 07-17 12:16 chase). Same signal, opposite value, decided by how far price has already traveled.

6. Momentum

The simplest indicator: current price minus the price N bars ago (Momentum(12) = price now − price 12 bars ago). Centered on a zero line.

• Positive = higher than N bars ago. Negative = lower.
• −0.61 just means “price is 61¢ below where it was 12 bars ago” — a mild drift down.

It’s a rear-view mirror, not a forecast. Negative doesn’t mean it’s about to go bearish — it means it already has been falling. Like the histogram, direction + size beat the raw sign:

• Big negative (−3, −4) = strong, real selling → confirms a short.
• Small negative near zero (−0.61) = pressure faded → quiet, indecisive, weak signal.

Use it as confirmation for the MACD cross: cross + momentum pushing the same way = a committed move. Momentum hugging zero = the move hasn’t committed yet → wait.

7. The manual method that worked (day one)

Same engine as the bots, two dials changed:

• Faster clock (1-min vs the bots’ 30-min): more signals, more whipsaw, smaller moves — but you catch both directions instead of waiting for one big swing.
• Enter right after the cross, once a few candles confirm the direction. This is the exact whipsaw guard the bots use (they act on closed 30-min bars, ignoring the in-progress bar) — done by eye instead of in code. Let the cross prove itself.
• Exit ½ to ¾ through the move — not greedy. The bots ride to the opposite cross and give back the tail of every move; taking the meat and leaving sidesteps that. It’s the histogram lesson in action: get out as the bars start shrinking, don’t wait for the flip.
• Momentum confirms the cross. MACD cross + momentum agreeing = go. Cross with flat momentum = the fakeout you skip.

8. Confirm to get IN, react to get OUT (the leader vs laggard rule)

The biggest lesson of day two (2026-07-16), and you spotted it yourself: you use several indicators to get IN, but only one — the fastest — to get OUT. Those are two different jobs.

• Getting in is about being right. A bad entry is an instant loss, so you demand agreement: momentum turning up and MACD green and price breaking the recent high. Confluence. If they don’t all line up, you skip it — that lineup-that-never-quite- forms is the fakeout that keeps chopping you.
• Getting out is about being fast. On a scalp the profit is already earned; the only question is keeping it. Reversals are quick, so you leave on the first fast gauge that rolls over and you do not wait for the slow ones to agree — by the time they do, the profit has walked out the door.

Why momentum leads and the MACD histogram lags — it’s in the math, not an opinion:

Momentum is the speedometer. The MACD histogram is that same drive smoothed twice — great for confirming a trend is real, useless for calling the top in real time.

The proof, live (2026-07-16 SOXL long):

• IN at $140.86 on full confluence — momentum turning up, MACD green, price broke the rolling 30-min high. All three agreed, so it wasn’t a fakeout.
• OUT at $142.42 — a trailing stop I’d raised from $141.89 up to $142.43 (banking an extra $0.54/share) triggered automatically at 11:26 AM as price rolled off the $143.38 peak. No staring, no flinching — the rule sold for me. My momentum gauge independently confirmed the top had rolled over minutes later; the MACD never agreed at all.
• The MACD histogram never turned red the whole way down. Fifteen minutes later price had round-tripped the entire spike — $143.4 → $140.61, below the $140.86 entry — and the histogram still read +0.42 green. Shrinking, but green.

Net result: +$15.62 (+1.11%) in 14 minutes on 10 shares. Waiting for the MACD to turn red before selling would have ridden that winner all the way back below the $140.86 entry into a loss. The trailing stop banked it near the top; the laggard never blinked.

The rule: Enter on confluence, exit on the leader. Never ask the slow indicator for permission to leave a scalp — the “confirmation” you’re waiting for is your profit leaving.

9. Discipline lessons (earned, not read)

• Cut losers as fast as you bank winners. On a 3× ETF, a loser you hold is what wrecks a day. Both losses on day one were cut in under 5 minutes.
• One clean move beats a pile of churn. Day one: the single well-timed early short (+$71.64) was the day; the other five trades roughly washed after two small losses. The money is in catching one real trend with conviction, not over-trading noise.
• The spread is a bigger tax at high speed. On an $8 scalp, a ~$0.20–0.40 bid/ask is eating a real slice. Your hit rate has to stay high. (This is why the bots use 30-min: fewer trips, less cumulative friction.)
• Don’t branch to a ticker you don’t have a feel for. Trying TNA (didn’t know its rhythm) gave a bit back; bailing in 5 minutes was the right instinct. Trade where your edge is.
• Your stop’s quantity must match your share count. Day 2: a 10-share long got a stop set for 1 share — 90% of the position rode naked. It only ended fine because I was watching and dumped the other 9 by hand. A stop covering a tenth of your position isn’t a seatbelt, it’s a sticker. Check the quantity every time.
• A signal firing is not a trade — it’s a request for your attention. A bot takes every ping; a person filters. Before acting on a confirmed break, check three things: is there runway (not 8 minutes to the bell), has it already moved (are you chasing a knife that’s done), and is it running into support/resistance (shorting into a level that keeps bouncing is the trap). Day 2’s EOD short fired clean and was still a pass on all three counts.
• Don’t confuse a good outcome with a good process. Day 2’s last trade was a correct read that paid 80 cents — right direction, but no juice left, upside-down reward-to-risk, six minutes to the close. It won, yet it was a thin wiggle, not a setup to repeat. Money made on a bad process is a loan the market calls back later.
• Protect a good day – when you hit your goal, STOP. Day 3 (2026-07-17): peaked at +$98 (the goal, hit by 11 AM), kept trading, and gave nearly all of it back on stopless chases – ended +$2.77. The split says everything: 10 disciplined trades made +$135, three no-stop chases lost −$131. The stop – and knowing when you’re done – was the whole difference. Giving back a good day hurts worse than a red day, because you had it in your hand.

10. Pattern Day Trader rule — 2026 update

The old $25,000 PDT minimum is gone. The SEC approved eliminating it on 2026-04-14; it took effect 2026-06-04 (amendment to FINRA Rule 4210). The “pattern day trader” designation and the 3-day-trades-per-5-days restriction were scrapped, replaced by a real-time intraday-margin framework.

• No day-trade count limit anymore.
• The only floor now is the standard $2,000 margin-account minimum.
• The number to watch is keeping account equity above $2,000 — not trade frequency.
• Brokers can phase this in through 2027-10-20, but Robinhood has already adopted it.

Sources: FINRA Notice 26-10 · Schwab · NerdWallet

Short-sale rejections on volatile days (learned 2026-07-20): a short sale needs the broker to LEND you the shares first. On big-volatility SOXL days (the +8% gap → −6% waterfall session), Robinhood’s lendable inventory runs out or their risk desk flips the symbol to no-new-shorts — then every “market short sell” is Rejected regardless of size (10 and 15 both bounced), while buys and normal sells work fine. It is not your buying power and there is nothing to fix; borrow usually returns when the tape calms. Rule: if the short side is blocked, the answer is “no trade” — never improvise a different way to be short. Expect this on exactly the days shorting feels most tempting. (The “Dec 31, 1969” date in RH’s rejection email is their epoch-timestamp bug — cosmetic.)

11. Day 1 scorecard (2026-07-15, first manual session)

SOXL was down ~11% (semis selloff). This scorecard originally logged +$78 across 6 trades and called it a clean green day. It wasn’t a lie — it was the day frozen at its high-water mark. When the P&L-calendar logging bug was fixed on 2026-07-16, the raw Robinhood fills revealed the full session: 16 round-trips, -$139.63. The morning was disciplined; the afternoon was the real lesson.

The arc — every SOXL round-trip, with the running total:

Plus one TNA long (72.18 → 71.61, -$14.25), for a true day of -$139.63 across 16 round-trips — 5 wins, 11 losses.

The lesson (§9, paid in full): trades 1–5 built +$92 in under an hour — a clean, disciplined morning. Then ten more round-trips of churn gave back $217 and buried the day at -$139.63. The old scorecard didn’t deceive on purpose; it stopped writing at the peak, the way memory does. This is §9’s “one clean move beats a pile of churn” in the raw — and it’s why Day 2’s rule (bank the one good trade, then stop) matters so much. The money was always this red; the broken ledger was just hiding it behind blank rows.

12. Day 2 scorecard (2026-07-16, second manual session)

SOXL was deep in the red (down ~15% on the semis rout) and chopped inside a tight range most of the session. Three trades — and the lessons mattered far more than the +$1.27.

Record: 2 wins, 1 loss, +$1.27 — essentially flat. But flat is a win here: the day included two rookie mistakes (chased a breakout, fumbled a stop) and the rules caught both. Trade 1 was the clean one — full confluence entry, and a trailing stop banked it at the top on the momentum roll (the worked example in §8). Beginners blow up on a day like this; breaking even is the skill showing.

Day 2’s technical lesson — failed breaks (traps) and the closed-bar fix:

A breakout is only real if a bar closes beyond the line. Trade 2 fired when price spiked through the 30-min high intrabar to $141.63 — then instantly collapsed back into the range. That’s a bull trap: a failed breakout that sucks buyers in at the top. Trade 3 was its mirror — a bear trap: price closed below the low, then immediately jumped back above it, trapping the shorts (fading it for the bounce was the correct read).

The fix that dodges both: wait for a 5-minute bar to actually CLOSE beyond the level — don’t chase the wick. A forming bar is a rumor — it flickers green↔red every tick (watched one flip live: a “3rd green bar” turned red mid-sentence). A closed bar is the news. Up to five minutes slower, but it filters out the spikes that reverse. The break-watcher was rebuilt mid-session to demand a closed-bar break for exactly this reason.

The through-line of Day 2: every red mistake was survivable because a rule — a stop, a cut, a filter — was standing behind it. The winners came from waiting for confluence; the losses came from jumping early. Same lesson as Day 1, paid for a second time.

13. Day 3 scorecard (2026-07-17, third manual session – the stop lesson)

The most violent tape yet: SOXL gapped down ~14% at the open on a Texas Instruments earnings bomb ($117.80 low), staged a full V-recovery back to green ($144), then rolled over again. Thirteen round-trips, both directions, read beautifully. Peaked at +$98.37 – basically the $100/day goal, all from discipline – then gave nearly all of it back in the last 90 minutes. Final: +$2.77.

The whole day in two numbers:

Every winner had a stop. Every loss had none. The stop – just the stop – was the entire difference between a +$135 day and a +$3 day. Trade 12 is the lesson: a naked short held 21 minutes as it ran $5 against it, then panic-covered at the top of the bounce. A buy-stop at $134 caps that at −$21; without one it was −$78. That one missing stop cost $57 by itself.

Two rules, seared in with real money:

1. When you hit your goal, STOP. The peak was +$98 at 11:12 AM – goal hit, done. Every trade after that was give-back.
2. The stop goes in the same second as the entry – especially when chasing. A chase with a stop is a small, defined risk. A chase without one is the account-killer.

The trap to never fall for: when a stopless loss reverts in your favor after you bail (SOXL dropped right back to $134 minutes after the $137.83 cover), the lesson is NOT “I should have held.” It’s “I should have had a stop.” “It came back” is luck – the next one keeps running and wipes you out. The stop removes the guessing entirely; that is its only job.

Still green (+$2.77), still light-years from Wednesday’s −$139.63. The skill is proven – +$135 of it, both directions, on the wildest tape of the week. The fix is the simplest thing in trading: never skip the stop. This was the day that stopped being words and became muscle.

14. Day 4 scorecard (2026-07-20, the gap round-trip – “the controls got me”)

SOXL gapped +8% overnight, ran to +10%, waterfalled −6% intraday, V-bounced, and by the close had given back essentially the whole gap – the second full round-trip in three sessions (Friday’s TXN crash was the mirror). The decade’s finding held again: gap moves fade; the middle of the day is a donation box.

Ledger truth (canonical, from raw RH fills): −$37.44 · 29 round trips · 20W/8L (71% win rate, red day).

The three: −$26.55 (6:50 top-buy, pre-stop era), −$48.97 (the 145.90 15-lot whose sell-stop rejected twice and never got replaced), −$56.10 (20 shares held 44 min through a $2.81 slide, stopless, while distracted by work). One autopsy, three bodies: a position sliding with no working stop. The reads went 20-for-29; the mechanics lost the day.

What worked (milestones, all first-evers):

• The watcher’s first full live loop: BEAR ping 7:55 (confirmed closed-bar break), BULL pings 8:30/8:40/9:00 – the 8:40 signal (momentum crossing ZERO) was entered within a minute, stop-managed, and banked. Signal → human → seatbelt → machine exit.
• Trailing-stop craft: from “where are the stops?” at 6:45 to seven-step manual trails and native trailing-stop orders by 10:00. Multiple exits executed by resting orders, including one while away from the screen (−$3.40 instead of a −$26 repeat).
• Live order-type fluency: stop vs limit vs stop-limit vs trailing, learned in-session (with tuition: a “buy@139.50 limit 141” ticket that was really “pay up to 141”).
• Self-honesty: the hidden −$26.55 was self-discovered and self-reported; the day’s full ledger was requested by the trader, not extracted by the coach.

Mechanics failure catalog (why the Order-Ticket Trainer now exists): 2 shorts rejected (no borrow, vol day – see §10), sell-stops rejected in the typing race x3, Qty=1 fat-finger x2, one stop believed-placed that never existed, one marketable-limit misread. Six distinct failure modes, all now reproduced in ~/Projects/chart_trainer/order_sim.html where they cost nothing.

The new protocol (self-imposed by the trader – the best kind):

1. Max 5 orders/day (≈ 2 trades + a trail edit) until the gate is passed.

2. Entries only on a referee ping or pre-placed tripwire – no market clicks.

3. The stop is part of the entry; the seatbelt indicator never flashes red; every order verified in Open Orders (“the click is not the truth – the account is”).

4. Gate back to full trading: three consecutive Order-Sim sessions with zero mechanics errors, seatbelt never red, ≤5 orders each.

5. Session windows: first ~2 hours + the EOD window. The morning trader was +$41 by 9:08; the next four hours gave back −$79. Session length, not the market, took the day.

6. The sizing ladder (self-authored 2026-07-21, on SPY as the training vehicle): 1–2 shares → 3–4 shares (after one clean day) → 5 shares (after a clean week + the sim gate) → 10 shares (after another clean week). Rungs are earned by clean days, not by calendar; any breaker day drops a rung and the climb is re-earned. Endpoint math: 10 SPY shares × $1.00 trail = $10 risk/trade = exactly one breaker unit. Two companion rules learned the same morning: every trade is a two-order dance (RH has no brackets – entry fills, THEN protection, within seconds; a tripwire that fires unattended = a naked position, so entry orders only rest while at the screen), and trail ≈ 1.5× the instrument’s typical 5-min bar range (SPY ≈ $0.75–1.00; the trail is “peak giveback,” not “loss from entry” – risk is controlled by SHARES, not by strangling the trail).

   THE COMPLETE RUNG CARD (clarified 2026-07-22 after a good-faith rules dispute – the order budget existed in the protocol but not in the ladder table the trader consults; a rule not in the player’s rulebook isn’t broken in bad faith):

   Dimension	Protocol-day rule
   Size	rung maximum (going smaller is always fine)
   Orders	5 PLACED orders/day = ~2 trades (entry + protection each) + one adjustment. Auto-fires of resting stops/trails do NOT count; every order YOU submit (incl. cancels-and-replaces’ new order) does
   Entries	referee ping or pre-placed tripwire, two-key checked
   Exits	the trail decides, OR named signals spoken before the click (3-signal set; mirror for shorts)
   Breaker	first red ≥ $10 = session over
   Vehicle	SPY (+SH for downside, same rules)
   Clean day	all of the above + zero naked minutes + Open Orders empty when flat
   Practice days (unlimited reps at 1–2 shares) exist only when DECLARED before the
   open – never retroactively.

   Overnight-room rules (written 2026-07-22, cost −$6.02 to learn): stops and trails SLEEP outside regular hours — for longs too. NOTHING protects a carried overnight position; the only seatbelt in that room is being FLAT. Therefore: flatten before stepping away, no exceptions (“work called and I forgot” is the guaranteed failure mode); phone timer at entry; no overnight positions on on-call nights. Day structure that survives a working life: 6:30–8:00 AM = active window (full attention, protocol trades); after 8:00 = long-only (SPY up / SH down) so every position carries a resting sell-trail while work has your eyes.

   CARD v2 — THE TWO-ACCOUNT SYSTEM (signed 2026-07-23, supersedes the rung card above): the discovery that broke the deadlock was thinkorswim paperMoney — live data, pro order types (incl. OCO brackets RH lacks), zero dollars. The rep-hunger that kept blowing the order budget (13 placed orders vs 5, three days running) now has a free venue, so the real account becomes a sniper rifle:

   Account	Role	Limit
   Robinhood (real)	sniper	3 round-trips/day (~7 placed orders; replacements count, auto-fires don’t)
   thinkorswim paper	gym	unlimited — live tape, all experiments, bracket practice
   Real-trade requirements unchanged: ping/tripwire entries, protection in seconds,
   trail = 1.5× instrument breath (SPY $0.75–1.00, SH $0.05–0.07), qty checked,
   licensed exits, breaker at −$10, flat+empty before stepping away, overnight room
   = limits only and never carried. Vehicles (real): SPY + SH; QQQ/PSQ paper-only
   until formally added. Ladder v2: promotion now requires a clean AND GREEN
   day (user’s addition); a clean-but-red day HOLDS the rung (honest tuition never
   demotes — it just doesn’t promote); breaker or rule-break days drop a rung. The
   Week-2 gate accepts 3 clean paper sessions from either the Order-Sim or TOS,
   played under protocol rules (≤5 orders, zero mechanics errors, seatbelt never red).

Perspective: −$37 bought the system’s first end-to-end day, six mechanics lessons, a purpose-built simulator, and a self-authored risk protocol. Wednesday’s stopless lesson cost −$139.63 and taught one thing. This one cost a quarter of that and taught everything else.

15. Day 5 scorecard (2026-07-21, first ladder day – “green, tiny, protected, ended by choice”)

First day of the sizing ladder (SPY, 1-2 shares, declared practice day). Ledger truth: +$0.92 net · 28 round trips · 18W/10L · worst trade −$1.26.

The headline number is the tail: Monday’s worst trade was −$56.10 (naked, 44 minutes). Today’s was −$1.26, capped by a resting trail in seconds — a 44× improvement in damage control in one day. Zero naked minutes across 28 trips.

The experiment held again, all day: every licensed action (referee-ping entries, tripwire entries, three-signal exits, exhaustion covers, trail-managed rides) netted green; the three biggest losses (−$1.13 off-vehicle SOXL short against a bull ping, −$1.26 countertrend chase, −$0.86 mumble-fade) were all freelance countertrend twitches. The edge is measurable and it lives entirely inside the license.

What got built today:

• Watcher re-pointed to SPY (spy-watcher / watcher, symbol now an argument) — removing the SOXL temptation at the source. Referee and plan now aim at one ticker.
• The two-key system: referee ping (key 1) + named-dials freshness check (key 2). Passes on stale/scrape pings are licensed decisions. User vetoed the 0.05% conviction filter — during training, every ping is a decision rep; the human IS the filter (comment in soxl_rh_watcher.py preserves the quiet option).
• Licensed-exit standard v2: exits by trail, or by three NAMED signals (momentum through zero + red/green flip + cross), spoken before the click; mirrored for shorts (momentum bottoming & hooking up = cover). “Momentum slowing” while positive is a tick narrated, not a signal. Grade at click-time, never by what price did afterward.
• Confirmation-lag truth: on grind days a closed-bar referee certifies push ENDS (5 pings, 5 push-ends, margins 3-41c). The tool’s edge concentrates on thrust days; grinders pay HOLD and punish activity. “10 minutes ago was perfect” is hindsight’s siren – live, those moments read identical to the mumbles.
• RH mechanics codified: no OCO (resting order = always the protection; targets = price alert + manual), cancel-before-sell two-step, after-hours = limits only (market/stop orders queue and meet the morning gap), orphaned entry-trails after a manual close = unattended entry orders — Open Orders must read EMPTY when flat.
• VWAP added (annotate-only trial): a LOCATION dial — the day’s volume-weighted fair price. Above = buyers in control. Nearly every freelance loss today was a short taken ABOVE VWAP on an up day. Three days of annotation to earn its seat.

Ladder status: Rung 1 complete and clean (one asterisk: the pre-agreement SOXL short). Tomorrow = Rung 2: 3-4 SPY shares under PROTOCOL rules — 5 orders max. The unlimited-rep phase is over; the selectivity phase begins.

16. The racecar decision (2026-07-24 – Week 1 capstone)

Ten days after the first manual trade, with the account GREEN on the morning (+$0.38, protected, ended by choice), the trader ran the honest math and chose reality:

• “1% a day” on $3,400 compounds to ~$41,700 in a year – a 12x. The best edge six years of data produced is RSI2 dip-buy at 28% CAGR; the best buy-and-hold (SMH) is 41.5% CAGR. Those are 0.10-0.14% per day. The dream target was 7-10x beyond the best verified edge on the books.
• The reframe that matters: 41.5% CAGR is not the consolation prize – it beats nearly every professional fund in existence. The “boring” plan IS the ambitious plan.
• The self-diagnosis, verbatim: “you get that high, that endorphin rush of I won, I won… I can’t race a racecar before learning how to drive correctly.” The rush is the product retail is sold; itemizing its cost is the exit door.

The decision: real-money day trading PAUSED by choice – “paper for a long while.” Not quitting; sequencing:

• Core capital deployed per the Finding-10 plan (SMH now, SGOV parking, reserve waits for the RSI2 dip signal, $999 free-margin carry in SGOV under the $1k line with the two rules: margin lives ONLY in SGOV, checked monthly).
• Skill development moves to the free venues: TOS paper (live data, brackets), the Order-Sim, the watcher’s pings as observation reps, the gate (0/3).
• Re-entry condition for real day trading: a sustained paper track record of positive expectancy under full Card-v2 protocol. The ladder waits where it was left.

Week 1 totals: ~-$75 net tuition across 10 days, ~110 round trips, six distinct mechanics failure modes found and fixed, a complete risk architecture built (stops -> trails -> brackets -> breakers -> budgets -> two-account system), and one trader who did the arithmetic most people avoid for a decade. Cheapest trading education on record.

17. The trend-morning exception (signed 2026-07-28)

The paper-phase re-affirmation, with one negotiated, WRITTEN carve-out – born the morning SOXL crashed -19% and five small with-trend real shorts banked +$4.88 (~0.14% of the account: exactly one day of the evidence pace, which is the point).

Real-money day trades are allowed ONLY when ALL of these hold:

1. A trend morning: a big directional day (gap/waterfall/ramp), in the ACTIVE WINDOW only (6:30-8:00 AM MST). No midday, no afternoon – the session-windows finding stands: mornings pay momentum, middays collect donations.
2. WITH the trend only. VWAP-side agreement + momentum/MACD aligned. No counter-trend, no bounce-catching, no “it has to turn.”
3. 2-3 shares max, every trade protected per Card v2, max 5 trades.
4. Breaker: first red >= $10 ends it; any red SESSION suspends the exception until re-earned by a clean week. The privilege is revocable by its own ledger.
5. Same-day self-report. The books see everything anyway.

Everything else stays paper (TOS gym unlimited), and the gate (3 clean sessions) still governs any fuller return. Rationale: an honest small allowance kept beats a clean rule broken – the 5-order card taught that; quiet violations train rule-breaking, negotiated exceptions preserve the rulebook’s authority.

18. The Schwab sandbox (opened 2026-08-03)

§17 lasted two days as written. The ledger audit (2026-07-30) showed the real activity had outgrown it – 25 trades on Jul 29, sizes to 10-13 shares, a 12:58 PM entry – while staying GREEN (~+$35/wk, ~+1% of account, scorecard reconciled against the canonical ledger). Per §17’s own logic (a kept rule beats an outgrown one), the renegotiation happened by STRUCTURE instead of by clause: the day-trading impulse moved to a ring-fenced ~$225 cash account at Schwab (Individual *912).

The deal:

1. Blast radius is the account. Worst case ever = lunch money. Robinhood (718316607) stays core capital: SMH/SGOV plan, bots, no manual scalps.
2. Cash account mechanics replace the trade cap. The real limit is SETTLED cash (T+1). Structure multi-trade mornings as separate settled slugs (e.g. 3 x 1 share), never re-trade same-day proceeds – a good-faith violation is a strike, 3 strikes = 90-day lockdown. Correction (2026-08-03): “no PDT in a cash account” is no longer a reason for anything – PDT was eliminated everywhere on 2026-06-04 (see §10), so a margin account has no day-trade count limit either. What the cash account still genuinely buys is the hard blast-radius cap (~$225 is all that can ever be lost) and forced settlement discipline; what it costs is the T+1/GFV constraint, which margin does NOT have. The real functional argument for keeping bracketed scalps at Schwab is OCO support: Schwab has it, Robinhood does not on equities (RH offers trailing stops only).
3. Every trade wears an OCO (first-fit params from the 60-day 9:37-10:30 ET TQQQ backtest: sell limit +0.6%, trailing stop 0.8% – the LIMIT earns, the TRAIL protects; both are median-morning numbers, too tight for V-bottom opens).
4. Gap check before entry: the open-momentum edge lives on gap-UP mornings (58% vs 38% win on gap-downs; flat = coin flip). Flat/down mornings are practice, and should mostly happen in paperMoney.
5. Self-report with fills, same day. Schwab does NOT flow into trade_ledger.csv (fill_logger is Robinhood-only) – this account runs on self-reports, which is itself the discipline being trained.

Session 1 (2026-08-03): +$0.51 (+0.23%), no GFVs, quit green by choice. Lessons bought cheap: OCO leg quantity must MATCH the position (+2 in, -1 protected = one naked share); editing a working trailing stop re-anchors it, and dragging any stop to/above the bid fires it instantly as a market order. Homework before the next live session: ten reps of the full routine in paperMoney (buy -> attach OCO -> modify the stop via the OCO group edit) until boring.

Session 1, the OTHER half – the part the sandbox was built to prevent. The same morning, 18 round trips (36 fills, 6:48-8:49 AM MST) ran in the ROBINHOOD MARGIN account, not the sandbox: SOXL +$5.36, TQQQ -$0.18 = +$5.18. Full day across both accounts: +$5.69. Reconciled from raw fills, flat at the close. Stats: 14W/4L (78% win), best +$4.56, worst -$4.64, avg +$0.29/trade, $8,044 notional churned = 0.064% return on churn.

Three things this says, recorded as behavior rather than intention:

1. Rule 1 of the deal (“Robinhood… no manual scalps”) did not survive day one, including trades placed after §18 was written that morning. Per §17’s own logic – a kept rule beats an outgrown one – the honest move is to RENEGOTIATE §18 to match where scalping actually happens, not to leave a clause the record pretends is being followed. Open question to settle deliberately: is RH margin a sanctioned second scalp venue (bigger size, trailing stops only, no OCO), or is the sandbox the real boundary and this was a lapse?
2. Holding beat scalping ~4:1 on the very same tape. SOXL ran $105.60 -> $116.31 (+10% in 50 minutes). Two shares bought at the first scalp and held to the last exit = +$21.42; the 18 round trips extracted +$5.18 from the move they were sitting inside. This is Finding-family evidence in share form: high win rate, tiny wins, the one big move missed.
3. The shorts were the whole drag. Longs +$7.66, shorts -$2.48 (-$4.64 and -$1.90 were both shorts into a tape running 10% up). Consistent with §8 (confirm to get in) and the gap-UP-only edge in rule 4 above – fading a strong trend morning is the losing side of this strategy.

No compliance issue from the 18 trades: PDT is gone (§10), so trade count is not restricted in a margin account. What replaced it is intraday maintenance margin (25% of long market value); peak exposure here was ~$464, nowhere near a constraint. Note RH scalp fills DO reach trade_ledger.csv via fill_logger.py, so unlike Schwab this half of the day is independently auditable.