2026-09-07 · midday

Labor Day midday, and this is the sixth read of Friday's finished session. The book is byte-identical to this morning's - same 32 contract ids, not one mark moved, NLV cash gross and excess liquidity equal to the cent. Same three structural rails, every runbook trigger clear, the long-leg repair still $18,770. Nothing about the positions changed, so this run went and tested the one claim the morning page leaned on. It leaned on it too hard. This morning's page said the H100 floor had held at $1.99 an hour for eight consecutive three-hourly captures. Two captures have landed since: $2.27 at 08:07, then $1.89 at 11:07. Up 14 percent, then down 17 percent, on a day with no market open and no news. The smoothing argument itself survives that and gets stronger with a number on it: measured across the whole record, the 24-hour average has a coefficient of variation of 11.4 percent against the daily median's 21.1 percent, and it moves 20 percent or more day over day on 1 day in 18 against the daily series' 9 in 19. On the smoothed series not one day in the entire record would have fired a 20 percent fall. So the right reading is not that spot has found a floor - it plainly has not - but that the 20 percent threshold is meaningful on the smoothed series and meaningless on the raw one.

What to do

280P x5 and 300P x8 sit above the $260 long-leg ceiling.
BUY TO CLOSE 5 GOOGL Dec15'28 360 PUT / SELL TO CLOSE 5 GOOGL Dec15'28 280 PUT
one spread, limit, day · ~$18,770 debit ($37.5392 net per spread on this run's marks)
Your playbook: “long leg never above $260 (must sit at or above the $230-250 floor zone, never below $230)”
The same standing order, repriced from this run's own marks: 360P at $68.6067 against 280P at $31.0675 is $37.5392 net, a sixth consecutive identical read. Both legs still carry Friday's closing marks, so an identical reprice is the correct result on a shut book rather than a stuck number. Only the 5x 280P leg is reachable by this order; the 8x 300P leg has no written repair. It is a Tuesday order.

Still over, unchanged

These fire every run until you act on them. No order attached.

rail	now	limit
long_leg_band	280P x5, 300P x8	$230-$260 band	13 contracts outside
net_cash_ratio	-0.065	amber -0.05, red -0.10	amber
gross_leverage	1.741x	amber 1.30x, red 1.50x	red
top2_name_delta_ratio	0.67x	amber 0.70x, red 0.85x	clear
total_delta_ratio	0.94x	amber 1.30x, red 1.50x	clear

Why each one reads the way it does

long_leg_band
The rail the action item above addresses. No leg traded, no mark moved since Friday's close.
net_cash_ratio
Cash -$93,230.12 against NLV $1,428,489.17, both identical to every read since Friday's close.
gross_leverage
Gross $2,487,660.42 over NLV $1,428,489.17, unchanged. Carried caveat: this divides GROSS position value by NLV against thresholds the IPS wrote for delta notional, and the delta rows in this same table read 0.67x and 0.94x, so the red reading overstates true leverage.

delta by name

Delta notional per name as of 2026-09-07: stock at delta one, each option leg at the Black-Scholes delta implied by its own mark. Top two: GOOG, MU.

name	delta notional	of NLV
GOOG	$808,598	56.6%	1543 sh, 16 legs
MU	$142,622	10.0%	525 sh, 2 legs
META	$88,589	6.2%	144 sh, 0 legs
VOO	$87,042	6.1%	123 sh, 0 legs
RDDT	$79,258	5.5%	513 sh, 0 legs
NVDA	$51,002	3.6%	285 sh, 2 legs
INTC	$28,767	2.0%	300 sh, 0 legs
AMZN	$25,830	1.8%	100 sh, 0 legs
BRK B	$20,220	1.4%	40 sh, 0 legs
EQT	$11,034	0.8%	200 sh, 0 legs
TSLA	$353	0.0%	1 sh, 0 legs
DIS	$105	0.0%	1 sh, 0 legs

Everything else clear

Nearest trigger is GOOG - tranche 2 weakness (nearest written level, and what rails.py reads) at $317, 6% away.
TSLA -6.2%, now $352.89. No written rule for it.
MU +5.9%, now $1,014.91. No written rule for it.

All 26 triggers swept, with today's number

price trigger	now	fires at	distance
GOOG - tranche 2 weakness (nearest written level, and what rails.py reads)	$335.72	≤ $317	5.6% down
GOOGL - tranche 3 confirmation	$338.72	≥ $360	6.3% up
GOOGL - tranche 2 weakness	$338.72	≤ $317	6.4% down
GOOG - tranche 3 confirmation	$335.72	≥ $360	7.2% up
GOOGL - thesis break, close regardless of price	$338.72	≤ $280	17.3% down
MU - roll the collar	$1,014.91	≥ $1,250	23.2% up
GOOGL - Rule 8 crash powder / washout buy	$338.72	≤ $260	23.2% down
GOOGL - do not chase strikes past here	$338.72	≥ $430	26.9% up
GOOGL - Rule 6, close every short put	$338.72	≥ $434	28.1% up
GOOGL - long spreads near max	$338.72	≥ $500	47.6% up
GOOGL - shares capped by the 600Cs	$338.72	≥ $600	77.1% up

level trigger	now	fires at	distance
Dec-28 30-delta put IV	35.7%	> 43%	+7.3 pt
GOOGL IV, crash trigger	35.7%	> 50%	+14.3 pt
11x Dec-28 300/400 call spread, mark vs width	42.05%	>= 90%	+48.0 pt
8x Sep-27 300/440 call spread, mark vs width	38.51%	>= 90%	+51.5 pt
10x Dec-28 350/500 call spread, mark vs width	29.29%	>= 90%	+60.7 pt
GOOGL Dec-28 340P extrinsic	$56.49	< $5	$51.49
Excess liquidity / NLV	74.41%	< 40%	+34.4 pt
NLV change on the week	+2.50%	-8% in one week	+10.5 pt
Defined max loss vs program cap	$175,059	$200,000	$24,941 headroom

calendar	date	away
Wash-sale window opens for the 6/05 GOOG loss lots	2026-09-17	10d
Micron fiscal Q4 print (not a runbook trigger row)	2026-09-30	23d
Alphabet Q3 print - scheduled thesis adjudicator	2026-10-28	51d
Reserve deploys at market if no crash by this date	2026-10-29	52d
Earliest date the GPU 90-day tripwire test can evaluate at all	2026-11-16	70d
First GOOG lot vest, collar put leg 9	2027-03-13	187d

Spot prices

	last	day
GOOG	$335.72	-0.9%
GOOGL	$338.72	-1.0%
MU	$1,014.91	+5.9%
AMZN	$258.30	-0.2%
BRK B	$505.50	-0.5%
DIS	$105.18	-1.8%
EQT	$55.17	-0.8%
INTC	$95.89	+4.6%
META	$615.20	+0.7%
NVDA	$229.49	+0.5%
RDDT	$154.50	-1.0%
TSLA	$352.89	-6.2%
VOO	$707.66	-0.4%

The check itself

found this run the smoothed GPU series is measurably quieter than the daily one, and this morning's '$1.99 floor' was not a floor.
This morning's page argued the fix for the noisy daily series already sits in the three-hourly captures, and pointed at eight consecutive captures printing $1.99 or $2.00 as evidence. Two captures have landed since it was written: $2.27 at 08:07 and $1.89 at 11:07, up 14 percent then down 17 percent, on a day with no US market open and no news. So the streak was not a floor and should not have been read as one. The smoothing argument itself is unharmed and now carries a number. Computed over the whole record from gpu-spot-intraday.csv against gpu-spot-daily.csv: the daily median has a coefficient of variation of 21.1 percent, ranges $1.84 to $4.01, and moves 20 percent or more day over day on 9 of 19 transitions. The 24-hour average across days with at least four captures has a coefficient of variation of 11.4 percent, ranges $1.93 to $2.94, and moves 20 percent or more on 1 of 18 transitions, with zero falls of 20 percent or more anywhere in the record. Median day-over-day move is 6.3 percent smoothed against 16.3 percent raw. That is the whole case in one line: on the smoothed series a 20 percent fall would be an event, and on the raw series it is a Tuesday. Nothing changed - the tripwire's definition is a written rule and this check does not rewrite written rules to fit the instrument.
found this run the intraday series inherits the same missing history, so smoothing does not shorten the wait.
gpu-spot-intraday.csv starts 2026-08-19, one day after the daily series. Moving the tripwire onto the 24-hour average would make the threshold meaningful but would not make it evaluable any sooner: the 90-day lookback is starved on both series until roughly 2026-11-16 either way. Worth stating plainly so the two problems are not confused - one is a measurement problem that a better statistic fixes today, the other is a history problem that only time fixes.
will happen again the cron has no trading-day guard - six runs against one session.
The schedule is '35 6,12 * * *' with no timezone anchor and no market-calendar test. Saturday twice, Sunday twice, Labor Day twice: six runs, all against Friday 2026-09-04's single finished session. Five of the six could only confirm the first. Reported since 09-05 and unchanged, because the cron is Gabriel's to edit.
will happen again deploy.sh still needs an arm64 node pinned ahead of it on PATH.
/usr/local/bin/node is a Mach-O x86_64 binary and sits ahead of /opt/homebrew/bin, while wrangler's cached workerd is darwin-arm64 only. Worked around per-run again with PATH=/opt/homebrew/bin:$PATH, for a fourth consecutive run. Still not fixed inside deploy.sh.
will happen again markets.dottl.xyz has no Access gate.
An unauthenticated request from this machine returns the full page - NLV, every position, every strike and every quantity - to anyone with the URL. book-check.dottl.xyz is gated behind Cloudflare Access; this one never has been. Not changed unilaterally: gating it is a call about who you want to be able to open the link.

Rails that could not be evaluated

The GPU tripwire is defined at a 20 percent fall on a series that moves 20 percent or more on 47 percent of days. Measured this run, the 24-hour average would cut that to 6 percent of days with no 20 percent fall anywhere in the record - but the threshold is a written rule and this check does not change it.
The GPU tripwire cannot evaluate at all until roughly 2026-11-16 on either the daily or the smoothed series - both start mid-August and the lookback is 90 days.
The SemiAnalysis composite is five months stale at 2026-04 $2.82 and the runbook's tripwire prose is defined on it, while later months are lock-icon paywalled.
2026-08-29 is still a hole in gpu-spot-daily.csv - 21 rows across 22 calendar days.
stressed_excess_liquidity - needs IBKR Risk Navigator (-30% price, +50% IV) with the TWS gateway up. Never yet measured, and the IPS amber's 7-day stress re-run has never run.
short_dec28_340p_extrinsic is still listed unmeasured in rails.json because it needs a live Dec-2028 chain pull, though the runbook's own version of the row was swept by hand this run and is clear at $56.49 against the $5 trigger.
gross_leverage divides GROSS position value by NLV against thresholds the IPS wrote for delta notional, so the red reading overstates true leverage. Unchanged and unresolved.
rails.py reads GOOGL levels off the GOOG quote, which is why the tranche-2 and tranche-3 distances differ between the two tickers in this report.
rails.py keys option-leg matching on contract_description with no contract_id fallback.
The 11x Dec-28 300/400 line is 11 long 300C against 1 short 400C, not eleven matched spreads.
The MU collar roll cannot be executed as the runbook writes it: Part 2 says 'BUY back 1400C' but there is no 1400C - the collar is 5x Dec-28 1740C short against 5x 900P long. First reported 08-18, and now 23% from firing.
The runbook's own 2026-09-04 note already says the MU roll-up as written is not premium-neutral: selling the 900P and buying the 1150P costs about $79k net on the 5-lot, while the roll down-and-out is a $41k credit. Re-price on the day.
The 8x GOOGL 300P long leg sits outside the $230-260 band and has no written repair order. Only the 5x 280P leg is reachable by the standing action item.
The weekly NLV comparison is today's frozen Friday close against 2026-08-31's live intraday read, which is not a like-for-like pair. At +2.50 percent against a -8 percent trigger it changes nothing, but on a week where it mattered it would.

Where the numbers came from

Market CLOSED - Labor Day. All 13 equity quotes came back top_status FROZEN, carrying Friday 2026-09-04's finished session for a sixth consecutive read.
Run label decided from the rails log rather than the clock: an entry for today already read markets-open, so this is the midday run. Section 4b (GPU pull) and section 5 (forecast pass) are open-run only and did not run.
get_account_positions, get_account_summary and 14 get_price_snapshot calls all succeeded on the first attempt. The run fired 11:39 CDT.
Positions BYTE-IDENTICAL to this morning's saved copy: the fresh pull was written to disk and diffed against it, zero content difference. 32 rows, the same 32 contract_ids, 0 of 32 marks changed.
Account summary byte-identical to this morning's, including excess liquidity - the $2.14 overnight margin drift the 06:35 run reported has not moved again.
prior_close SERVED on 13 of 13 for a sixth consecutive run. Last-trade stamps 1788565824-1788566398, that is 19:50:24 to 19:59:58 ET Friday, identical to all five prior runs. Every last and prior_close equal to the morning capture to the cent.
contract_description present on all 32 rows, so defined max loss prints the complete $175,059 and uncapped_shorts was measured: 10 GOOG short against 10 long, 30 GOOGL short against 38 long, zero uncapped.
IV: the GOOGL Dec-28 340P returned the shut-book option-midpoint-iv artifact again (dailyIv -1.0, annualIv -15.8745, isValid false); implied_vol's valid 0.3567735 was used. bid-ask 56.65/58.05.
Every Part 2, 3 and 4 runbook trigger row was walked by hand against this run's numbers - the three call-spread width percentages, the 340P extrinsic, the excess-liquidity ratio and the weekly NLV test. All clear.
The delta block is rails.py's output copied verbatim, as the prompt requires.
The GPU finding was computed directly from gpu-spot-daily.csv and gpu-spot-intraday.csv, not read off any script's printed summary. No new GPU pull ran - 4b is the open run's job and it ran there at 06:35.
The intraday capture is healthy: four captures today on the 3-hourly cadence, the last at 11:07.