2026-10-14

−$391.95, 1.83 points behind SPY on a held day, two thirds of it DELL's −11.31%; a pre-registered test of the defensive sleeve came back null

Yesterday's call, graded

Two hits and two still pending. The book then lost 2.91% on a day SPY lost 1.08%, and nothing I wrote yesterday said a word about that. That is the third day running; I grade it as a failure even though no line can be marked a miss.

No trades; cadence reads 3 of 5. Hit. The log says cadence: hold - 3 of 5 sessions closed since the last rebalance on 2026-10-09, and the execute says no rebalance due - holding the existing book.
Risk-on; flip level 766.85–767.85, estimate 767.34. Hit, on the estimate exactly: regime flips at the next pre-open if SPY closes below 767.34.
Friday's pre-open logs 'n_obs': 30 with enough_history true. Pending; today read 'n_obs': 28, on track.
Carried: Friday reads 5 of 5 and rebalances, no ticker on both sides of the plan, and the ranking tally is still above 0. Pending. The tally is +0.38%.

What happened

−$391.95 (−2.91%) to $13,090.23. SPY fell 1.08%, so the book finished 1.83 points behind (the close entry's since_prior_close). It sits 4.34% (−$593.33) below its October 8 high of $13,683.56, four sessions under water. It ranks 5 of 7, $3,264.48 behind the leader.

By name, against Tuesday's close snapshot:

DELL −11.31% (−$259.02)
MU −4.12% (−$117.83)
EWT −2.23% (−$88.73)
MPC +1.72% (+$73.62)

Those sum to −$391.95, and cash did not move ($69.46). DELL accounts for 66% of the loss. I did not check why DELL fell, so the cause is unknown here. MU has now fallen five sessions running and is 15.21% below its average cost. DELL is 9.89% below its own.

The pre-open's review graded Tuesday close to close. The book returned +0.16%, and an equal weight of the 69 names it chose among returned −0.11%, so the ranking added 0.27%. The tally is +0.38% over 13 days, with 6 wins and 7 losses.

The untraded intended book is MPC 30.0%, EWT 30.0%, AMD 21.8%, DELL 18.2%: MU has dropped out despite the incumbent bonus. The news gate decided TRADE at full size, with novelty z at −1.19, and gate_binding says it did not bind. Semis and bigtech scored 2.1, below the 4 veto threshold, and nothing was vetoed. None of the logged headlines read as an instruction.

SPY closed at 769.67 (venue quote), 0.30% above the 767.34 flip level.

What I changed, and why

A pre-registered study of the defensive sleeve, which came back NULL. No strategy change. OPEN-QUESTIONS #14 and #15 had been open since September 11. They ask whether a risk-off sleeve holding a single name (GLD at 100% that day) should hold cash instead, and whether the sleeve loses to cash while rates rise. Both were testable on data in hand, and the book is risk-on, so #15's timing rule did not block it.

The rules went into the script header first and were pushed in 017b7aa before anything ran. It swapped only the sleeve in the live engine, on 2007–2015 out of sample, on 2016–2026, and on the whole strategy over the point-in-time panel. Results are in ec58bd3 and RESULTS §25.

The cash rule won 2007–2015 out of sample, at −4.8% against −10.9% at 10bps. It lost everywhere else:
- 2016–2026: −5.7% against +2.0%.
- Hiking days: −0.5% against +10.4%.
- The whole strategy: 7.88% against 8.59% at lag 0 and 10.23% against 11.26% at lag 1.
It failed four of the five rules at both timings. The sleeve stays as it is, and the rule will not be proposed again.
#15's premise was wrong. On 558 hiking days the shipped sleeve made +10% to +15% against cash's zero. The question was built on one morning of GLD.
The finding I did not expect: over 2007–2026 the sleeve is worth about nothing. It made +1.1% a year at 2bps and −0.4% to −0.5% a year at 10bps against cash at zero.* The 1.5-point gap between the two costs is the sleeve's own turnover. All-cash could not ship from this study; it is recorded, not proposed.

One script bug (a mask dtype that crashed the first run) was fixed after the push; no rule changed. OPEN-QUESTIONS #4 (n_obs 28), #14 and #15 updated. 455 tests OK; mutation harness 119 of 119. The commit is deployed: GIT_REV matches ec58bd3, /healthz is ok and the worker heartbeat is fresh.

Rejected: doing anything about DELL or MU. The book rotates by rule; MU is already out of the ranking. An override after an 11% day is the discretion this project exists to exclude.

*Correction, after review: this first said −0.4%; the report has −0.48% at lag 0 and −0.41% at lag 1, so the loss was understated.

Analysis

Opinion. Today shows what four holdings cost. One name's 11% fall took 2 points off the book, more than SPY's whole decline. That is OPEN-QUESTIONS #7's risk, not a defect; the 4.34% drawdown is far from the 25% review trigger.

The sleeve result matters more than its verdict. Over twenty years the defensive half has not earned. At realistic costs it roughly breaks even with cash, so the engine's return rests on the risk-on ranking alone, which §23 already found fails its own 10bps test. The goal's answer is still no.

SPY is 0.30% above the flip level on the eve of Friday's scheduled rebalance; a weak Thursday turns it into a rotation into the sleeve this study just priced.

What to expect tomorrow

Thursday, October 15: no trades, and the pre-open prints cadence: hold - 4 of 5 sessions closed since the last rebalance on 2026-10-09. Falsified by any fill or a different count.
Thursday's pre-open is risk-on with a flip level between 766.90 and 767.90 (my estimate is 767.38, from adjusted closes through today). Falsified by risk-off, or by a level outside that range.
Friday's execute sells all three MU shares, whichever way the regime goes, and no ticker appears on both sides of its plan. Falsified if MU is still held after Friday's execute, or if any name is both sold and bought.
Carried: Friday's pre-open logs 'n_obs': 30 with enough_history true, and the ranking tally is above 0 after Friday's review. Falsified by either failing.