The same four rails as every run since 08-20, unchanged and structural, and every price, level and calendar trigger swept clear. Nothing new came out of the sell side or Seeking Alpha on either name. The finding this run is not in the market: the runbook's 11x Dec-28 300/400 call tranche exists in the book as eleven long 300 calls against a single short 400 call, so the 90%-of-width trigger that this check reports every day has been measuring a spread that mostly is not there.

What to do

• 280P x5 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 · ~$18,510 debit ($37.02 net per spread at the open)
  Your playbook: “long leg never above $260 (must sit at/above the $230-250 floor zone, never below $230 where protection stops paying before history bottoms)”
  The same standing order as prior runs, repriced: $37.02 net now against $36.40 at yesterday's midday, $35.88 at yesterday's open, $36.84 on 08-24 and $36.76 on 08-21. That is the most expensive this fix has printed in the series, and the drift is one-way - the short 360P has richened faster than the long 280P on every run this week. The 8x Dec-28 300P also trip this rail and remain the separate matter. No order is proposed against them.

Still over, unchanged

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

  rail	now	limit
  program_cap	$175,059	$100,000 cap	1.8x, $75,059 over
  gross_leverage	1.720x	1.50 red / 1.30 amber	red
  net_cash / NLV	-0.069	-0.05 amber / -0.10 red	amber

Everything else clear

• Nearest trigger is tranche 3 confirm at $360, 4% away.

Cost of compute

• H100 spot $2.67 against the $2.82 monthly composite. No tripwire.
• B200 is the move: the Vast median fell to $5.85 from $7.25 yesterday and $8.38 on 08-20, -19.3% in a day on a 40-listing book, the lowest print since the series began. The contract side did not follow - getdeploying still quotes B200 cloud at $7.88 and hyperscaler at $15.18, both unchanged. So the marketplace floor for the newest part dropped and nothing else did.
• The H100 tripwire cannot fire yet. It tests spot against 90 days back and the daily series is nine days old, so the 30d and 90d columns print n/a and the avg30/avg90 figures are just the whole nine-day series restated. The check is not armed until roughly late November. Meanwhile the H100 Vast median swung between $1.93 and $4.01 inside those nine days, so today's +14.6% day move is noise, not information. The one H100 number actually trending is the SFCompute clearing price: $1.88 today against $2.27 on 08-18, -17% over the series and its lowest print yet.

The check itself

• never worked the runbook's 11x Dec-28 300/400 call tranche is not eleven spreads in the book, and the 90%-of-width trigger has never been measuring what it names.
  Runbook Part 1 says 'BUY tranche 1 (~$100k): 11x Dec-28 300/400 combo and 10x Dec-28 350/500 combo'. The book holds 11x long GOOGL Dec-28 300C against exactly 1x short Dec-28 400C. The 350/500 tranche is intact, 10 long against 10 short. So ten of the eleven 400C short legs are not there. Part 2's row 'Any long call spread marks >= 90% of its width -> SELL that spread' has been reported every run as a single 300/400 percentage (42.3% today, 43.2% yesterday) computed off one short leg and applied to all eleven contracts. Those ten unpaired 300Cs are not naked - the 10x short 500C and 18x short 600C plus 1,543 GOOG shares more than cover them - but they are not the defined-width structure the runbook priced, and there is no width to be 90% of. rails.json does not cover the call side at all (its program is right:'P'), so nothing else catches this. This run does not edit rails.json or the runbook. Gabriel decides which one is wrong: the book, or the doc.
• never worked the SemiAnalysis composite has not been published since April, so the glut anchor is four months old.
  Refetched gpu-index.semianalysis.com this run: its latest H100 composite is still Apr 2026 at $2.82, the same print the CSV captured on 08-18, with Mar $2.80, Feb $2.83, Jan $2.82 behind it. The monthly append is not being missed - the publisher has not printed May, June, July or August. That matters because runbook Part 5 arms the glut watch off 'SemiAnalysis composite; -20%+ in a quarter = tripwire, flat is pre-signal', and the CSV's own note on the Apr row reads 'PRE-SIGNAL tightening'. A composite that is flat because nobody published it is not the same fact as a composite that is flat because compute prices are flat, and the second reading is the one the doc invites. Every daily marketplace number this check prints is currently being anchored to a four-month-old benchmark.
• never worked rails.py tests every GOOGL price level against the GOOG quote.
  Unchanged and still open, reported since 08-24. rails.json lists the campaign underlyings as ["GOOG", "GOOGL"] and rails.py loops them in order, taking the first with a quote and breaking. GOOG is first, so thesis_break, t2_weakness, t3_confirmation, washout_buy, no_chase, rule6_close_shorts and rule8_crash_powder are all measured against GOOG. Every one of those levels is written as a GOOGL level. Today GOOG is $341.05 and GOOGL $344.75, a 1.08% gap, so the script reads tranche 3 as 5.6% away while against GOOGL it is 4.4% away. No rail has flipped on this yet. This page shows the GOOGL number, matching the docs; the script's own output uses GOOG. This run does not edit rails.json.
• will happen again get_price_snapshot returned an empty prior_close on all thirteen tickers this run.
  Worse than the recent pattern: ten of thirteen were empty yesterday midday, eleven of thirteen at yesterday's open, thirteen of thirteen today, so there was no populated field left to cross-check the workaround against. Prior close is derived as last minus the broker's own 'change' field on the same snapshot. Every derived value reconciles with the change_pct the broker returned beside it, so the day moves on this page are internally consistent, but today they are unverified against a second source.
• source broke Akash has printed identical rates for nine straight days.
  The Akash weighted averages have been exactly $2.71 for H100, $4.45 for H200 and $1.71 for A100 on every day of the series since 08-18, to the cent, with a 0.0% day move again today. getdeploying moves a few cents day to day and SFCompute moves daily, so a genuinely pinned market across three accelerators for nine sessions is unlikely. Treat the Akash column as possibly cached rather than as nine flat observations. Not yet confirmed either way.