the machine room
the engine sat out the nasdaq's worst day in eighteen months with every put blocked, and the puts i switched back on friday come with a rule that keeps them off on monday.
last sunday this column said the overdue count would be zero, or this block would lead again. it leads again
The record file has not changed since . It holds 66 calls: 10 hits, 7 misses, 2 partials, 47 open, a resolved hit rate of 57.9% with a partial counted as half. The one new call () is on Nvidia's third-quarter revenue, due Nov 24. Of the 47 open calls, 17 are now past their own deadlines, up from 12. The five that lapsed this week are CDW's working-capital build (Nov 3), Corsair's memory spread and Insight's fee layer (Nov 5), and Phison's margin and the 512Gb NAND wafer call (Nov 6). One of those has its answer already: the Phison earnings page published on found its own margin call held, and the record's Phison row still reads open. The root cause has not moved: calls go into the record automatically each Wednesday and come out only by hand. Last week's fix, a Sunday sweep that drafts each grade with its source attached, was described here and not built. The fix, made smaller: the sweep goes on the schedule as a job that only drafts, and I sign. Until it runs, this block stays at the top of the column with the count in it.
what it did this week
Nothing, for the third week in four. This time the grade is in what it was kept from doing.
the halt blocked 34 call candidates, and three closes later 8 of the 9 graded had gone the wrong way
There was no trade, so there is no winner. What the halt got right is again on the call side. The governor's weekly grader logs every option candidate that cleared the engine's own entry tests and was stopped only by the halt, then checks what the stock did afterwards, signed in the trade's direction. Counted once per stock, direction and session, it blocked 34 call candidates. The next day 5 of 27 graded had moved the right way, averaging −2.24%. Three closes on, 1 of 9, averaging −10.78%. Lumentum calls flagged near $692 were 22% against three closes later. Why it worked: a call needs the stock to rise inside the option's life. The halt fires because the index is below trend, and this week the index kept falling.
the same grader says the blocked puts were right 9 times in 12, and the best of them were refused on wednesday, the day before the drop
The halt also blocked 26 put candidates, contracts that gain when a stock falls. The next day 14 of 22 graded had moved the right way, averaging +2.79%. Three closes on, 9 of 12, averaging +8.25%. Wednesday's five were graded on Thursday's bar: Micron +20.7%, Marvell +18.6%, Nvidia +10.6%, Tesla +9.4%, Merck flat. Those are stock moves in the put's direction, not option prices, because the grader still archives no option marks. The losing side is real too: Reddit and Shopify puts went the wrong way the next day, and Friday's Nasdaq bounce of about 1% is the move a long put pays for.
Three switches stood in front of every one of these. First the halt, still scoped to block all entries until Friday. Second, the operator put block from Oct 29, the deciding refusal on 138 option rows. Third, a fault found Friday: from Wednesday, the flag that allows live option entries reached the options compute as false on 2 of 14 runs Wednesday, 11 of 14 Thursday and 13 of 14 Friday, though the flags file says true. It fails closed, so it cost nothing this week, and the halt hid it. A fix went in Friday evening; Monday's first options run is its first real test.
the new guard measures how far the index has fallen, when the risk on a put after a crash is what the option costs
What changed on Friday. At 13:37 PT, at my call, live puts came back on, the halt's scope narrowed to long-only (so calls and stock buys stay blocked on tier-3 days and puts may trade), and the per-trade ceiling stayed at $1,500, doubled from $750 on Monday. I added one guard: no new put while the Nasdaq-100's last close is 5% or more below its 50-day line. It fails closed, meaning that when it cannot read the index it refuses.
The guard, run against this week. Each day's entries read the previous close. Monday read −4.50% and Wednesday −4.82%, so both pass; Tuesday read −5.59%, so it refuses; Thursday read −3.75%, a pass; Friday read −9.64%, a refusal. Wednesday's best puts would have got through. Monday opens with the index 8.4% below the line, so the guard refuses every put until the index climbs back to about $680. In a replay of 148 halted runs it refused 22 candidates, and none of them would otherwise have become an entry, so it has never been tested on a real trade.
The root cause. The guard stands in for a real worry: after a crash, options are expensive. Their prices carry implied volatility, the market's fear premium, and after a day like Thursday that premium is high. A single put bought then pays for the fear, and a one-day bounce like Friday's takes it back even if the stock goes nowhere. That is a price problem, not a direction problem, and the sleeve's own record shows it. Its drawdown sits in mid-delta contracts, those with a delta of 0.35 to 0.50, further from the money and more exposed to time decay and fear premium: 3 wins, 7 losses, 5 scratches over 15, −$1,444. Contracts nearer the money are 4–4–2 over 10, +$541.80, and the four puts of the last four weeks are 2–1–1, +$703. A distance rule refuses good and bad trades alike, and it opens again in the conditions where options are cheaper.
The fix. Switch structure at the line instead of switching off. Above the 5% line a near-the-money single put is allowed, as now. Below it a bearish entry goes in only as a put debit spread: buy a put near the money and sell a cheaper one further down. The sold put gives back some of the inflated fear premium, and the spread caps both the cost and the gain. Contract count comes from the dollar loss at the invalidation level, the price that proves the trade wrong, inside the $1,500 ceiling. It runs in paper first against the same blocked put candidates the grader already logs, graded in option premium, with its exit test written down before the first entry.
the lessons it wrote down, and what changes
Last Sunday's five tests, marked first. The scheduler held. The rest did not move, and one of them moved the other way.
the afternoons came back, and the option price archive, the put-spread study and the cash read did not
Last week's tests, marked. (a) Runs through the close on both sleeves, with a session-close row every day: PASS. Every day's options file ends with a run after 15:46 ET. (b) The halt grade prints at least one line in option premium: FAIL. Every halted option row reads "no intraday forward marks archived". (c) The put-spread study has a written test and a paper entry: FAIL. Puts came back live as single contracts instead; the paper review that was meant to earn them back had 0 of 10 episodes counted. (d) Zero $0.00 first-pass cash reads: FAIL. Friday's digest still shows the stock sleeve's paired runs reading $3,900 and then $0.00, and the options runs rotating between three different readings, one of them empty. (e) Any option entry carries its dollar-loss-at-invalidation figure and costs under $750: UNTESTED. There was no entry, and the ceiling is now $1,500.
The lessons it logged. One, on sitting out: after Thursday the no-trade tally reads 27 days where sitting out saved against 25 where it cost, net +33.14R on proxy trades. R is the dollar amount a trade is sized to risk. Last week cost days led 25 to 24. The engine's note: one index-drop day moves the count by one, so the caution is still roughly neutral. Two, on a zero-trade week: it wrote that a halt week with no trades is not evidence the halt helped, because none of the 197 blocked rows could have become a trade with every other gate also shut. That is the honest read of the call-side win above. Three, on the stock gates: the dealer-wall refusals, which skip a stock when a large options position sits too close above the entry, no longer clear break-even. Last week they read 70%; five rows graded this week went 1–4, and the bucket is now 15 wins, 11 losses over 26 decisive, 57.7%, against a 58% break-even. The engine withdrew its proposal to loosen that threshold.
What it adjusted about itself. Nothing new. The ten one-point confidence markdowns stand, and no flag was raised.
prove the entry flag, write the spread route, size by the stop before the bigger ceiling is ever used, and archive option prices
One: prove Friday's flag fix on Monday. The digest counts how many options runs saw entries enabled; the bar is every run. Two: the put debit spread route below the depth line (Section I), in paper, with its test in writing: five episodes, a positive average per episode in premium, and a worst case set in advance. The bar is set before the result, so the result cannot move it. Three: risk-based sizing, third week carried. The ceiling doubled on Monday and sizing did not change, so the next trade can be a bigger copy of Oracle's: $990 of premium with no stop-based size behind it. Contract count comes from the dollar loss at the invalidation level, and the ceiling is only a cap on that. Four: archive the option bid and ask on every blocked, entered or shadowed run, so the halt grade and the put route both grade in premium. From Monday the grader's new reverse review also starts scoring every put the engine actually enters under long-only.
How we'll know it worked next Sunday: (a) entries reach the options compute enabled on every run; (b) any put entry carries its premium, its dollar loss at invalidation and a size derived from it; (c) the halt grade prints at least one line in option premium; (d) the spread study has a written test and at least one paper entry; and (e) zero $0.00 first-pass cash reads. If the depth guard refuses every put all week, the column says so on the first line, with what those puts did.
four live changes in one week, none of them reached by a trade
Most of what moved this week was mine. The halt kept every change from meeting the market.
the ceiling doubled, puts came back, a guard went in, and the same-day option rule changed
| change | what it does | status | what it did this week |
|---|---|---|---|
| Option ceiling $750 → $1,500 Mon Nov 2 | Doubles the most premium one option entry may spend. | live | No entry used it. Sizing below the ceiling is unchanged. |
| Puts on, halt long-only Fri Nov 6 | Lifts the Oct 29 put block; tier-3 days block calls and stock buys but not puts. | live | Arrived after Friday's last entry window. |
| Put depth guard Fri Nov 6 | No new put while the Nasdaq-100 closes 5%+ below its 50-day line. | live, untested | Refuses all puts from Monday's open, with the index 8.4% below. |
| Entry-flag dropout Wed – Fri | The live-entries flag reached the options compute false on most runs. | fix installed, unproven | Hidden by the halt; first test Monday. |
| Same-day options Fri Nov 6 | "Never buy" became "sold by 7:45 am PT"; the order gate refuses such entries from 7:30 am PT. | live at the gate | The engine still proposes nothing under two days to expiry, and no forced-sell path is built. |
Two research reads, both negative. A separate chart-pattern paper sleeve has never closed a trade. Its frozen setup, backtested over 154 trades, averaged −0.03R, because stops sat 5–7% away and every position was flattened the same day, so almost nothing reached a stop or a target. I paused it on Friday until a different stop or hold tests positive. A 16-month backtest of a same-day options trend signal also found no edge, with or without the 7:45 deadline. Last week's worst fault, the afternoon scheduler stall, did not recur.
the three verticals, in brief
The newsletter's demand-and-credit read on semiconductors, cyber, and construction: this week's headline, and the dated checkpoint that would confirm it or crack it.
chips had their second-worst day on record, and micron fell 17%
On Thu Nov 5 the Philadelphia semiconductor index fell about 9%, after a drop that briefly passed 11%; reports of the close run as high as 12% (24/7 Wall St., Grafa). Micron fell 17.46% to $656.39 and Nvidia 9.26% to $181.60. Coverage tied it to the midterm results and fears over AI and data-center rules (Fox Business, Bloomberg). Korea's Kospi fell more than 6% and the Nikkei entered a bear market on Nov 6. The semiconductor fund on the engine's feed closed Thursday 19.9% below its 50-day line. The record's NAND wafer call was due Nov 6, and I could not find TrendForce's November wafer update. The last read, Oct 28, had 512Gb TLC spot down 9.19% to $16.883. The call is open and overdue.
the security names fell with the chips on thursday, with no news of their own
On Thu Nov 5 CrowdStrike fell about 9–10%, Cloudflare about 9% and Okta about 11%; Palo Alto's drop is reported at 8% and 10% in two pieces from the same source (TIKR). None had company news. On the engine's feed the cybersecurity fund closed Thursday at $91.01, 9.6% below its 50-day line. This is the group trading as one block for the second week running, the opposite of the dispersion call. The four-name spread check due Nov 6 was not computed for this column, so it carries.
data-center construction hit a record $96.5 billion rate in september, as total building fell
The Census release on Mon Nov 2 put private data-center construction at a $96.5B annual rate in September, the highest since the series began in 2014 and about 90% above a year earlier (AGC). Total construction fell 1.9% to $2.186T; private non-residential rose 1.1%, mostly on data centers and power. The figure is preliminary. Last week's checkpoint set the bar at up 60% or more for "the build is finding power". It cleared that by 30 points, so the call holds for September. Thursday's selloff was partly a bet that the rules change from here.
And the newsletter, on itself: the record's state leads this column again, because last week's promise did not hold. One thing to add. The construction checkpoint above was met from a public release, and the Phison row can be graded from a page this site published on Friday. The evidence is already in hand. The Phison grade has not been entered in the record.