The track record.

Households funded the spending increase from savings, not credit: of $1,185bn of consumption growth in the year to April, $591bn (~50%) came from the saving rate falling 5.5% to 3.0%, against $106bn (9%) from total consumer credit — roughly five to one

CONFIRMED — the pre-registered falsification test (household-hhdc-q2-preregistration) resolved 11 Aug 2026: the Q2 HHDC printed cohort delinquency transitions roughly stable vs Q1 — card 90+ by age 18-29 +0.41pp offset by 30-39 -0.42pp, auto flat, aggregate 30+ flows inside their two-year plateau — so households had not pivoted from savings to credit. The sequencing claim survives its first dated checkpoint; the ongoing watch passes to the revolving-credit and saving-rate-floor triggers.

Q2 print mapping, stated pre-print: >=$1.86B revenue (+85%) is slow-fade evidence; consensus $1.81B (+80%) is one tick down the curve; <$1.78B is the first faster-than-breakeven datum; an FY26 guide above ~$7.9B is the strongest slow-fade signal available tonight

SCORED 2026-08-04 in pltr-fade-measured — revenue $1.935B (+93%, accelerating) cleared the $1.86B slow-fade bar and the $1.81B consensus by a wide margin; FY26 guide $8.15B cleared the $7.9B bar. The strongest possible slow-fade reading: the pre-registered re-weight fired (bull 25%->35%, bear 30%->20%), fair value restated $112 -> $149.

Stage one of the ladder: ~2% signed by the recess, against a market at ~3c — with 'a filed cloture motion or a locked September date' named in advance as the pre-recess bull tells

CORRECT, in near-agreement with the market. H.R. 3633 was not law by the recess — resolved No 10 Aug 2026 (Kalshi Sep-1 leg at 1c). Both named bull tells fired in the final pre-recess hours: Thune moved to proceed and filed cloture on 8 Aug 2026 (CR S4557), and the Senate locked the cloture vote for 2:15 p.m. ET, 15 Sep by unanimous consent. The disagreement now sits in the open stages, inverted: Kalshi Oct-1 leg 11c vs house ~18%, 2026 leg 23c vs house ~35% — the house is above the market for the first time in the franchise.

PCE three-month annualized momentum has risen every month from 2.7% (Nov 2025) to 6.3% (May 2026) — six readings, six increases, no interruptions — while the market prices only ~28% odds of a July 2026 hike and ~71% hold

SPLIT — the market half scored clean (the 71% hold came in). The momentum half broke the next morning: June PCE, released 30 Jul, printed 3.07% three-month annualized against 6.31% in May, ending the six-month climb on the seventh reading. Accurate as description at publication; inverted as a forward signal inside 24 hours.

None of the Fed's three dials argues for a cut: headline PCE +4.1% YoY, unemployment 4.2% with initial claims 187k (a 15-year low), HY OAS 2.79pp near three-year tights, Chicago Fed NFCI -0.55

CONFIRMED on the conclusion, SOFT on one input — the committee held and not one member voted to ease, so no dial argued for a cut. But the note's labor dial leaned on the 187k print as a 15-year low; it was genuinely the lowest since 2011 and yet a single-week trough after five weeks of 209k-230k. The 4-week average through 25 Jul is 202,750, above the note's own 200k threshold. Correct call, cherry-picked input — logged so the next note quotes the 4-week average.

Policy is negative in real terms — effective fed funds 3.63% against 4.1% headline inflation — while the 10-year at 4.69% is doing the tightening on its own

NARROWED — the June vintage cut headline YoY to 3.67%, so the real policy rate went from roughly -0.5pp to roughly flat (3.63% vs 3.67%). The negative-real-rate framing was true on May data and is now marginal. The 10-year leg held: 4.67% on the decision, 4.68% the day after, against 4.69% at publication.

The cut branch is dead on this data: prediction markets price cuts at a fifth of a penny, and there is no version of the July 2026 meeting where the Fed eases

CONFIRMED, emphatically — the 29 Jul FOMC held at 3.50-3.75% with three dissents (Hammack, Kashkari, Logan) all preferring a 25bp HIKE. Zero votes to ease. The 'coiled spring' read was right, but it surfaced in the vote tally, not in the statement language the note told readers to watch: inflation stayed 'remains elevated,' never shifting to 'has risen.' EXTENDED 2026-09-16: the September SEP goes past the note's own window — the 2027 median fed funds dot is 4.125%, identical to the 2026 median, i.e. no cuts anywhere in 2027, with 8 of 18 participants ABOVE the median at 4.375%. The cut branch is not merely dead for July; it is absent from the committee's own three-year path. The statement language the note told readers to watch DID finally move, one meeting late: the supply-shock/energy attribution was deleted outright and replaced with 'The Committee will deliver price stability.'

Strongest counter-case (carried honestly): core three-month momentum has decelerated three straight months from 4.8% to 3.5%, and the 5-year breakeven at 2.18% has been falling since spring — the expectations anchor is the single best argument this is not the 1970s

VINDICATED — the note's self-argued counter-case is the reading the data endorsed. Core three-month annualized fell to 2.89% in June, a fourth straight deceleration, and the 5-year breakeven sat at 2.24% on decision day versus 2.18% at publication: no anchor repricing in either direction. The hedge outperformed the thesis. PARTIALLY UNWOUND 2026-09-16 — the anchor leg has slipped: the 5-year breakeven printed 2.41% on 15 Sep against 2.18% at publication and 2.24% on the July decision day, ~20bp of drift the wrong way in seven weeks. The core-momentum leg held its side (3m annualized 3.05% on the July vintage, a fifth straight deceleration from 4.76% in February) and the Fed hiked anyway. That is the mechanism correction this note owes: the committee's reaction function is level-and-duration ('inflation too high for too long'), not the three-month momentum gauge the note was built on. Correct on direction at both events, wrong on instrument at both.

Bear claim 1 — 'The gamblers have left crypto for good': speculative money now goes to prediction markets and zero-day options instead, permanently

MOSTLY REJECTED. True this month, but 'for good' fails — speculative money has 'left Bitcoin forever' before every recovery in its history, and the prediction-market boom is smaller than it looks (~85-87% of Kalshi volume is sports). Weather, not climate.

Bear claim 2 — 'Bitcoin's power is worth far more running AI, so the energy is leaving mining for good'

TRUE, BUT ABOUT THE WRONG THING. The 10-20x gap and the conversions are real and permanent — for the mining stocks. But the difficulty adjustment makes mining more profitable whenever miners leave, so the network heals itself. A fact about ten companies' business models, not about the asset.

Sub-claim: energy is not actually leaving mining — hashrate is near its all-time high (~1,043 EH/s) even though revenue per unit of hashpower is down ~46% Y/Y

REJECTED — the middle step of the bear argument mostly hasn't happened, and even if it did the last step doesn't follow.

Sub-claim: the signed AI contracts are NOT a source of miner flexibility — ten companies, 19 deals, ~3.9 GW, ~$96B, almost nothing renewing before 2031

THE BULL CASE AS USUALLY TOLD FAILS. The accurate version that holds: flexibility lives in the ~20 GW of secured-but-uncontracted pipeline (~5x what is under contract) plus a yearly signing rhythm.

As the record 2024-25 capex vintages land, reported depreciation must grow sharply at every high-dose name regardless of schedule

CONFIRMED (Q2'26): depreciation accelerated 4-for-4 — MSFT +56%, META +46%, GOOGL +43%, AMZN +42% (AWS alone +67%) — against control stock Apple at +16%.

Guided capex holds or rises; the model assumed Microsoft would spend ~$115.0bn in FY2026

CONFIRMED: Microsoft FY2026 actual capex $115.9bn per the FY26 10-K — a 0.8% miss against the published assumption.

Companies keep five-to-six-year server lives through the fastest hardware-obsolescence cycle in computing history — thirteen disclosed extensions since 2020 against exactly one correction the other way (Amazon, Feb 2025)

CONFIRMED: no server-life change disclosed at any of the five plus Apple in the Q2'26 filings. Microsoft instead extended building lives 15 to 25 years effective FY2027 — the fourteenth extension against one reversal.

Rank the companies by dose (capex / net income) and you rank them by earnings overstatement — nine concordant pairs out of ten — and the dose is rising at every hyperscaler

CONFIRMED: every dose rose — MSFT 0.63 to 0.87, GOOGL 0.69 to ~1.19, META 1.15 to ~1.96, AMZN 1.70 to ~2.34; Apple's control dose held at 0.07.

FV ~$87/share (bear $32 / base $78 / bull $160) vs $109.86 — roughly 20% below the market

RIGHT WITHIN EIGHTEEN DAYS, THEN WRONG. HOOD closed 86.56 on 31 Jul 2026 - 44 cents below the 87 estimate, fourteen trading sessions after the call - then reversed and traded 118.89 on 3 Sep, 8.2 percent above the call and 41 percent above fair value. The estimate was reached; the recommendation has not paid. Correction: fair value should have read 84.53, not 87 - see the balance-sheet claim below.

The reported ~53x P/E understates the true multiple: FY25's 10.7% effective tax rate flatters it; normalized at 24%, the trailing multiple rises to ~62x

HOLDS, NARROWING. On trailing earnings through Q2 2026 the stock is 52.3x reported and 59.1x normalized at a 24 percent tax rate - the same argument, worth about seven turns of multiple instead of nine. The reason the gap is closing is that the effective rate is normalizing on its own: 10.7 percent in FY2025, 17.9 percent for H1 2026, 19.2 percent in the June quarter, 14.2 percent trailing. The claim survives and gets weaker every quarter.

Reverse valuation: $109.86 requires ~$3.2-3.8B of normalized net income within four years — 18-24%/yr compound growth, uninterrupted, through a Fed easing cycle. $110 is not a valuation, it is a growth commitment

HOLDS, HARDER. At 118.89 the market pays 105.5bn for the franchise net of corrected cash, which requires 4.2bn of normalized net income at a terminal 25x - 23.1 percent compound growth for four years - or 3.5bn at 30x, 17.6 percent a year. The hurdle is about three points a year steeper than at the call, and it now has to survive a Supreme Court petition against 11.9 percent of revenue.

The decoupling is demonstrated, not promised: Q1'26 crypto revenue fell 47% while total revenue grew 15%; rolling 90-day BTC beta fell from ~1.22 to ~0.88

CONFIRMED, AND HARDER THAN WE FORECAST. The June quarter repeated the test with a faster tape: cryptocurrency revenue fell 37.5 percent year over year while total revenue grew 32.3 percent - the growth beneath the collapsing line more than doubled versus the quarter we cited. Measured 90-session BTC beta is 0.94, against 0.88 stated at the call and 1.22 before. This is the claim the record has most vindicated.

Event contracts went $3M -> $104M in a year (~10% of revenue) but rest on contested legal ground; flexing that line swings fair value by roughly $15-25/share

CONFIRMED, BOTH HALVES, AND THE LEGAL HALF GOT WORSE. Event contracts reached 156m in the June quarter - 11.9 percent of revenue, above the 10 percent we described, and for the first time more than the entire crypto business at 100m. The contested ground moved against the industry inside six days: the Ninth Circuit created a circuit split on 28 Aug 2026 (KalshiEX v. Assad, No. 25-7516) and New Jersey petitioned the Supreme Court on 2 Sep (Flaherty v. KalshiEX). Robinhood is not a party to either case and had already stopped new Nevada sports contracts in Dec 2025 - but its own Ninth Circuit appeal was argued the same day as Assad, and its Nevada and Washington non-enforcement agreements are keyed to that litigation. The stock rose 14.0 percent from the ruling anyway.

Net interest is now the largest revenue line (~$359M/qtr, +24%) and is the one engine with a macro headwind: ~$294M of annual pre-tax net revenue per 100bp of rates

CONFIRMED, AND THE EXPOSURE GREW. Net interest is still the largest single line at 389m in the June quarter, up 9.0 percent. The disclosed sensitivity to a 100bp move rose to 350m of pre-tax revenue from the 294m we cited - up 19 percent in one quarter and 42 percent in a year, now 6.7 percent of annualized revenue. The headwind has not arrived: effective fed funds was 3.62 percent on 13 Jul and 3.63 percent on 1 Sep. The rate risk is larger than we said, and still unspent.

Balance sheet: corporate cash ~$5.0B against no funded corporate debt; net corporate cash ~$5.2B ($5.64/share) is a real floor under the equity

WRONG, AND IT WAS WRONG ON THE DAY WE PUBLISHED. Robinhood closed 2.2bn of 0.00 percent convertible senior notes due Oct 2029 on 25 Jun 2026, disclosed on Form 8-K on 23 Jun (pricing) and 25 Jun (closing) - twenty and eighteen days before this note published on 13 Jul. We built the balance-sheet layer off the March quarter and did not sweep the intervening 8-Ks. Correcting only the error and changing nothing else, net cash falls from 5,165m to 2,965m, or 3.25 a share instead of 5.64, and probability-weighted fair value falls from 86.94 to 84.53. The error ran in the company favour: overstating net cash overstates fair value, so every conclusion in the note was slightly kinder to the price than our own method warranted. This claim was never registered as one of the note six tracked claims; it is added here because the record should carry it. Terms, for completeness: zero coupon, 174.42 conversion price at a 65 percent premium, principal cash-settled, capped calls, and 290m of the proceeds used to repurchase 2.7m shares at 105.71.

The real risk is a slow funding drain: a permanent ~$1.55B/yr preferred cash coupon against a zero-yield asset with no operating free cash flow

UNDERSTATED — mstr-q2-2026 restates the coupon at $1.76B/yr, 13% higher in one quarter.

No margin-call channel exists — the debt is unsecured with no BTC pledged

PARTIAL MISS — mstr-q2-2026 finds 26 June WAS a margin event in shadow leverage built on top of STRC, where broker-dealers cut advance ratios and cascaded liquidations. Leverage migrated outside the balance sheet and was not modelled.

P(H.R. 3633 signed into law on or before 31 July 2026) ~13%, range 5-20% — a weighted blend of the Kalshi before-Aug contract (~18%), an outside-view reference class (~8%), a conditional-gate model (~6.3%) and converted full-year markets

CORRECT, and correctly below market. The before-August contract expired near zero (<1c on ~878k volume) with no Senate vote scheduled; resolved No on 1 Aug 2026. Both the direction of the below-market call and the reference-class logic behind it (a bill that had missed every prior deadline) scored. Scored in clarity-act-endgame v2.0.