Actionable insights — Weekly SSR: the dated de-risking plan, the per-megawatt contract valuation, the trim-half-into-the-gap rule, the negative-EV net-cash gates, the mispriced-as-tech short, and the positioning-bounce test

1. Set the de-risking date from the political calendar, publish it in advance, and let seasonality size it

1. Separate the reason a risk is currently quiet from the risk itself. Ask what is suppressing it — if the answer is a temporary constraint (munitions, an election, a funding window) rather than a resolution, put a date on the constraint's expiry and treat everything before it as borrowed calm.
2. Convert the date into a position schedule, not a forecast: name the months you will stay invested and the month you begin cutting. Announce it to the people who follow you before you act, so the decision cannot be rationalized later by whatever the tape is doing.
3. Check the schedule against base rates for the same calendar setup rather than trusting the narrative. Pull the median index return and the median peak drawdown for the equivalent historical years — the drawdown is the number that determines whether you hedge, because a positive median return with a deep median dip is a hedging problem, not a selling problem.
4. Express it as hedges added on strength plus tighter selectivity, not as a wholesale exit. "Add some hedges back on further market gains and get more selective."
5. Name the specific outcome that makes it worse, so you can update on evidence: which chamber flipping, which escalation resuming.

• Any sign the munitions/political constraint has lifted early (a resumption of large-scale strikes rather than sanctions); the Nov 3 result by chamber; Hormuz transit normalizing or, per Niles, Iran "holding the Strait of Hormuz hostage past the midterms"; and the 30-year auction — a repeat of last week's weak 5.3% tail is the funding-side confirmation.

2. Value a long-dated compute contract per megawatt — then buy the market's impatience about the start date

1. Reduce the headline contract value to a rent per megawatt per year (total contracted revenue ÷ megawatts ÷ years). This is the only number comparable across deals and across operators.
2. Apply the NOI margin the structure implies — if the tenant funds its own equipment and the landlord supplies shell and power, that is 80-90%, i.e. real-estate economics, not hosting economics.
3. Subtract the build cost per megawatt and capitalize the net operating income at a defensible multiple to get contract value per share. Compare that with the whole market capitalization: if one contract is worth most of the company, the rest is free.
4. Now read the timing schedule in the press release. Find the first-revenue date and the full-deployment date. Every year of delay is a year the market will discount, and an impatient market is where the entry comes from.
5. Enter after the announcement spike fades back toward the pre-announcement level, and size it as speculative — a contract that starts in two years is a financing and execution risk, not an earnings stream.

• The construction and energization milestones between now and Dec-2027 (the gap where the thesis can break); how the build is financed; whether the 286 MW upsize option is exercised; and the read-across signal from the same industry — CRWV's active power tripling to 1.5 GW against a $104B backlog is the benchmark for what "contracted megawatts actually delivered" looks like.

3. Trim half into an earnings gap in a speculative position — bank the windfall, keep the backlog

1. Decide the rule before the print: for a small, speculative, high-beta name, a pre-market gap of a defined size (here ~40%) triggers a fixed 50% trim — not a judgement call made while the quote is moving.
2. Execute in the pre-market, into the gap, and publish the alert. The liquidity is there precisely because the news is fresh.
3. Keep the other half against the part of the news that is multi-year. Separate the one-quarter numbers from the backlog/pipeline numbers: a revenue beat is a quarter, a backlog jump is years.
4. Re-check position size rather than conviction. The reason to trim is that a 40% gap has silently turned a small position into a large one — the trim restores the intended sizing, and says nothing about the thesis.

• Backlog-to-revenue conversion in the next two prints (a 533% backlog jump only matters if it becomes billings); whether the $14B pipeline converts to bookings; the buyback actually executing; and margin behaviour as low-margin legacy work is displaced by data-centre orders.

4. Screen for a negative enterprise value — then require an unwound cash burn and a path to the cash

1. Screen micro-caps where net cash ≥ market capitalization. That is the entry ticket, not the thesis — a negative enterprise value only means you are buying cash at a discount with the business attached.
2. Gate one: has the burn stopped? Cash inside a loss-making company is a melting asset. Require evidence the money-losing unit has been shut, sold or unwound — not a promise to fix it.
3. Gate two: express both halves per share. Split the value into net cash per share and business value per share, so you know which one you are actually underwriting.
4. Gate three: is the earnings stream concentrated? A single-product profit stream is exactly why the market refuses to capitalize it — accept the discount as structural, not temporary, and size accordingly.
5. Rank by how the cash returns to shareholders. Absent buybacks, dividends or a sale, cheapness can persist indefinitely.

• Any capital return announcement (the catalyst that closes the gap); further divestitures at disclosed prices; quarterly cash balance versus the prior quarter (the melting test); and, for GRVY, franchise revenue concentration and any new title.

5. Short the label, not the multiple — ask what the business would be called if it were named honestly

1. Find companies whose market classification and balance sheet disagree. Read the loan book / revenue mix, then write down the sector you would assign it with no name attached.
2. Identify the concentration: what single end-market or counterparty class does the asset side actually depend on? A one-sector book has no internal diversification to absorb a downturn.
3. Identify the macro sensitivity the label hides. A technology multiple prices growth; a lending book prices rates and credit. Where those diverge, the mispricing is the whole trade.
4. Quantify downside as the re-rating to the honest peer group, and state the macro condition it depends on so the thesis is falsifiable.
5. Size it as a core short only if the mispricing is structural rather than a single bad quarter.

• Non-performing loans and reserve builds in the concentrated segment; deposit costs and net interest margin; any disclosure that re-segments revenue toward the payments label (which would validate the market's multiple); and, for the promoter pattern, new share registrations or ATM programs following a price spike.

6. Compare your income book against the whole plain-vanilla fixed-income menu — then stress it by issuer quality, not in aggregate

1. Every year, print the year-to-date total return of each conventional fixed-income class — bank loans, EM hard and local, high yield, munis, MBS, investment grade, treasuries — and put your own book's return next to them. This is the benchmark test that decides whether a specialist book earns its complexity.
2. When a client asks about a rate shock, refuse to answer in aggregate. Split the book by issuer quality and give a separate drawdown estimate for the riskiest tier and the utility/regulated tier.
3. State the shock levels explicitly (a 10-year above X, a 30-year approaching Y) so the estimate is testable rather than reassuring.
4. Ask whether a policy reaction function caps the shock. If the level that damages your book is also the level that forces an official response, the drawdown is short-lived and the correct action is to hold through it, not to hedge it away.
5. Keep the highest-quality tier large enough that the whole book survives the tantrum you cannot avoid.

• The 10-year through 5% and the 30-year through 5.5% (the stated trigger levels); the next 30-year auction tail after last week's 5.3%; whether any official-sector pushback actually appears at those levels; and the spread between low-grade and utility preferreds as the tell that the market is discriminating by quality.

7. Strip non-operating mark-ups before you accept an earnings season — and re-derive the growth rate yourself

1. Take the index's headline beat rate and aggregate surprise, then find the two or three companies responsible for most of it — extreme index-level statistics almost always resolve to a handful of names.
2. For each, split reported EPS into operating and non-operating. Unrealized gains on private stakes are value changes, not cash — no share was sold.
3. Recompute the index's surprise and growth rate excluding those names and use the ex-figure as your working number.
4. Separate revenue growth from earnings growth as the honesty check: revenue cannot be marked up. If revenue growth is genuinely strong, the underlying quarter is real even when the EPS optics are inflated.
5. Then ask what happens to the comparison: a mark-up can only repeat if the private company raises again at a higher price.

• Whether Anthropic's next round is marked higher (the only way the gain repeats); 2027 consensus EPS being revised for the non-repeat; the ex-IT-and-energy revenue line as the real economy read; and the divergence he flags underneath all of it — nominal GDP growth racing ahead of aggregate weekly payrolls, where "one of these two things will happen" and "the first thing to break will probably be AI growth, and then second will be the consumer."

8. Trade consumer behaviour, not consumer attitudes — and check who is doing the spending

1. When a sentiment survey and a spending series disagree, discard the survey. "Today, we are most interested in consumer behavior, not consumer attitudes." Sentiment worked in past cycles; it has not for five years, so it is no longer a forecasting input.
2. Ask which decile is spending. If the top of the income distribution accounts for the majority of consumption, aggregate spending tracks asset prices, not wages — and the median household's misery is irrelevant to the retail-sales print.
3. Re-read the savings rate accordingly. In a strong economy, a falling savings rate is a confidence signal from the wealthy, not a distress signal from the poor.
4. Identify what would actually break it: the same wealth effect running in reverse (equities and home prices), an energy or food shock, or a labour-market crack — not a worse survey.
5. Test the read against company results, not more macro data. Pick the single reporter whose customer base is the marginal consumer.

• Retailer commentary on consumer weakness in the coming week rather than the macro print; the savings rate against household net worth; and the fourth cause of the July miss — labour-market cooling — which he says "will likely see that in the August numbers," i.e. the print that separates calendar noise from a real slowdown.

9. Decompose a data miss into calendar distortion, price effects and genuine demand before you trade it

1. Go straight to the category contributions, not the headline. One or two lines usually explain the whole surprise.
2. Strip calendar and promotional distortions — a promotional event moved between months creates a mirror-image beat and miss with no change in demand.
3. Strip nominal price effects. Where the series is not inflation-adjusted, a fall in the price of a good (gasoline) shows up as a fall in spending on it.
4. Separate pull-forward (incentives rolling off, refunds spent) from a change in the spending rate.
5. Only what survives all three is a demand signal — and rank the surviving causes by which will still be there next month.

• The August print with the Prime Day mirror image removed; the control group (ex autos, gas, building materials, food service) as the GDP-relevant line; restaurant spending as the discretionary tell (it rose +0.5%); and memory-driven goods prices as a new, non-demand source of nominal distortion.

10. Underwrite an arb by the legs retired — and read a rising break fee as a commitment signal

1. List every discrete obstacle between announcement and close — each antitrust filing, each state or sector regulator, each affected constituency with standing to object, each vote.
2. Each week, mark which legs were retired rather than re-forecasting the odds. Spread tightening should follow legs falling, not headlines.
3. Watch for the acquirer buying off an objecting constituency with a contractual concession — that is how a blocked deal becomes a settlement, and it converts opponents into supporters who will lobby for approval.
4. Track process facts that raise the buyer's cost of walking away: a go-shop expiring, a break fee stepping up, an HSR re-filing that resets a clock to a known date. These are commitment signals, and they are public.
5. When a competing bid fails, read why. Financing, change-of-control covenants and a required equity rollover are structural failures — they do not come back at a higher price.
6. Keep a dated milestone list (votes, waiting-period expiries) as the working calendar for the whole book.

• Whether the California suit actually settles (the last large WBD leg); the Sept 8 HSR expiry on Roku/Fox; the Sept 8 and Sept 14 votes; and, on Caesars, the Sept 14 HSR expiry after the buyer's re-filing.

11. Read the "Background of the Merger" section — it prices how thin the bidder list really was

1. When a proxy or S-4 lands, go to the background narrative before the financials. It is the only public account of the auction.
2. Count the parties contacted, the parties that engaged, and the parties that bid. A large canvass with one bidder means the announced premium is the ceiling, not the floor — there is no topping bid coming.
3. Trace the price path from first indication to final terms, and note the consideration mix. It tells you how much the buyer had to concede, and therefore how motivated it is.
4. Note the reasons the declining parties gave — competing priorities, financing, integration — as a read on the industry's appetite for the whole asset class.
5. Use the same document as training material: the sequence of committee formation, market check, diligence and board approval is the anatomy every future deal repeats.

• Any go-shop or topping-bid provision in the merger agreement (given only one bidder ever emerged, the odds are low); the exchange-ratio exposure created by the 40% stock component; and the same background-section discipline applied to the Caesars proxy, where it revealed the failed Icahn bid.

12. Distinguish a positioning bounce from a fundamental one — and check the rental market before shorting a shortage

1. After a forced liquidation, mark the date and measure each name's bounce off that low. That number, not the valuation, is the risk carried into the next print.
2. Watch how the tape reacts to unambiguously good results. A beat-and-raise on both revenue and EPS that sells off means the news was already in the bounce — the mechanical repair is finished.
3. Rank by bounce size, not by fundamentals: the biggest bouncers are the biggest post-earnings risks even when the quarter was fine.
4. Before shorting on valuation, price the underlying product in the secondary market. Rental or spot prices for the scarce input tell you whether demand is real independent of the equity narrative.
5. If the input price is rising, a valuation short is a bet on sentiment. Wait for the input price to turn.

• GPU rental and on-demand rates turning down (the signal that the shortage, and therefore the short case, has changed); the same forced-liquidation arithmetic running the other way in software, where the unwind of the short-software/long-semis book produced TEAM +35% and a WDAY LBO bid putting "a floor on the software sector" via IGV; and Nvidia's residual-value guarantees — capped at 25% per project — as the place the circularity charge would resurface if lease rates fall.