2026-09-21 · midday

Gabriel traded. Fifty NVDA shares filled at $227.64 at 12:39:11 PDT, nine minutes before this run's book pull and the first fill in the account since 2026-09-16 - it is what moved cash from -$53,891.02 to -$65,274.02, reconciling to the cent against $11,382.00 of stock and $1.00 of commission. Away from the book, the forecasting pipeline's coverage gate refused today's commit over one INTC file, and on inspection the refusal is a front-expiry substitution rather than a loss. The rails themselves are where they were: the same two structural fires, and every price, IV, liquidity and runbook trigger clear.

What to do

13 long puts sit above the $260 long-leg ceiling: 280P x5 and 300P x8.
BUY TO CLOSE 5 GOOGL Dec15'28 360 PUT / SELL TO CLOSE 5 GOOGL Dec15'28 280 PUT
one combo x5, limit near mid, day · ~$17,026.07 debit ($34.0521 net per spread at this run's marks)
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, $161.32 cheaper than this morning's $17,187.39 and $724.38 cheaper than Friday's $17,750.45, because the 360P mark keeps falling faster than the 280P's as GOOGL rises. It addresses the 280P x5 leg only; the 8x Dec-28 300P has no written repair order anywhere, so this order leaves 8 of the 13 offending contracts in place.

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
gross_leverage	1.6900x	1.5x red (1.3x amber)	0.1900x over red
top2_name_delta_ratio	0.61x	amber 0.70x, red 0.85x	clear
total_delta_ratio	0.86x	amber 1.30x, red 1.50x	clear

Why each one reads the way it does

long_leg_band
The rail the action item addresses. No leg has been traded on it - today's fill was NVDA stock, not a campaign leg.
gross_leverage
Seventh consecutive rise: 1.6546, 1.6614, 1.6673, 1.6756, 1.6814, 1.6829, now 1.6900. This is the first rise in the sequence with a cause that is not purely marks: gross rose $36,700.48 since 06:43, of which $11,374.50 is the 50 NVDA shares bought on margin and the rest is the META and INTC revaluation. NLV rose $15,339.21 over the same window.

delta by name

Delta notional per name as of 2026-09-21: 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	$764,131	50.3%	1543 sh, 18 legs
MU	$157,500	10.4%	525 sh, 2 legs
META	$107,581	7.1%	144 sh, 0 legs
VOO	$87,769	5.8%	123 sh, 0 legs
RDDT	$81,854	5.4%	513 sh, 0 legs
NVDA	$63,004	4.1%	335 sh, 2 legs
AMZN	$25,872	1.7%	100 sh, 0 legs
EQT	$10,010	0.7%	200 sh, 0 legs
INTC	$8,656	0.6%	300 sh, 2 legs
TSLA	$374	0.0%	1 sh, 0 legs
DIS	$104	0.0%	1 sh, 0 legs

Everything else clear

Nearest trigger is GOOGL - tranche 3 confirmation (nearest written price level) at $360, 1% away.
No position moved more than 5% today. Biggest was -0.0430.
INTC +12.4%, now $122.04. No written rule for it.
META +12.3%, now $747.09. No written rule for it.
RDDT +5.8%, now $159.56. No written rule for it.

All 29 triggers swept, with today's number

price trigger	now	fires at	distance
GOOGL - tranche 3 confirmation (nearest written price level)	$356.18	≥ $360	1.1% up
GOOGL - tranche 2 weakness	$356.18	≤ $317	11.0% down
MU - collar roll level	$1,047.87	≥ $1,250	19.3% up
GOOGL - no chase, let the program shrink	$356.18	≥ $430	20.7% up
GOOGL - thesis break, close regardless of price	$356.18	≤ $280	21.4% down
GOOGL - Rule 6, close every short put	$356.18	≥ $434	21.9% up
GOOGL - washout, stop selling spreads and buy stock	$356.18	≤ $260	27.0% down
GOOGL - Rule 8, deploy the $100k crash reserve	$356.18	≤ $260	27.0% down
GOOGL - long spreads near max, apply the would-I-buy-it-fresh test	$356.18	≥ $500	40.4% up
GOOGL - shares capped by the 600Cs, roll the GOOG collar up	$356.18	≥ $600	68.5% up

level trigger	now	fires at	distance
Net cash / NLV is the closest ratio rail to its band, and today's purchase is what moved it	-0.0430	-0.05 amber	0.0070 of NLV, about $10,650 of further margin buying. It sat at -0.0358 at 06:43; the 50 NVDA shares moved it -0.0071 in one trade. It has already touched amber twice this month, on 09-14 and 09-15 midday.
Every long call spread against the runbook's 90%-of-width sell rule	45.3% is the highest	90% of width	Dec-28 300/400 45.3%, Dec-28 350/500 32.6%, Dec-28 300/600 30.4%, Sep-27 300/440 43.9%, Sep-27 300/445 43.2%. Nothing is half way there.
Dec-28 30-delta put IV, the alternate tranche-2 trigger	36.09%	43%	6.91 points. The designated option-midpoint-iv field returned isValid TRUE for a second consecutive run after four runs of negative volatility.
GOOGL Dec-28 340P extrinsic, the assignment watch	$51.13	$5	The put is $16.18 out of the money, so the whole $51.13 mark is extrinsic. The runbook notes this row only fires below roughly $250 in 2028.
Program cap against the campaign's defined-loss ceiling	$175,059	$200,000	$24,941 of room, 87.5% used. Unchanged in shape from this morning; no campaign leg traded today.
Excess liquidity against both the dollar floor and the ratio rails	$1,148,481, 75.6% of NLV	$650,000 floor, 40% amber, 25% red	It rose $11,799.79 across the session despite the purchase.
Top-2 name delta and total delta, both clear	0.607 and 0.860	0.70 / 1.30 amber	GOOG and MU carry $921,631 of the $1,306,857 total. Today's NVDA buy added about $11,375 of delta notional to a name that is sixth by size.
NLV against the Part 3 one-week freeze test	$1,519,529.99	a fall of 8% in one week	Struck against the 2026-09-14 open at $1,423,507.12 this is +6.75%, a rise. No verdict turns on the baseline because the test is one-directional.
rails.py evaluates every GOOGL price rail against GOOG's price, and this run did not silently correct it	script used $352.19 (GOOG)	$356.18 (GOOGL)	The $3.99 basis error is worth 1.1 points of distance on every GOOGL level. On the tranche-3 row it is the difference between 2.23% away and 1.07% away. Every rail's verdict is the same on both bases today. rails.json was not edited.

calendar	date	away
MU fiscal Q4 print, after the close - finviz carries it as Sep 30 AMC	2026-09-30	9d
Earliest clean sale date for the 6/05 GOOG loss lots if only bought calls count as replacement property	2026-10-01	10d
Earliest clean sale date if a WRITTEN call is also replacement property	2026-10-15	24d
GOOG print, first scheduled thesis adjudicator	2026-10-28	37d
Deploy the $100k reserve at market if no crash has come	2026-10-29	38d
GOOG capex guide, second scheduled adjudicator	2027-02-03	135d
GOOG lots vest long-term, add collar put leg 9 (+66 sh)	2027-03-13	173d
GOOG lots vest long-term, add collar put legs 10-11 (+200 sh)	2027-04-02	193d
GOOG lots vest long-term, legs 12-14 (+350 sh) - 193 of them clock-stopped by the 09-14 collar	2027-06-05	257d
GOOG lots vest long-term, leg 15 (+50 sh) - all 50 clock-stopped by the 09-14 collar	2027-06-25	277d

Spot prices

	last	day
GOOG	$352.19	+2.3%
GOOGL	$356.18	+1.9%
MU	$1,047.87	+3.2%
META	$747.09	+12.3%
INTC	$122.04	+12.4%
RDDT	$159.56	+5.8%
NVDA	$227.40	+2.3%
AMZN	$258.72	+2.0%
TSLA	$374.25	+2.7%
VOO	$713.57	+1.7%
DIS	$104.41	+1.7%
EQT	$50.05	+0.1%

The check itself

broken The forecasting pipeline's coverage gate refused today's commit.
ops/status.py --check exits 12: "latest daily record status is 'refused'". The record is today's, written at 12:49:58 PDT, so the job ran and the outputs are fresh - this is not a staleness failure. gate_examined 44 of 190 candidates, 146 skipped, gate_losses 1, committed False, commit None. The gate's own line: "check_no_data_loss: HEAD -> working tree: 44 file(s) checked, 146 skipped, 1 losing coverage". The gate examined a real, non-zero set of files, so its verdict means something.
diagnosed The single losing file is data/shared/derived/iv_INTC_book_regen.json, and the loss is a front-expiry substitution, not a deletion.
COMMIT_REFUSED-2026-09-21.md quotes: "LOSS data/shared/derived/iv_INTC_book_regen.json - core_vol_surface.by_expiry: keys removed: 2026-09-25, 2026-10-16", then the full collapse of both expiries - 2026-09-25 calls 71 -> 0 and puts 69 -> 0, 2026-10-16 calls 56 -> 0 and puts 49 -> 0 - repeated across core_vol_surface, surface, fitter.forward_strikes and diagnostics.repricing_error. Comparing HEAD to the working tree directly: HEAD carries nine expiries beginning 2026-09-25 and 2026-10-16; the regenerated file carries nine expiries beginning 2026-09-28 and 2026-10-23. The two front expiries were REPLACED, not dropped, and the expiry count is identical. The remaining diffs are ordinary chain churn: 2027-01-15 puts 55 -> 54, 2027-03-19 puts 56 -> 54, and diagnostics.dropped 37 -> 30, which is fewer rejections, not more.
broken Every option chain has been running on the Tradier fallback since 2026-09-17, and that is the likely cause of the substitution above.
The log's FALLBACK DATA line names all thirteen tickers on Tradier for both chain and iv30. Alpha Vantage's HISTORICAL_OPTIONS endpoint now answers "This is a premium endpoint. You may subscribe to any of the premium plans" - twelve such refusals in today's log. It appears in the 2026-09-17 and 2026-09-18 logs with the same count of twelve and in neither the 09-15 nor the 09-16 log, so the gating started on 09-17 and the primary chain source has been dead for four sessions. The committed INTC surface predates it; the regenerated one is Tradier's expiry ladder.
broken The deploy step failed, so stocks.dottl.xyz did not receive today's build.
"=== 12:49:56 FAIL deploy (exit 1) ===" on stocks_dashboard/deploy/deploy.sh. Wrangler 4.136.1 reached "Uploading... (205/968)" against the existing stocks-dottl-xyz Pages project and returned "Failed to upload files. Please try again. Error: {})". Its own log is at ~/.wrangler/logs/wrangler-2026-09-21_19-49-41_756.log. The wording is a retryable upload error, not a rejection. Note that the commit refusal is independent of this: the gate refused, so there was nothing new to publish either way.
noted Thirty-one of the thirty-three failed steps are the chronic set, unchanged from 09-18.
steps_ran 124, steps_failed 33. The 09-18 record carried the same failure list with steps_failed 27 and still committed, so those are long-standing: house_forecast for AMZN, GOOGL, META, MU and NVDA are all blocked, and street / scorecard / book_ev fail across BRK-B, DIS, EQT, RDDT, TSLA and VOO. Genuinely new today are the four HD steps - street, scorecard, sync_positions and book_ev - which suggests HD is being onboarded, plus deploy and check_no_data_loss.
defect The refusal report hands the reader a recheck command pointing at the wrong repository.
COMMIT_REFUSED-2026-09-21.md says to recheck with /Users/gabedottl/code/options_trading/backend/.venv/bin/python -m ops.check_no_data_loss -v. That interpreter is in code/options_trading; the job, the gate and every path the report itself quotes live in code/financials. The interpreter exists, so the command fails on the module rather than on the binary, which makes it look like the check is broken rather than the path. Small, but it is the one command the report tells a cold reader to run.

Rails that could not be evaluated

THE OPEN RUN'S 'last fill still 2026-09-16' IS NOW STALE and would have been the wrong thing to carry forward. NVDA filled at 12:39:11 PDT. This is the case the twice-daily cadence exists for, and it is worth noting that nothing in the check would have caught the trade if the cash line had not moved - the rails all pass, the position count is unchanged at 35, and get_account_orders is empty because the order completed. The quantity diff and the trades pull are what found it.
MU still has NO written price level anywhere below spot, and the fiscal Q4 print is now 9 days out. It is the second-largest delta in the book at $157,500 and the runbook's only MU price row is the $1,250 roll trigger, 19.29% above. A downside print has no written response.
NVDA has no written rule either, and it is now a position Gabriel is adding to. The check reported the fill and the delta and proposed nothing, which is correct under the written rules, but the gap is real: a name being actively traded with no rail is the one place this check is blind by construction.
META's prior close is still unreconciled between two IBKR surfaces. The connector's change basis implies Friday at 665.22; the 09-18 and weekend runs recorded 665.75 from a get_price_history RTH bar. The same 0.53 gap the open run logged, unchanged. No verdict turns on it - META has no written rule and both bases clear the 5% move threshold.
get_price_snapshot's prior-close field returned an EMPTY OBJECT on all thirteen requests again, both runs today. Every prior close is derived as last minus change from the same response. It has now failed this way on consecutive runs and should be treated as the connector's normal behaviour rather than a one-off.
rails.py prices every GOOGL rail off GOOG. Reported by the open run, still unfixed, and rails.json was not edited by either run. It is a $3.99 basis error today, worth 1.1 points of distance on every GOOGL level.
stressed_excess_liquidity has still never been measured. It needs Risk Navigator at -30% price and +50% IV with the TWS gateway up, and it is one of two rails in the file that no run has ever evaluated.
short_dec28_340p_extrinsic is marked unevaluated by the script because it wants a live Dec-2028 chain pull. This run computed it from the position mark instead - $51.13 against a $5 floor - and reports it as clear, but by a different method than the rail specifies.
The financials pipeline's primary options source has been dead for four sessions without anything raising it until the gate happened to refuse. The FALLBACK DATA line was in the 09-17 and 09-18 logs and no run reported it, because a fallback that silently succeeds looks like a green run.

Where the numbers came from

Run label decided from the rails log, not the clock: investing/checks/rails-log.jsonl already carried a 2026-09-21T06:43:50-07:00 markets-open entry, so this is the MIDDAY run. Section 4b (GPU) and section 5 (forecast pass) were both SKIPPED by that rule, and gpu and forecasts are empty objects in this record.
THE MARKET IS OPEN. Book pulled 19:48Z / 12:48 PDT, 42 minutes before the 13:00 PDT close. All twelve equity snapshots returned top_status REALTIME with is_close FALSE - an improvement on the FROZEN payloads the weekend runs had to work around.
PRIOR CLOSES WERE DERIVED, NOT ESTIMATED. get_price_snapshot's prior-close field returned an EMPTY OBJECT on every request, so each prior close is last MINUS change from the SAME snapshot response - two fields of one payload, never two reads stitched together. All twelve reproduce this morning's open-run prior closes to the cent, and eleven of twelve reproduce Friday 2026-09-18's recorded RTH close to the cent. META is the exception and is written up in gaps.
THE FILL WAS VERIFIED THREE WAYS, not inferred. get_account_trades DAYS_7 returns three prints under order 1653556762 at 2026-09-21T19:39:11Z - 39, 1 and 10 shares of NVDA at $227.64 on BATS. The position diff against this morning's file shows NVDA 285 -> 335 and no other quantity changed across the 35 rows. And the cash line reconciles exactly: $11,382.00 of stock plus $1.00 of commission against a cash move of -$11,383.00, residual $0.00. get_account_orders is empty because the order completed, not because nothing was sent.
IV came from option-midpoint-iv annualIv 0.3608647459008823, isValid TRUE - the designated field, valid for a second consecutive run. implied_vol annual_iv 0.36021790375047236 agrees to three decimals. top_status on the option was DELAYED and last 50.81 had is_close FALSE; bid-ask 50.30 / 52.10.
RUNBOOK PARTS 2 THROUGH 4 WERE SWEPT ROW BY ROW from this run's own marks, not carried from the morning. The five long call spread percentages, the nine price-distance rows, the two IV rows, the 340P extrinsic row, both excess-liquidity rows and the NLV week row were each recomputed against GOOGL $356.18 and MU $1,047.87. Every row is clear. The NLV week comparison is struck against the 2026-09-14 open at $1,423,507.12, giving +6.75%; Part 3 tests for a FALL of 8%, so no verdict turns on the baseline.
THE SCRIPT'S SPOT BASIS IS STILL WRONG and this run did not silently correct it. rails.py reported every GOOGL price rail against GOOG's $352.19. The watch_price block is stated on the correct GOOGL basis of $356.18 and the discrepancy is written up in watch_level and gaps. rails.json was not edited.
THE PIPELINE DIAGNOSIS WENT BEYOND THE STATUS LINE. ops/status.py --check exits 12; its record, the COMMIT_REFUSED-2026-09-21.md report and ops/logs/2026-09-21.log were all read, and HEAD was compared to the working tree directly with git show to establish that the INTC expiry keys were substituted rather than dropped. The substitution check was also run across every iv_*_book_regen.json in the tree; INTC is the only such file that exists, so this is not a systemic pattern hiding behind one example.
READ-ONLY throughout on ~/code/financials: nothing was edited, committed, reverted, rerun or cleaned, the working tree was left exactly as the job left it, and the LaunchAgent was not touched. rails.json, build_markets_report.py and googl-mu-runbook.md were likewise not edited. The only writes this run made are positions.json, summary.json, quotes.json, this report, the rails-log line and the check-in block.