Yolo-Insider-Probe Playbook (forked scientist)

Generated from LEDGER.json by ledger-stats · as of 2026-10-09 local. Do not hand-edit; prose narrates, this block is the record.

Open positions

None.

Forked from: yolo · Seeded: 2026-07-26 (user-approved fork of the 2026-07-26 session's fork proposal — the first under the disciplined-scientists rule) · A/B protocol: money-loop §3 — run beside unchanged yolo out-of-sample; promote/kill on evidence, never merge back by vibes.

The ONE changed rule (the hypothesis)

Parent yolo's golden signal is RSI < 30 + insider buying, golden cross intact. This fork drops the golden-cross-intact requirement:

Entry: RSI < 30 + material insider buying — regardless of chart damage (death cross / collapse state allowed).

Hypothesis under test: the insider-buying signal alone carries edge on damaged charts, where the discount is biggest. The motivating setup: TSLA 2026-07-24 — RSI 28.1, $205.3M unchanged insider buying, death-crossed, which parent yolo correctly passed on under its rules.

Everything else is INHERITED from yolo — deliberately

• Exits: identical to yolo (lib/exit-rules.ts aliases this book to yolo's rules): hard −20% stop, +20% target, +15% partial-trim watch. A fork changes exactly one rule so the A/B isolates it.
• Sizing: probe units — 2.5% of book ($2,500) per entry, half-again on a second qualifying signal, never more than 5% in one name. Damaged charts earn probe size, not conviction size.
• Universe, cash discipline, and session protocol per research/playbook/CADENCE.md (Tier A daily) and the base rules in research/trading/paper/README.md.
• Disciplined-scientist rule applies here too: this book's written rules are THIS file; an idea outside them is a new fork proposal, not a trade.
• Scope — user call 2026-09-25: INSIDER-ONLY. This book's entry water is the golden-insider leg alone (RSI < 30 + material insider buying, chart damage allowed). Parent yolo's capitulation leg (RSI < 25 + $10B) stays in yolo and is never an entry here (user call 2026-09-25, ops/tasks/TASKS-ENGINE.md #yolo-insider-probe scope row).

Current Regime Read

2026-09-25: the bond rout hit a fresh extreme: the 30-year yield is at its highest since 2004 and mortgages are at 7.45% (CNBC, Bloomberg, September 24). The index doesn't care: SPY $771.35, RSI 55.9, 7.4% above its 200-day; VIXY $16.61, gate open. The damage is rate-shaped. Utilities and real estate are both at RSI 24.1, and equal-weight (36.1) and small caps (34.9) are soft. Financials held up. The insider universe printed zero golden signals for a ninth straight week, and the carried buying list is frozen for a fourth. The one material insider name under 30 (GSHD) is stale and net-sold.

Self-Critique

Flat at $100,472.24, +$472.24 realized, 2 closed, 2 green, cohort n=1. As of 2026-09-25 the desk ruled this book insider-only, which ends the capitulation ambiguity I carried for four sessions. The book is now a clean test of one thing, and that thing is rare. Mostly-cash is the honest expected state. Every near-miss so far has died on missing infrastructure (my own narrow universe, the tradability door, no thesis artifact, stale filings), never on the insider signal itself failing.

Next Session Priors

• No positions. Inherited exits unchanged: −20% hard stop, +20% target, +15% partial-trim watch.
• Read the box receipt first, and ignore its capitulation rows until the box is amended. They are parent yolo's.
• KRMN: RSI 36.8, $35.69, 48.8% below its 200-day. Its September 18 director cluster ($1.04M, three directors) ages daily. If it re-breaks 30 while that buying is still current, the thesis read of the 8-Ks and the short report must exist first, or it's a PASS again.
• DY: RSI 29.7, but the buying is $453K, under the $1M bar. Only matters on a new filing.
• GSHD: no tracking. Stale, net-sold, untradeable, golden cross intact.
• Check recency and net selling on every golden-insider ENTER by hand; the box doesn't.

Session log

Sessions before 2026-08-18 are archived to _archive/sessions/ (07-26, 08-04, 08-13).

2026-08-18 — The Rout Didn't Reach the Gate Either

THINK: One probe, still alive: TSLA is $336.93 versus $313.03 entry, +7.6%, book value $100,190.88. Exit-check (desk-wide) says HOLD — 0 SELL, 0 TRIM across 117 positions. The inherited stop is $250.42, trim-watch $359.98, target $375.64 — nowhere close. Outside the book, today is a genuinely large one-day dislocation: AI-infrastructure names down 5-9% median, 105 names moved ≥7%, but VIX 14.63 and VIXY barely moved (+0.21%). If there was ever a day for a damaged-chart insider-buying name to show up, this reads like the setup — so I checked hard.

ANALYZE: Ran the actual gate, not the vibe. Fresh 2026-08-18 insider-universe (59 names) and tech-insider-buys (42 names) summaries, sorted by RSI. Floor is META at RSI 36.2 — death-crossed, -4.42% today, -9.27% 7d, genuinely damaged, but 6+ points above the gate. RBLX next at 37.6. Every other insider-buying name is RSI 40+. Zero names below 30. TSLA itself — the name that started this whole fork — is RSI 46.2 today, still well clear of its own charter level. The AI-infra rout hit optical/memory/neocloud names hardest, and none of those carry a rated insider-buying signal in these two universes to begin with; the fork's water and the rout's epicenter didn't overlap today.

DECIDE: PASS. Hold TSLA; zero buys, zero sells, zero trims. RSI<30 is the gate and nothing cleared it, rout or no rout. Cash stays $97,500.

REFLECT: Fifth session, same clean zero, and today's is the most informative one yet because the market handed a real dislocation and the gate still didn't fire — this is exactly what a disciplined two-part entry rule should do: not chase "insider buying" and "big red day" into the same trade just because both words showed up on the same tape. Cohort remains n=1. Next session, check RSI first, insider-buying receipt second, same as always.

Trades Executed:

2026-08-20 — Sixth Session, Same Clean Zero

THINK: One probe, still alive: TSLA $345.13 vs $313.03 entry, +10.25%, book value $100,256.37 (paper positions yolo-insider-probe). Cash $97,500. Exit-check already ran desk-wide with --execute today (MRNA/WMT sold elsewhere, not mine). Stop $250.42, trim-watch $359.98, target $375.64 — nowhere close. Tape is a red index day (SPY -0.84%) with gold/crypto/oil bid — not obviously the kind of day that cracks a damaged chart into RSI<30.

ANALYZE: Ran the gate, not the vibe. Fresh 2026-08-20 insider-universe (59 names) and tech-insider-buys (42 names) summaries, sorted by RSI. Floor is AVGO at RSI 36.2, death-crossed but not below 30 (insider-universe.json/tech-insider-buys.json). Next is META at 37.0. Every other name is RSI 40+. Zero names below 30 — same as every session since the 07-26 charter. TSLA itself is RSI 50.4 (mag7.json), well clear of both its own charter level and any re-add signal.

DECIDE: PASS. Hold TSLA; zero buys, zero sells, zero trims. RSI<30 is the gate and nothing cleared it. Cash stays $97,500.

REFLECT: Sixth clean zero. Cohort remains n=1 — the market has not handed this experiment a second qualifying signal since the charter, and forcing one to generate data would defeat the point of the A/B. Next session, check RSI first, insider-buying receipt second, same as always.

Trades Executed:

2026-08-31 — Half Off the Table on the Probe

THINK: One position, one job. TSLA from $313.03 on 2026-07-26, now $367.95. +17.54%. The A/B fork exists to test whether insider-flagged entries beat the parent book's water, and this one has run.

ANALYZE: exit-check flagged it [partial-profit] and it's the only TRIM on the entire 152-position sweep. TSLA closed +5.5% today, tied for the best day on the focus list — but it's still 26.2% off its 52-week high with a three-month number of −11.5%. A strong week inside a chart that hasn't repaired. That's a profit to bank, not a trend to ride.

DECIDE: Sold 3.993 shares at $367.95. Realized +$219.29, +17.54%. 3.99 shares still open — the probe keeps its exposure, the fork keeps its data point, and the gain is no longer theoretical.

REFLECT: First realized P&L for this book and it's a win, which proves nothing yet at n=1. What it does establish is that the partial-profit rule fires cleanly on a fork book without me arguing with it. Next session prior: watch the remaining half against the 26.2% drawdown — if TSLA can't hold above the entry on a red tape, the rest comes off.

Trades Executed:

2026-09-01 — Seventh Clean Zero, And The Runner Does The Work

OBSERVE: The fork is one position and a lot of cash: TSLA 3.993455 shares at $313.03, now $356.09, +13.76%, and $98,969.22 uninvested. Half came off the table last session; this is the runner. Entry gate, both universes (insider-universe + tech-insider-buys, 70 names between them): zero under RSI 30. The lowest prints are EOSE 35.2 (death-crossed, collapse), FSLR 35.9 (death-crossed, collapse), PINS 36.5, ENPH 38.4. Last session's near-miss was META at 36.2; tonight META is not even the low. Nothing is within five points of the gate. That matters for what this fork is actually testing. My one changed rule drops the golden-cross requirement — it does not touch the RSI < 30 leg. EOSE and FSLR are precisely the damaged charts this book exists to be allowed to buy, and they still fail, because they fail on RSI, not on chart damage. Seven sessions in, the binding constraint on this fork has never once been the rule I forked. That is the finding, and it is a slightly embarrassing one: the A/B cannot generate a cohort because the shared leg is the bottleneck.

DECIDE: PASS on entries. No action on TSLA — $356.09 sits $3.89 below the +15% trim-watch at $359.98, well under the +20% target at $375.64, and nowhere near the $250.42 stop. No freelancing between the levels, as written.

REFLECT: n is still 1. Seven sessions, one probe, half-trimmed. If the RSI leg keeps gating every candidate before the insider leg is ever consulted, this fork will never produce evidence about the hypothesis it was seeded to test — and that is a design problem with the fork, not a market problem. Worth raising at the next money-loop pass: either the probe waits however long it takes for a genuine RSI<30 insider name, or the fork is mis-specified and should have loosened the RSI leg instead of the cross leg. I am not going to quietly widen it myself — that would destroy the only thing the A/B is good for.

Trades Executed: none — PASS.

2026-09-04 — NO RSI, NO SEND

Written 2026-09-07 local for the September 4 settled session; run 2026-09-07-repair-closure-eow.

OBSERVE: Flat book. Cash and total value are $100,472.24; realized P&L is +$472.24. The scoped exit pass checked zero positions, with zero fills, trims, or actions, and the September 4 book breaker is clear. SPY closed $770.19 at RSI 55.6, QQQ $718.96 at RSI 53.6, and VIXY $17.00 at RSI 31.5.

RULE CHECK: All 59 insider-universe names and all 42 tech-insider-buys names have sound September 4 summaries: 70 distinct names, zero missing and zero below RSI 30. PINS is nearest at 31.9, then CRDO 34.2, AVGO 38.0, and DDOG 38.7. The fork permits chart damage; it still requires RSI <30 plus current material insider buying. The five buying-table names also miss RSI: TSLA 50.9, CRSP 50.7, ASAN 45.2, INTC 51.7, TSM 56.8. Repeated buying dollars are not a newly verified filing.

DECIDE: PASS. Zero new exposure. Every name in the adopted hunting ground fails the RSI gate, so current insider materiality, thin-thesis admission, VIX buy permission, sizing, tradeability, and shared-plan admission were not reached or claimed clear. The digest's seven sub-30 names sit outside these two insider hunting lists; this review does not claim there is no insider buying elsewhere. No fill means no decision/action handoff is outstanding.

NEXT: Cash is a valid result. Watch PINS and CRDO for an actual RSI breach, then verify current material insider buying and the company thesis before pursuing a plan. TSLA is no longer open, so the old runner levels and discretionary exit prior are historical. The inherited coded exits remain -20% hard stop, +20% target, +15% TRIM, sector-death SELL, and -15% near-stop WATCH. No rule change earned.

Trades Executed:

2026-09-08 — PINS Keeps Creeping Toward 30 And Still Isn't There

OBSERVE: Flat book, $100,472.24 cash, +$472.24 realized lifetime. All 101 names across the two adopted hunting lists have sound September 8 summaries.

RULE CHECK: Nearest name is PINS at RSI 30.6 — tighter than 09-04's 31.9, but the rule reads "<30," not "close to 30." Then CRDO 33.6, ISRG 35.1, LLY 36.9, DDOG 37.6. Zero qualify.

DECIDE: PASS. The fork isolates exactly one variable — chart damage allowed — and that variable never gets tested if RSI never crosses the line in the first place.

NEXT: Keep watching PINS. If it prints sub-30 alongside a fresh buying filing, that's the first real test of the hypothesis since TSLA.

Trades Executed: none — PASS.

2026-09-11 — The Box Saw ONON Three Nights Ago And I Was Looking At The Wrong List

THINK: Flat book, $100,472.24, all cash, +$472.24 realized lifetime, two closed rows both green. Scoped exit-check checked zero positions — nothing to sell, nothing to trim, nothing to watch, breaker clear inside both budgets. So this session is entries or it is nothing. Tape backdrop: SPY $764.29 / RSI 50.3, QQQ $714.88 / RSI 50.8, both still above their 200-day with double-digit years (summaries/holdings-etfs.json, etf-ideas.json). VIXY $17.30 / RSI 39.6 (summaries/market-etfs.json) — the inherited regime gate is open, no half-sizing, no defensive posture. Nothing about the index level says hide.

ANALYZE: Ran the box first, per the September 8 rule — persona-entries yolo-insider-probe, receipt kept at data/state/persona-entries/yolo-insider-probe/2026-09-11.json. It screened 1,143 rows down to 35 candidates and printed 6 ENTER, 2 UNDECIDED, 27 PASS.

Here is the thing: not one of the six ENTERs came from my rule. All six fired on capitulation — RSI<25 plus a $10B market cap — which is parent yolo's leg and consults no insider data whatsoever. CLX 23.0, LHX 24.9, HRL 23.2, CASY 19.0, TJX 17.8, RCL 24.4. My leg, golden-insider, printed zero. Every PASS row says the same three words: insider value missing.

I checked that independently instead of taking the box's word for it. Nineteen buy-side Form 4 clusters at or above $1M in research/api-data/filing-watch.json (dataDate 2026-09-11), crossed against settled RSI: the lowest is ONON at 32.3, then GSHD 34.6, BABA 36.6, BX 38.2. Nothing under 30. Both adopted hunting lists say the same — the insider scan reports a sixth straight cycle with no sub-30 name, closest PINS at 30.8, and the buying table narrowed from five names to three (ASAN, INTC, TSM). PINS, the name my own last two priors told me to watch, went the WRONG way: 30.6 on September 8, 30.8 now. Still not there.

So what about the six the box did hand me? Every one of them dies at thesis admission, and I want the reasons on the record because they are not the same reason. TJX: the desk's own retail roundup this session rates it ⚠️ Avoid and says in plain words "genuinely broken, not a bounce setup," with the 30-day damage resting on nothing newer than a three-week-old downgrade — 1Y -10.52%, alpha -26.74, 18.3% under its 200-day. LHX: no live campaign covers it, the actions digest calls it the board's clearest broken-trend name in a full collapse regime, 1Y -12.5%, alpha -28.72, 21.7% under its 200-day. RCL: zero corpus artifacts at all, broken on every window, 1Y -25.72%. CLX: zero corpus artifacts at all, 1Y -29.46%, alpha -45.68. HRL is the interesting one — it IS a key ticker in the livestock-disease lane, but that lane's own August 28 entry files HRL's weakness as "company-specific earnings news, not thesis evidence either way," names ELAN as the clean expression, and holds. A perspective that names a ticker while explicitly declining to read it as evidence is not a receipt to buy it. And CASY never got that far: paper check says tradeable false, not in any watchlist — the door is shut before the argument starts.

Look at what these six have in common. Five of six are negative on the year with negative alpha AND below their 200-day. These are not violent pullbacks inside intact trends — the setup I am supposed to want. These are broken long trends printing oversold on the way down. CASY is the single exception (1Y +11.1%, 2Y +64.84%), and it is exactly the one I am not allowed to touch.

The two UNDECIDED rows, TQQQ and SOXL, the box could not decide because nothing on disk supplies prior-bar RSI for a reclaim cross. That is mine to judge, so I judged it: TQQQ sits +14.6% above its 200-day with golden cross true and 1Y +48.52%; SOXL +10.2% above, golden cross true, 1Y +317.62%. You cannot reclaim a line you never lost. RSI 49 and 48.3 are not crossing 40 from below. No reclaim, no entry.

And now the part I do not enjoy writing. I pulled the three prior box receipts. On 2026-09-08, 09-09 and 09-10 the box printed ONON as ENTER on golden-insider — RSI 28.8, then 27.4, then 27.5, a $3,987,581 insider buy cluster, golden cross false, sized at the 2.5% probe unit. Golden cross FALSE. That is the exact setup this fork was seeded to test — a damaged chart the parent would refuse and I am allowed to take — and it is the first one since TSLA on July 26. My September 8 session wrote "Zero qualify," because I was reading only insider-universe and tech-insider-buys. ONON lives on the retail watchlist. The box screens the whole corpus; my self-narrowed hunting ground does not. Seven sessions of me writing "the RSI leg is the bottleneck, the fork can never generate a cohort" and it turns out at least part of the bottleneck was my own universe. Today ONON is RSI 32.3 and the window is shut.

Two honest caveats before I flog myself too hard. That cluster is dated 2026-08-17 to 2026-08-17 and the dollar figure is byte-identical across all three receipts — it was three weeks stale by September 8, and my own September 4 prior says do not carry old buying totals as fresh filings. And ONON has no company profile, no perspective, no current artifact — nothing in the corpus would have admitted it, and there is no news file for it at all, so I cannot even check whether its drawdown had a cause. The likely honest outcome is nominated-not-taken, not a fill I lost. But I did not reach that judgment on September 8. I never saw the row.

DECIDE: PASS. Zero buys, zero sells, zero trims, zero shorts. Cash stays $100,472.24. Every capitulation nominee fails thesis admission on its own evidence; CASY fails the tradability door before that; both leveraged rows fail a reclaim that is not happening; my own insider leg has no candidate at any price today. No fill means no decision or action handoff is outstanding.

One thing I am NOT doing: taking a capitulation name just because the box lit it up. The box is the nominator, the receipt is the admitter, and today there is no receipt behind any of them. Note also that my written rules say everything except the golden-cross waiver is inherited from yolo, which means capitulation is arguably my water too — but eight sessions of practice in this book have only ever operated the insider gate on two lists. That scope question did not need resolving today because admission killed all six regardless. It will need resolving the first time a capitulation name shows up with a real thesis behind it, and I am raising it rather than quietly deciding it in my own favor.

REFLECT: Ninth session, ninth clean zero on the hypothesis leg — but for the first time the zero is not the whole story. The finding is that my hunting ground has been narrower than my box, and a qualifying setup walked past the window three nights running while I was looking at two watchlists. That is a coverage bug in my practice, not a market problem and not a rule problem. The box receipt is the fix and it is already on disk; I just have to actually read it, which is what today's session did for the first time.

Trades Executed:

Evolved Rules — proposed, NOT adopted this session

No rule change taken. One scope clarification requested, because it is a practice question and not mine to settle alone:

1. Hunting ground. This book's sessions have read only insider-universe and tech-insider-buys. The coded box screens every settled summary row and is what surfaced ONON. Proposal: the box receipt's candidate set becomes this book's universe of record, and the two insider lists become a cross-check rather than the boundary. Evidence: ONON ENTER on 2026-09-08/09/10 on golden-insider at RSI 28.8/27.4/27.5, missed entirely by the 09-08 session.
2. Is capitulation this book's water? The written playbook inherits everything but the golden-cross waiver, which reads as yes; eight sessions of practice read as no. Requesting a ruling before a capitulation name arrives with a receipt. Not deciding it in the direction that permits a trade.

Neither goes into the active rules today. A fork changes one rule at a time, and this book's whole value is that its one rule stayed isolated.

2026-09-18 — TENTH CLEAN ZERO, AND THE ONE NAME THAT QUALIFIED ISN'T BUYABLE

OBSERVE: Flat book, tenth session. $100,472.24, all cash, +$472.24 realized across two closed rows, both green. The scoped exit pass checked zero positions — nothing to sell, trim or watch — and the breaker is clear inside both budgets. So this session is entries or it is nothing. Backdrop: the Fed raised rates a quarter point Wednesday, the first increase in three years, and Friday was quarterly expiry. SPY $761.69 / RSI 49.2 and QQQ $721.45 / RSI 55.9, both above their 200-day with double-digit years. VIXY $17.09 / RSI 39.4 and the desk's vol gate at 14.81, CALM — the inherited regime gate is fully open, no half-size, no defensive posture. Nothing about the index says hide.

RULE CHECK: Box first, per the 09-11 correction that this book's universe of record is the box receipt and not my two self-chosen watchlists. It screened 1,113 rows to 50 candidates: 6 ENTER, 2 UNDECIDED, 42 PASS, and once again not one ENTER came from my rule — all six fired on the inherited capitulation leg, which consults no insider data whatsoever. Every PASS row reads "insider value missing."

I cross-checked the insider leg by hand rather than trusting that silence, exactly as on 09-11. The Form 4 file carries 35 buy-side clusters for the desk date. Joined against settled RSI, the lowest is KRMN at 31.0, then UUUU 33.6, ONON 34.7, GSHD 35.0. Zero names under RSI 30 have insider buying. Tenth clean zero on the hypothesis leg.

KRMN is the closest call this book has had since the charter and it is worth writing down properly. Its cluster is dated 2026-09-18 — same day, three directors, $1,036,485. That is not a stale carry, which is the exact failure my 09-04 prior told me to check for and the reason the ONON row in September was probably never a real fill. The chart is the shape this fork exists to be allowed to buy: golden cross false, 49.2% below its 200-day, −45.38% on the year with alpha −60.39, in a collapse regime. My rule waives the chart damage. It does not waive RSI, and 31.0 is not under 30. And then the door closes anyway: the tradability check says not in any watchlist, tradeable false. Same wall CASY hit on 09-11. If KRMN cracks 30 on Monday I cannot act on it, and that is a coverage problem, not a rule problem.

The six capitulation ENTERs — CRS 23.8, CLX 21.5, TXRH 21.0, CASY 23.8, RCL 22.6, DTE 24.8 — all die, and the reasons are worth separating. The scope question I raised on 09-11 is still unresolved: my playbook inherits everything but the golden-cross waiver, which reads as capitulation being my water too, while ten sessions of practice read as no. I said then I would not settle it in the direction that permits a trade, and I'm not settling it tonight either. Underneath that: not one of the six has a corpus artifact admitting ownership, and the two that have artifacts argue against — Royal Caribbean is rated ❌ don't-buy by the travel desk with its 200-day breach widening from 9.7% to 14.5% in a week, and the retail roundup says of Casey's cohort that not one name in it is a deep pullback in an intact trend. And the part that actually matters for THIS book: three of the six — CRS $54.4M, CASY $29.1M, TXRH $3.6M — carry insider SELL clusters. A fork whose entire hypothesis is that the insider signal carries edge does not buy names insiders are selling. CRS has a fourth problem: parent yolo has held it since August and is down 17.11% on it. A fork that buys its parent's open position makes two books one bet and turns the scoreboard into a lie.

The leveraged rows need their own paragraph because one of them actually fired. TQQQ is dead on Rule 10 — five kept receipts show RSI 45.3 → 42.4 → 42.5 → 51.4 → 54.1 with price 11.6% to 16.5% above its 200-day the whole way; no cross from below, no reclaim of a line never lost. SOXL is different: the reclaim completed this week. My own receipts show RSI crossing 40 from below on 09-15 (39.5 → 40.2) and price reclaiming its 200-day on 09-17 (−6.8% → +2.6%), and today it sits +10.1% above with RSI 51.7. The inherited Rule 10 fired. I am still passing: the leverage lane is the same unresolved scope question, this book has never once operated it, the trigger day was Wednesday rather than today, and the fund closed +7.71% on the session — buying a 3x daily-reset fund the day after a 7.7% pop is the exact chase the parent named when it took this rule on 09-04. But I am logging it as a cost, not a shrug: this is the first time the unresolved scope question has cost this book a completed, rule-legal trigger. That makes it a live desk decision instead of a filing-cabinet question.

DECIDE: PASS. Zero buys, zero sells, zero trims. Cash stays $100,472.24. My own leg has no candidate at any price; every capitulation nominee fails the unresolved scope question, thesis admission, and — for three of them — my own insider evidence; TQQQ has no trigger; SOXL has one I am not authorized by practice to take. No fill means no decision or action handoff is outstanding.

NEXT: KRMN is the watch, and the ask is a watchlist, not a rule. Fresh same-day director buying on a chart 49.2% under its 200-day, one point above the gate, and untradeable at the door. Retire ONON (34.7, moving away, stale cluster plus a new small sell) and PINS (33.3, moving away, $6.09M sell cluster) from the priors — both went the wrong way and both now carry insider selling. One prior closes clean: ONON's news has been pulled since (September 11), so the standing request from last session is answered. Inherited exits remain −20% hard stop, +20% target, +15% partial-trim watch. No rule change earned.

Trades Executed: none — PASS.

Evolved Rules — NO CHANGE. The fork's one variable stays isolated. Three items escalated, none adopted: (1) scope question #1, carried from 09-11 and still unresolved — is capitulation this book's water? (2) scope question #2, NEW and now load-bearing — is the LEVERAGE lane this book's water? The inherited Rule 10 completed a legal two-leg trigger on SOXL this week and this book passed on practice grounds alone; first time the open question cost a real trigger. (3) Coverage ask, not a rule change — KRMN is untradeable at the door: fresh same-day director buying, RSI 31.0, golden cross false, 49.2% below its 200-day, the cleanest hypothesis-shaped candidate since the July charter, and it sits in no watchlist. Second time (CASY, 09-11) the door closed before the argument. If the desk wants this A/B to generate a cohort, the box's candidate universe and the tradable universe have to be the same universe.

2026-09-23 — MY RULE FIRED TWICE, AND I STILL CAN'T WRITE A THESIS FOR EITHER

THINK: Flat book, eleventh session. $100,472.24, all cash, +$472.24 realized across two closed rows, both green. Tonight's scoped exit pass checked zero positions — no fills, no trims, no watches — and the breaker is clear inside both budgets. So it is entries or nothing. Backdrop, per the desk's September 23 summaries and the day's actions digest: the rate shock escalated — the five-year Treasury breached 5% for the first time since 2007 and the long-bond fund closed at a record low (Bloomberg, CNBC, September 22–23), a week after the Fed's quarter-point hike. The index shrugged: SPY $767.81 / RSI 53.0, still 7.0% above its 200-day with a +15.77% year; VIXY $16.79 / RSI 38.7, no hedging bid. The damage is rate-shaped and it is everywhere underneath — utilities in capitulation (XLU RSI 23.5), nine of eleven sector funds down on the month.

ANALYZE: Box first, as the 09-11 correction requires. It screened 1,124 rows to 75 candidates and printed 17 ENTER, 3 UNDECIDED, 55 PASS — and for the first time since the July charter, two of the ENTERs came from MY rule, not the inherited capitulation leg.

KRMN — the cleanest hypothesis-shaped candidate this book has ever had, and it clears every gate but one. RSI 26.8 (first went under 30 on September 22 at 27.4 — no session of mine was written that night). Insider leg: the Form 4 file carries a September 18 director cluster, $1,036,485, three directors, open-market purchases — five sessions old, current by my own standard, not a carry. I read the three filings myself: one director did ~$1.01M of it; the other two bought token lots ($10K and $19K). Material by the rule's letter, concentrated in one buyer. Chart damage: total, which is exactly what this fork waives — golden cross false, 52.8% below its 200-day, −51.9% on the year, alpha −67.67, collapse regime, 72.1% off its high. And the door that slammed on 09-18 is open: the tradability check now says tradeable, in the watching list. The coverage ask I filed last week got answered.

Then admission, and it dies there. I searched the whole corpus for one artifact that argues for owning Karman: no company profile, no deep dive, no campaign (drone-defense doesn't carry it), no scan discussion tonight. What IS on file: two 8-K captures (August 28, September 16) that are bare pointers with no content read into them; a September 9 social search that is a promo video and a chatbot description; an August earnings record (a small beat); and the September 9 market brief, which says Karman dropped 11.3% on a short-seller's cash-flow questions and a new CFO. That is the only substantive company-level sentence in the corpus, and it is a bear case nobody has answered. The news receipt was pulled September 9 (newest article August 7), so it covers neither the director buying nor the −35.66% month; stock-events holds only earnings rows. "A ticker mention is not a receipt" — and a Form 4 is my timing signal, not my thesis. If the insider leg could admit its own entry, this book would be a pure insider-follower and the thesis rule would be decoration. PASS on admission. It is the right answer and I hate it.

GSHD — rule-shaped on paper, fails on currency before anything else. RSI 23.5. But its buy cluster runs July 28 – August 4, seven weeks old, one of the two buyers is a fund rather than an officer, and the same window carries a $17.6M sell cluster from three 10% owners that outweighs it. My 09-04 prior forbids carrying old buying totals as fresh filings; this is exactly that, and net selling besides. It is also untradeable — not in any real watchlist. Dead twice.

The fifteen capitulation ENTERs — NEE, CRS, BKNG, LHX, HRL, D, DTE, EXC, PEG, BAC, FIS, CVS, JBS, TXRH, RCL — all fire on the inherited RSI<25 + $10B leg, which consults no insider data. Whether that leg is my water is the scope question I raised on 09-11 and it is still an open user question. A setup that needs it answered is a documented PASS, not a ruling I get to make in my own favor. Same for TQQQ and SOXL (UNDECIDED on the reclaim cross): the leverage scope question is equally open — and on the merits neither is a reclaim tonight anyway, both 25–28% above their 200-day with RSI near 60, nothing lost to take back. XLU is UNDECIDED because the capitulation leg can't read a market cap for a fund and there's no insider leg for a fund; it lives in capitulation water regardless. PASS.

The adopted insider lists agree with all of this: tonight's insider scan and tech insider-buys scan show the buying list frozen for a third straight week at the same five carried names (TSLA, CRSP, ASAN, INTC, TSM), golden signal zero for an eighth week. The action this week is off-list, which is why the box is my universe of record.

DECIDE: PASS. Zero buys, zero sells, zero trims. Cash stays $100,472.24. KRMN clears RSI, current material insider buying, and tradability, and fails thesis admission — no corpus artifact supports owning it and the one company-level fact on file is an unanswered short report. GSHD fails currency, is net-sold by 10% holders, and is untradeable. Every capitulation and leverage row waits on the two open scope questions. No fill, so no decision or action handoff is outstanding.

REFLECT: Eleven sessions and the diagnosis changed again. For ten I blamed the RSI leg, then my own narrow universe, then the tradability door. Tonight all three cleared on one name and the binding constraint turned out to be the research layer: the box can find a hypothesis-shaped setup the desk has never written a word about except a short-seller's. That's not a rule problem and I won't propose loosening admission — a fork that buys on a Form 4 alone is testing something else. It's a coverage ask: if the desk wants this A/B to produce a cohort, a KRMN-class name needs a read of its 8-Ks and the short report before the gate opens, not after. No rule change earned.

Phase 5a: exit-check on this book — 0 positions checked, 0 SELL / 0 TRIM / 0 WATCH, breaker clear; nothing to decide.

Trades Executed: none — PASS.

Evolved Rules — NO CHANGE. Carried, not adopted: (1) capitulation scope — open user question since 09-11; (2) leverage scope — open user question since 09-18; (3) the KRMN coverage ask is ANSWERED (tradeable since this week), replaced by a research ask: a thesis read of Karman (the August and September 8-Ks, the short report, the new CFO) so the next qualifying night ends in a real admission decision instead of an empty file.

2026-09-25 — THE DESK RULED: I'M AN INSIDER BOOK, AND THE INSIDER LEG HANDED ME ONE NAME I CAN'T BUY

THINK: Flat book, twelfth session. $100,472.24, all cash, +$472.24 realized across two closed rows, both green. Tonight's exit pass has nothing to check: breaker clear inside both the 5% daily and 25% drawdown budgets. So it's entries or nothing.

Backdrop from Friday's settled summaries:

• Index level, nothing says hide. SPY $771.35, RSI 55.9, 7.4% above its 200-day, +17.22% on the year. QQQ $744.50, RSI 64.7.
• Regime gate open. VIXY $16.61, RSI 37.6: no half-size, no defensive posture.
• Underneath, the bond rout. The 30-year yield closed at its highest since 2004 and mortgages hit 7.45% (CNBC, Bloomberg, September 24).
• Rate-sensitive funds washed out together. Utilities and real estate both sit at RSI 24.1. Equal-weight is at RSI 36.1 and small caps at 34.9.
• The adopted insider universe still has nothing. Golden signal zero for a ninth straight week, and the carried buying list is frozen for a fourth.

And the big one: the desk RULED. The probe tests the insider gate ALONE; the capitulation leg stays with parent yolo. The scope question I carried since 09-11 is closed, and it's closed in the direction I refused to decide for myself.

ANALYZE: Box first, as always. It screened 1,125 rows to 69 candidates and printed 14 ENTER, 5 UNDECIDED, 50 PASS.

• Thirteen of the fourteen ENTERs are capitulation: NEE, CRS, BCE, MCD, GD, LHX, HRL, D, DTE, DUK, EXC, PEG, TXRH. As of today that's officially parent yolo's water, not mine. The box hasn't caught up to the ruling yet; that fix belongs to the desk, not me. They aren't my candidates, so they don't even reach thesis admission.
• One ENTER came from my leg: GSHD, at RSI 28 with an $11.45M insider value.

I cross-checked the insider leg by hand anyway. The Form 4 file carries 35 buy-side clusters, 18 of them at or above $1M. Joined against settled RSI:

• GSHD is the only name under 30 with material buying.
• DY sits at 29.7, but its cluster is $453K, under the $1M bar.
• Next up: DTM at 31.2 ($218K), then BX 31.5 and ET 31.8.
• KRMN, last session's near-hit, bounced out: RSI 36.8, $35.69. Its September 18 director cluster is a week old now. Not tonight.

GSHD dies FOUR ways, and the last one is the one that matters to a scientist:

1. Currency. The buying runs July 28 to August 4, seven weeks back. One of the two buyers is a fund with no officer role. My 09-04 prior forbids carrying old buying totals as fresh filings.
2. Net sold. A $17.57M sell cluster from three 10% owners, July 30 to August 5, is bigger than the buy.
3. The door. Tradeable false, in no watchlist.
4. It doesn't test my rule. GSHD's golden cross is still TRUE. That makes it a setup parent yolo's own golden signal would see too. Even a fill would be zero evidence for the one rule this fork exists to isolate.

And there's no thesis artifact behind it anyway: an earnings calendar row, some monster-scan mentions and one brief mention. BLOCKED.

The UNDECIDEDs:

• TQQQ and SOXL are both far above their 200-day: +26.4% and +31.8%, RSI 63.4 and 61.6. You can't reclaim a line you never lost. They're also outside an insider-only book as I read the ruling.
• XLU and XLRE are funds with no insider leg, sitting squarely in capitulation water, which is now yolo's by ruling.
• HLSQ's summary row is stale as of 09-23 and it carries no Form 4 buy cluster, so it couldn't qualify on my leg anyway.

DECIDE: PASS. Zero buys, zero sells, zero trims, and no proposals to the lead. Cash stays $100,472.24.

• The capitulation nominees are out of scope by ruling.
• The leveraged rows have no reclaim and are out of scope as I read the ruling.
• My one insider nominee is stale, net-sold, untradeable and doesn't exercise the forked rule.

No fill, so no decision or action handoff is outstanding.

REFLECT: Twelve sessions, cohort still n=1: TSLA, the July charter. The ruling makes this book honest in a way it wasn't: it's now a PURE test of "insider buying on a broken chart." Most nights that means cash, because that setup is rare: nine straight weeks of zero golden signals in the adopted universe, two hits on the box's wider screen on 09-23 and one tonight. That's not a problem to engineer around. A rare signal is the thing under test, and a book that sits in cash waiting for it is doing its job.

Two observations, neither adopted:

1. The box reads insider VALUE with no recency window and no netting against selling. That's why GSHD printed ENTER two sessions running on a seven-week-old buy that big holders outsold. I'm catching it by hand every time. If the desk wants the box to carry that judgment, it's a box-rule question for the algo lane. I'm not loosening or tightening anything myself.
2. KRMN is still the only fork-differentiating candidate this book has seen since ONON (golden cross false, current director buying). It still has no thesis artifact, just a short-seller's unanswered cash-flow questions on file. The research ask stands.

No rule change earned.

Trades Executed: none. PASS; the lead replaces these with actual outcomes.

Evolved Rules — NO CHANGE.

• Scope question #1 (capitulation): RESOLVED by the user call of 2026-09-25, insider-only. The mandate and box amendments belong to the lead.
• Scope question #2 (leverage): read as covered by "insider gate alone", pending the lead's confirmation.
• Carried: the KRMN thesis-read request, and the box insider-recency observation above.

2026-10-02 — THE BOX HANDED ME BLACKSTONE BUYING BLACKSTONE, AND IT STILL HAS ITS GOLDEN CROSS

THINK: Flat book, thirteenth session. $100,472.24, all cash, +$472.24 realized across two closed rows, both green. Tonight's exit pass has nothing to check, no trims, no watches. Breaker clear, 0% daily, 0% drawdown. So it's entries or nothing.

Friday's settled tape:

• Index doesn't care. SPY $769.64, RSI 54.5, 6.8% above its 200-day. QQQ $749.58, RSI 65.4.
• Gate open. VIXY $16.72, RSI 41.
• Underneath, the bond complex is washed out end to end. TLT RSI 23.5, IEF 24.6, BND 24.5, LQD 22.7, and HYG all the way down at 17.7. Real estate (XLRE 21.7) and financials (XLF 25.9) are oversold too.
• Breadth is soft. Equal-weight 37.1, small caps 39.6.
• The adopted insider universe is still dead. Golden signals zero for a tenth straight week. The buying list is the same five names (TSLA, CRSP, ASAN, INTC, TSM) for a fifth straight week. The lowest RSI anywhere in it is NFLX at 31.1, and that one has selling.

ANALYZE: Box first. 1,252 rows screened to 78 candidates: 10 ENTER, 11 UNDECIDED, 57 PASS.

• Nine of the ten ENTERs are capitulation: BCE, MCD, RTX, NOC, GD, ENB, BAC, FIS, TXRH. That's parent yolo's water by the September 25 ruling, and the box still hasn't caught up. Not my candidates. They don't even reach admission.
• ONE ENTER came from my leg: BX. RSI 25.5, insider value $20.19M. On the box's face, the cleanest-looking golden-insider print in weeks. Then I read the clusters by hand, like every time, and it falls apart.

BX dies FOUR ways:

1. Currency. The buy cluster ran August 12 to 24, six weeks back. The 09-04 prior says old totals are not fresh filings.
2. Who's buying. Almost all of the $20M is "Blackstone Holdings IV L.P." filing as a 10% owner. That's Blackstone's own holding entity buying Blackstone. Most likely that's an affiliate transaction, not an insider putting money behind a conviction, though I haven't read the filing text. The only human buyer is a single director.
3. Net sold, by a LOT. August 4 to 13: $1.16B out across seven filings, from two Blackstone holding entities plus the chief accounting, legal and administrative officers. The buy is a rounding error next to it.
4. It doesn't test my rule. Golden cross TRUE, with the 50-day still above the 200-day. Parent yolo would see this exact setup. A fill would be ZERO evidence for "insider buying on a broken chart," the one thing this book exists to measure.

No thesis artifact anyway. The only company-level line on file is the September 24 brief noting all ten alt managers fell together. BLOCKED.

Cross-check the leg by hand. 18 material buy clusters on file, joined against settled RSI. BX is the only one under 30. UUUU sits at 30.5 on a July buy. GSHD has bounced to 33.3 and is still stale and net-sold. BABA is at 35.6. KRMN, my one real fork-shaped near-miss, is at RSI 38.5, $34.14, 50.4% under its 200-day, and its September 18 director cluster is two weeks old. DY is at 33.9 with $453K, still under the bar.

UNDECIDEDs:

• TQQQ and SOXL: 27.5% and 39.3% above the 200-day, RSI around 65. You can't reclaim a line you never lost. Outside an insider-only book anyway.
• HLSQ, HYG, LQD, TLT, IEF, FXE, VNQ, XLRE, BND: funds with no insider leg, sitting in capitulation water. That's yolo's.

DECIDE: PASS. Zero buys, zero sells, zero trims, no proposals to the lead. Cash stays $100,472.24.

• Capitulation nominees: out of scope by ruling.
• Leveraged rows: no reclaim, and out of scope.
• My one insider nominee: stale, likely an affiliate buy, outweighed fifty-to-one by insider and affiliate selling, golden cross intact, and no thesis on file.

No fill, so no decision or action handoff is outstanding.

REFLECT: Thirteen sessions, cohort still n=1 (TSLA, the July charter). Third week running the box has handed my leg one name, and third week running the box's insider read was the problem, not the signal. GSHD twice on a seven-week-old fund buy outsold by 10% holders. Now BX on an entity buying its own parent's stock, outsold by over a billion. The box takes the newest buy cluster's dollar value. It doesn't look at recency, doesn't net against selling, and doesn't check whether the "insider" is an affiliated entity. I catch it by hand every time and I'll keep doing that. Fixing it is a box-rule question for the algo lane, not a loosening or tightening I get to make. Mostly-cash is still the honest state for a rare signal.

Phase 5a: exit-check on this book: 0 positions, 0 SELL / 0 TRIM / 0 WATCH, breaker clear. Nothing to decide.

Trades Executed: none, PASS.

Evolved Rules: NO CHANGE.

• Carried, not adopted:
  • (1) The box should drop its capitulation leg for this book per the 09-25 ruling. That's the lead's job, still pending.
  • (2) The box's insider read needs a recency window, netting against selling, and an affiliate-entity check. Three straight false ENTERs: GSHD twice, BX. This is an algo-lane question.
  • (3) The KRMN thesis-read request stands: the 8-Ks, the short report, the new CFO.

2026-10-09 — FOURTEENTH ZERO: THE BOX ONLY READS CLUSTERS, AND THE ONE BROKEN-CHART "BUY" IT COULDN'T SEE STILL WOULDN'T HAVE CLEARED

THINK: Flat book, fourteenth session. $100,472.24, all cash, +$472.24 realized across two closed rows, both green. The exit pass has nothing to check: zero positions, breaker clear inside both the 5% daily and 25% drawdown budgets. So it's entries or nothing.

Friday's settled tape:

• The index doesn't care. SPY $778.57, RSI 59.9, 7.7% above its 200-day. QQQ $751.27, RSI 60.9. Bloomberg's October 9 read: big tech's rally ran into a fourth week as AI-bubble fears faded.
• Gate open. VIXY $16.3, RSI 39.3.
• Underneath, damage moved around. Chips cooled after OpenAI told investors its annualized revenue was about $50 billion (CNBC, October 8). The carriers broke after SpaceX agreed to buy 800 MHz spectrum (CNBC, October 8), and Verizon had its worst day since 2002 (CNBC, October 9). VZ closed $41.65 at RSI 21.8, 10.5% under its 200-day.
• Bonds climbed out of capitulation without reclaiming anything. TLT RSI 34.8, HYG 31.4. Small caps are soft: IWM RSI 38.5.
• The adopted insider universe is still dead.
  • Zero golden signals for an ELEVENTH straight week.
  • The buying list is the same five names for a sixth week: TSLA +$221.0M, CRSP +$26.6M, ASAN +$4.5M, INTC +$2.4M, TSM +$1.5M. The lowest RSI among them is INTC at 43.4.
  • The only name in the universe under 30 is EOSE (28.7), and it's a −$25.1M seller.

ANALYZE: Box first. It screened 1,262 rows down to 57 candidates: 3 ENTER, 3 UNDECIDED, 51 PASS.

• All three ENTERs are capitulation: VZ (RSI 21.8), BCE (17.1), BMRN (22.9). That's parent yolo's water by the September 25 ruling, and the box still screens it. Not my candidates.
• Every PASS is a setup fail, and every sub-30 row reads "insider value missing". Zero golden-insider rows. That's the first night in three sessions the box didn't hand my leg a false ENTER (GSHD, then BX).
• UNDECIDED:
  • TQQQ and SOXL: 26.4% and 16.1% above their 200-days, so there's no reclaim to make. Leverage is outside an insider-only book anyway.
  • 005380.KS: a stale foreign row with no insider leg.

Then I cross-checked the leg by hand, like every time.

• Clusters. 35 buy clusters on file, 16 at or above $1M.
  • The lowest RSI of ANY cluster is 32.4: DGXX and BW, both around $110K–$115K, well under the bar.
  • The lowest material cluster is GSHD at 34.1. It's still stale, still net-sold, and still has its golden cross.
  • KRMN: RSI 37.9, $32.59, 52% under its 200-day, and its September 18 director cluster is three weeks old.
  • DY: 36.2 on $453K.
• Then the individual filings, which the box doesn't read at all. Two names under 30 have a buy-side filing:
  • HRL, RSI 25.0. A director bought 10,000 shares in the open market on October 5, for $201K. It's real, fresh and coded as a purchase, but it's a fifth of the $1M bar. Not material.
  • ALMS, RSI 26.3, $7.16. It's 67.5% under its 200-day and 77.16% off its high, with golden cross FALSE. There's one director filing dated September 4: $10.73M across 986,721 shares. On its face, that's the most fork-shaped print since KRMN. It dies FOUR ways:
    1. Currency. It's five weeks old. The 09-04 prior says old totals aren't fresh filings.
    2. Direction. The single filing carries BOTH purchase and sale codes. I haven't read the net direction, and I haven't read whose money it is.
    3. No thesis. There's no profile and no deep dive. News was last pulled September 3, and the only item is a Q2 earnings line. Nothing on file explains the collapse. A screen nominates; a thesis admits.
    4. The door. ALMS sits only in the auto-generated discovery list, which the paper CLI doesn't count. It's untradeable, the same way KRMN was in September.
• VZ's seven Form 4s. These were filed October 9, dated October 8, the CEO's among them, and the digest says to read them before calling anything insider buying. VZ isn't my water, but I checked the filing record anyway:
  • The same seven officers have filed together every two weeks since July 16: July 16, July 30, August 13, August 27, September 10, September 24, October 8. The filing reader leaves the transaction codes blank on all of them.
  • That's a schedule, not a reaction to spectrum news that landed after the October 8 close. Most likely these are routine plan recordings. I haven't read the filing text, so that's a read, not a fact.
  • Separately, the head of the business unit has sold 1,100 shares a week since August.
  • And VZ's golden cross is still TRUE, so it wouldn't test my rule even with real buying behind it.

DECIDE: PASS. Zero buys, zero sells, zero trims, no proposals to the lead. Cash stays $100,472.24.

• Capitulation ENTERs (VZ, BCE, BMRN): out of scope by ruling.
• Leveraged and foreign UNDECIDEDs: out of scope, and there's nothing to reclaim.
• Golden-insider leg: no name qualifies. HRL is under the bar. ALMS is stale, carries both purchase and sale codes, has no thesis, and is untradeable.

No fill, so no decision or action handoff is outstanding.

REFLECT: Fourteen sessions, and the cohort is still n=1 (TSLA, the July charter).

Tonight the box was quiet, and wrong in a new way. It only reads buy CLUSTERS, so a single insider's big purchase never reaches it. ALMS wouldn't have cleared anyway. But a fresh, purchase-coded, single-director $1M+ buy on a broken chart is EXACTLY the hypothesis this book exists to test, and right now the box can't see one. That joins the recency, netting and affiliate gaps already filed with the algo lane. It's a box-rule question, not a loosening or tightening I get to make.

From now on my hand join covers individual filings, not just clusters. Every near-miss in fourteen sessions has died on staleness, netting, affiliates, a missing thesis or the door. None has died because the signal itself failed. Mostly-cash is still the honest state for a rare signal.

Phase 5a outcome: exit-check on this book: 0 positions, 0 SELL / 0 TRIM / 0 WATCH, breaker clear (inside 5% daily-loss and 25% drawdown budgets). Nothing to decide.

Trades Executed: none, PASS.

Evolved Rules: NO CHANGE.

• Carried, not adopted:
  • (1) Drop the box's capitulation leg per the 09-25 ruling. Filed with the algo lane, still pending: three straight nights of capitulation ENTERs.
  • (2) The box's insider read has no recency window, no netting against selling and no affiliate check. Filed. NEW facet tonight: single-filer purchases outside clusters are invisible to it (ALMS).
  • (3) The KRMN thesis read still stands: the 8-Ks, the short report, the new CFO.