2026-09-10 · midday

The one order on this page is $548.10 cheaper than it was this morning, and the reason is the reason this morning's page said had stopped working. The long-leg repair - buy back the 5 short Dec-2028 360 puts, sell the 5 long 280 puts - costs $19,448.05 now against $19,996.15 at the open run, the cheapest reading since Tuesday. Alphabet rose $3.87 across the same hours. That is $141.60 of debit per dollar of GOOGL, within two percent of the $139 figure yesterday's page published and this morning's page threw out. This morning's rejection was fitted on a 52-cent move in the stock, a sample small enough for a fraction of a vol point of skew to swamp it; today's move is seven times larger and the rule survives it. Both effects are real - the skew between $280 and $360 narrowed from 0.85 vol points to 0.55 this run and helped - but spot is the first-order term and should be treated as such. Reprice before sending either way. Nothing else broke. The same three structural rails are over. Every runbook trigger in Parts 2 through 4 is clear, and the nearest written level, tranche-2 weakness at $317, has backed off to 4.0 percent below GOOG from 3.0 percent this morning. Nothing was traded: 32 rows, the same contracts, zero quantity changes, cash unchanged to the cent for an eighth session. NLV is $1,409,049, up $8,331 since the open and down 1.0 percent on the week. Micron kept falling and is now 4.5 percent down on the day at $981.28, 27 percent below its collar roll-up level. Reddit is the day's largest gain at plus 6.85 percent and Intel the largest loss at minus 5.00 percent, which lands exactly on the 5 percent move-alert line without tripping it. Neither has a written rule, so neither gets a proposal. Two things about the check itself. The broker's real-time quote endpoint, which failed all nine attempts this morning, worked on all fifteen calls this run and stamped every equity REALTIME - so today's prices are read directly rather than reconstructed, and the campaign implied vol is the broker's own 35.74 percent rather than this morning's solved 36.76 percent. And yesterday's page published an explanation for the patchy prior-close field - that the service rate tracks the time of day - which today falsifies: the same midday slot served 1 of 13 today against 5 of 13 yesterday. That claim should not be carried forward.

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 · ~$19,448.05 debit ($38.8961 net per spread on this run's live marks)
Your playbook: “long leg never above $260 (must sit at or above the $230-250 floor zone, never below $230)”
Eleventh read of the same standing order, and the cheapest since Tuesday. The series: $18,769.60 on six frozen-market runs, $18,793.92 on Tuesday's live chain, $19,885.61 Wednesday morning, $19,570.59 Wednesday midday, $19,996.15 this morning, $19,448.05 now. This morning's page declared the $139-per-dollar-of-GOOGL sensitivity broken, on the grounds that GOOGL was 52 cents higher than the day before while the debit was $110.54 higher. Today's larger move argues the other way: GOOGL is $3.87 higher than at the open run and the debit is $548.10 lower, which is $141.60 per dollar, within two percent of the rejected figure. A 52-cent sample cannot test a rule of this size - a quarter of a vol point of skew moves the debit more than 52 cents of spot does. Skew did also help here: solving each leg off its own mark puts the 360 put at 36.96% and the 280 put at 36.41%, a 0.55-point gap against 0.85 this morning, so the strikes converged. The honest statement is that spot is the first-order term with skew as a real second-order one, and roughly $140 per dollar is a usable approximation that should still be repriced on the day. Not an argument for urgency in either direction - the rail has been open for weeks and Gabriel sets the timing. Only the 5x 280P leg is reachable by this order; the 8x 300P leg has no written repair. 15 minutes to the close as this is written.

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	-0.066	amber -0.05, red -0.10	amber
gross_leverage	1.741x	amber 1.30x, red 1.50x	red
top2_name_delta_ratio	0.66x	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 has been traded.
net_cash_ratio
Cash -$93,230.12 against NLV $1,409,049.45. The cash figure has not moved by a cent since 2026-09-04, an eighth consecutive session, so this ratio tracks NLV alone. NLV rose $8,331 since the open run and the ratio improved from -0.0666 to -0.0662 with nothing bought or sold.
gross_leverage
Gross $2,453,432.02 over NLV $1,409,049.45. Down from 1.755x this morning; gross fell $4,222 while NLV rose $8,331, so both terms moved the right way and neither was a trade. Carried caveat, unchanged: this divides GROSS position value by NLV against thresholds the IPS wrote for delta notional, and the delta rows read 0.66x and 0.94x, so the red reading overstates true leverage. The book and the written rule disagree; rails.json was not edited to reconcile them.

delta by name

Delta notional per name as of 2026-09-10: 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	$790,660	56.1%	1543 sh, 16 legs
MU	$141,988	10.1%	525 sh, 2 legs
META	$92,660	6.6%	144 sh, 0 legs
VOO	$85,677	6.1%	123 sh, 0 legs
RDDT	$80,269	5.7%	513 sh, 0 legs
NVDA	$49,363	3.5%	285 sh, 2 legs
INTC	$30,279	2.1%	300 sh, 0 legs
AMZN	$25,165	1.8%	100 sh, 0 legs
BRK B	$20,254	1.4%	40 sh, 0 legs
EQT	$11,010	0.8%	200 sh, 0 legs
TSLA	$364	0.0%	1 sh, 0 legs
DIS	$105	0.0%	1 sh, 0 legs

Everything else clear

Nearest trigger is tranche 2 weakness (nearest written level, and what rails.py reads) at $317, 4% away.
No position moved more than 5% today. Biggest was RDDT +6.85%.
RDDT +6.8%, now $156.47. No written rule for it.
Named, not acted on. RDDT is on the no-written-rule list, so this check proposes nothing for it. It closed at $146.44 yesterday and has extended from +2.09% at the open run to +6.85% now, the book's largest single-name gain today on a $80,279 line.
INTC -5.0%, now $100.93. No written rule for it.
Not fired by the script and reported anyway because of where it landed. The broker's own change field prints -5.00% against a 5.00% alert threshold; the arithmetic on last minus prior close is -4.998%, which is why rails.py left it off the move list. Named here so the near-miss is on the record rather than invisible. INTC has no written rule either way, so nothing is proposed.

All 27 triggers swept, with today's number

price trigger	now	fires at	distance
GOOG · tranche 2 weakness (nearest written level, and what rails.py reads)	$330.37	≤ $317	4.0% down
GOOGL · tranche 2 weakness	$332.76	≤ $317	4.7% down
GOOGL · tranche 3 confirmation	$332.76	≥ $360	8.2% up
GOOG · tranche 3 confirmation	$330.37	≥ $360	9.0% up
GOOG · thesis break, close regardless of price	$330.37	≤ $280	15.2% down
GOOG · washout, stop selling spreads and buy stock	$330.37	≤ $260	21.3% down
GOOGL · crash powder, deploy the $100k reserve	$332.76	≤ $260	21.9% down
MU · collar roll-up	$981.28	≥ $1,250	27.4% up
GOOGL · close every short put	$332.76	≥ $434	30.4% up
GOOGL · long spreads near max, apply the buy-it-fresh test	$332.76	≥ $500	50.3% up
GOOGL · shares capped by the 600Cs, roll the collar up	$332.76	≥ $600	80.3% up

level trigger	now	fires at	distance
Dec-28 30-delta put IV	35.74%	> 43%	+7.3 pt
GOOGL IV, crash trigger	35.74%	> 50%	+14.3 pt
11x Dec-28 300/400 call spread, mark vs width	40.44%	>= 90%	+49.6 pt
8x Sep-27 300/440 call spread, mark vs width	36.45%	>= 90%	+53.5 pt
10x Dec-28 350/500 call spread, mark vs width	28.10%	>= 90%	+61.9 pt
GOOGL Dec-28 340P extrinsic	$52.92	< $5	$47.92
Excess liquidity / NLV	74.81%	< 40%	+34.8 pt
NLV change on the week	-1.04%	-8% in one week	+7.0 pt
Defined max loss vs program cap	$175,059	$200,000	$24,941 headroom
Biggest single-name move today	RDDT +6.85%	>= 5%	fired, no written rule

calendar	date	away
Wash-sale window opens for the 6/05 GOOG loss lots	2026-09-17	7d
Micron fiscal Q4 print (not a trigger row; the Seeking Alpha pieces filed this morning all bear on it)	2026-09-30	20d
Alphabet Q3 print, scheduled thesis adjudicator - re-read the house view within 72h	2026-10-28	48d
No crash by this date: deploy the $100k reserve at market anyway	2026-10-29	49d
Alphabet capex guide, second scheduled adjudicator	2027-02-03	146d
GOOG lots vest long-term (+66 sh): add collar put leg 9, log in the straddle register same day	2027-03-13	184d

Spot prices

	last	day
GOOG	$330.37	+0.6%
GOOGL	$332.76	+0.6%
MU	$981.28	-4.5%
AMZN	$251.65	-0.3%
BRK B	$506.34	-0.1%
DIS	$105.08	+0.9%
EQT	$55.05	+0.8%
INTC	$100.93	-5.0%
META	$643.47	-1.6%
NVDA	$218.27	-2.4%
RDDT	$156.47	+6.8%
TSLA	$363.79	-1.1%
VOO	$696.56	-0.6%

The check itself

resolved the real-time quote endpoint is back.
get_price_snapshot failed all nine attempts across roughly fifteen minutes of this morning's run and was reported as unavailable for the whole of it. It answered all fifteen calls this run, first try, with no retries needed: thirteen equities, the GOOGL Dec-28 340 put, and one repeat. Every equity came back top_status REALTIME with an unhalted last-trade stamp inside a 68-second window. The practical consequence is that this page's prices are read rather than reconstructed from five-minute bars and the positions feed, and the campaign implied vol is the broker's own field rather than a solved estimate. The outage lasted one run and left no residue.
wrong yesterday's explanation of the patchy prior_close field does not survive today.
Wednesday midday's page concluded that prior_close service rate 'is a function of the time of day, not of the feed's health', on the evidence that the morning served 1 of 13 and the midday run seven hours later served 5 of 13. Today's midday run - the same slot, the same thirteen contracts, the same broker - served 1 of 13, EQT alone. Two observations at the same time of day differing by a factor of five is not a clock effect. What does hold, and is the part worth keeping, is the underlying finding: a low service rate is not evidence of a degraded feed. The twelve unserved names were derived as last minus the broker's own change field and ten of the twelve reproduce this morning's independently sourced daily-bar closes to the cent. The time-of-day story should not be carried forward.
wrong this morning's trailing-slash diagnosis of the page URL does not reproduce.
This morning's page corrected Wednesday's Cloudflare-edge-cache explanation and replaced it with a trailing-slash one: that markets.dottl.xyz/2026-09-10 serves the right page at 40,573 bytes while markets.dottl.xyz/2026-09-10/ serves the 6,675-byte site index. Tested against this run's deploy, the behaviour is the reverse and simpler than either story. The no-slash form returns an HTTP 308 with location /2026-09-10/ and a zero-byte body; the trailing-slash form returns 200 and the 28,366-byte page. That is Cloudflare Pages' ordinary directory redirect, which every browser follows, so both forms land a reader on the right page and neither is broken. What both this morning's note and Wednesday's were actually looking at is curl without -L reporting the redirect as an empty response. Neither an edge cache nor a slash convention is involved. The link below is the no-slash form the prompt specifies, and it resolves.
found this run the solved-IV method reads about a vol point high, and the bias is not constant.
This morning's run could not read implied vol from the broker and solved it instead off the GOOGL Dec-28 340 put's own mark, disclosing a measured bias of +0.68 vol points from a single back-test on the prior day. Today both numbers exist on the same leg at the same moment: the broker's option-midpoint-iv reads 35.74% and the same solve run on the same mark returns 36.68%, a bias of +0.94 points. Larger than the disclosed estimate and in the same direction. Nothing here changes a verdict - the nearest IV trigger is 7.3 points away, roughly eight times the error - but the per-leg vols quoted in this run's action item are solved values and are comparable to each other, not to the broker's 35.74%.
carried the midday slot still fires 25 minutes before the close, not the hour it is written to fire.
Reported as new on Wednesday and unchanged since: the written slot is '12:00 PM Pacific (one hour before the close)', the cron entry is '35 6,12 * * *', and the book here was pulled at 12:35 Pacific with the bell at 13:00. It cost nothing again today - every quote is live and every number is a live number - but the runway before the close is 25 minutes rather than the 60 the prompt describes, which matters for anything Gabriel would want to send before the bell. Not changed here: the cron entry is written configuration.
carried two account rails have still never been measured.
rails.py reports stressed_excess_liquidity and short_dec28_340p_extrinsic as unevaluated and this run treats neither as passing. The first needs IBKR Risk Navigator with the TWS gateway up and has never been measured. The second is different today: the script cannot evaluate it without a full Dec-2028 chain pull, but the specific leg it names was quoted live this run, and the runbook Part 4 sweep - which is a hand check, not the script - reads its extrinsic at $52.92 against a $5 trigger. That is reported in watch_level as a swept row, not as a rail the script cleared.

Rails that could not be evaluated

The GOOGL Dec-28 340 put returned top_status DELAYED while all thirteen equities returned REALTIME. Its mark is used for the campaign IV, the extrinsic row and the action item's per-leg vols. Every option mark in the book moved off the open run's, so nothing here is stale in the sense of unchanged, but the option chain and the equity quotes are not stamped as the same freshness.
prior_close was served by the broker for EQT alone. The other twelve are last minus the endpoint's own change field. Ten of the twelve reproduce this morning's independently sourced daily bars to the cent.
RDDT's prior close is corrected to $146.44 here. This morning's page carried $146.47, which came from an hourly-bar fallback after four failed daily calls; the daily bar was pulled successfully this run and reads $146.44. The correction moves RDDT's day change from +6.83% to +6.85% and changes nothing else.
NVDA's prior close still disagrees between the two feeds by 25 cents - $223.67 on the daily bar against $223.42 implied by the snapshot's change field. The daily bar is used, as this morning. Cause unproven; most likely a closing mark against a closing trade. NVDA is 4.4% of NLV, has no written rule, and reads -2.41% or -2.31% depending on the choice.
The per-leg implied vols in the action item are solved off each leg's own mark, not read. On the one leg where the broker's own IV is also available today, that method reads 0.94 vol points high. The vols are internally comparable and are not comparable to the 35.74% quoted elsewhere on this page.
The gross-leverage rail divides GROSS position value by NLV against thresholds the IPS wrote for delta notional. It reads 1.741x red while the delta rows read 0.66x and 0.94x, both clear. The book and the written rule disagree; the rail is reported as it fires and the file was not edited to match.
This morning's published trailing-slash explanation for the page URL is falsified. The no-slash form 308s to the trailing-slash form, which serves the page; curl without -L reporting an empty body was the whole of the symptom in both this morning's note and Wednesday's. See plumbing.
No GPU pull this run. Section 4b is the open run's job and ran at 07:16; the levels on this page's sister report are from that pull, not this one.
No forecast pass this run. Section 5 is the open run's job. Nothing on this page is a new sell-side or retail call, and the absence of one is not evidence that none was published this afternoon.
The SemiAnalysis composite anchor remains the 2026-04 print of $2.82, five months stale, and later months are paywalled. Carried from the open run; no monthly append was due.
The GPU tripwire still cannot fire before roughly 2026-11-16 on either series, because the 90-day lookback the written trigger needs does not exist yet. Carried from the open run.
Wednesday midday's published claim that prior_close service rate tracks the time of day is falsified by this run and should not be carried forward. See plumbing.
This morning's published rejection of the $139-per-dollar repair sensitivity was fitted on a 52-cent move in the stock. Today's $3.87 move gives $141.60 per dollar. The rejection does not hold at this sample size and the approximation is restored, with the caveat that skew is a real second-order term.
Nothing in rails.json was edited. Where the book and the written rules disagree, it is reported rather than reconciled.
markets.dottl.xyz still has no Cloudflare Access gate - the deploy reports 'Access: off (public)', so anyone with the URL sees the full book, the position sizes and the NLV. Carried, unchanged, and not fixed here because it is a configuration decision.
deploy.sh needed no PATH prefix this run, after seven consecutive runs that reported needing one. One clean invocation is not proof the underlying problem is gone.

Where the numbers came from

Run label decided from the rails log, not the clock: 2026-09-10 already carried a markets-open entry, so this is the midday run and sections 4b (GPU) and 5 (forecasts) were both skipped by rule.
Book pulled 19:35Z (12:35 Pacific), roughly 25 minutes before the close. 32 rows, the same 32 contract_ids as the open run, zero quantity diffs, all 32 marks moved including all 20 option rows. Nothing was traded intraday.
get_account_positions and get_account_summary each succeeded first try. get_price_snapshot succeeded on all 15 calls after failing all 9 this morning. get_price_history was used twice, for the RDDT and NVDA daily bars.
All 13 equity quotes are single snapshot calls stamped 19:37:12 to 19:38:20Z, inside a 68-second window, every one top_status REALTIME. The GOOGL Dec-28 340 put returned DELAYED.
The campaign implied vol is the broker's own option-midpoint-iv on the GOOGL Dec-28 340 put, 35.74%, read rather than solved for the first time in two runs.
rails.py fired 3, cleared 18 and left 2 unevaluated. The delta block is the script's own output, carried verbatim and not recomputed.
Runbook Parts 2, 3 and 4 were swept by hand against this run's numbers, including the three call-spread width percentages, the MU roll level, the IV crash trigger, the 340P extrinsic and the one-week NLV test. Every row clear.
The one-week NLV test uses the 2026-09-03 midday reading of $1,423,900.37 from the rails log as its anchor.
Two claims published earlier this week are corrected here: Wednesday's prior_close time-of-day explanation and this morning's rejection of the repair sensitivity.
Nothing in rails.json was edited and no forecast record was added, resolved or removed.