There are 27 new buy signals from the week that closed on Friday, and this is their second day of life: the moment you see which have started working and which have not. Below are 12 of them, ranked from best to worst by return from the signal, and the horizon throughout is weekly swing trading. Three families hold them together: gold and its miners, large-cap technology, and the travel-linked names.

Tuesday's session is a textbook case of how a single variable can move the whole expectations curve. On Friday a very weak jobs print — 23,000 positions lost — had wiped out bets on a Federal Reserve hike in September and pulled the US ten-year down to around 4.61%. Three sessions later the yield sits above 4.72% and the thirty-year has cleared 5%. Nothing happened to the economy in between: crude rose, for a fourth consecutive session, with the Strait of Hormuz still paralysed.

On the board, and price action trading gave it away before the headlines did, refining did the heavy lifting. Marathon PetroleumLong gained 5.03%, ValeroLong 2.85%, ConocoPhillipsLong 2.35% and ChevronLong 0.90%. This is the second straight day for the theme, and it is the confirmation we were looking for: a geopolitical premium that lasts one session usually fades, one that lasts two starts becoming a price.

The technical analysis that matters, though, is about flows, and it is the same read we wrote on Monday. On ExxonMobilLong and on the Vanguard energy ETFLong weekly money keeps leaving while price rises: on Tuesday they closed at plus 0.01% and plus 1.11%, effectively flat, on the very day the sector ran. On ChevronLong and ConocoPhillipsLong the flow holds, and they duly followed the theme. Telling distribution from accumulation inside the same sector rally is the only way to choose among eight names rising together, and it is the work an ai trading system does by scanning the entire universe before picking a direction.

One note on entry price, because that is where the edge is lost: Marathon PetroleumLong travelled roughly five times its own typical swing in a single day. The theme can be right and the moment wrong, and that is precisely where risk management matters more than direction.

The day's most spectacular move came from Riot PlatformsShort, with Riot stocks up around 17% in the session and as much as 25% after the close, after leasing its Rockdale campus in Texas to Anthropic: 191 megawatts of capacity and $9.1bn of contracted revenue through June 2048, rising to $16.1bn with renewal options. As a trading strategy read it confirms a structural trend — bitcoin miners turning themselves into compute providers — and it dragged the whole group along: TeraWulfShort up 3.40%, IRENShort up 2.61%, Applied DigitalShort up 2.17%.

On the same wave, after the close, CoreWeaveShort reported quarterly revenue of $2.6bn, an annualised run-rate of $10.4bn, a backlog up to $129.2bn and a 59% adjusted operating margin. The stock had already added 2.42% during the session.

Here our position sits on the opposite side of the move, and that comes first: the weekly signal on all five of these names is a sell, and on Tuesday every one of them rose. It is worth explaining why we do not turn it on a headline. A twenty-year contract changes one company's income statement; it does not change the price structure of the other four, which are what they were the day before. Our protocol works on the weekly signal, and a signal does not flip because a competitor signed a deal: it flips when price does, and the level where that happens is written in each instrument's note.

The day's biggest loser was AlphabetShort, down 3.84%: Alphabet stocks carried the session's heaviest bill. Three things weighed together: capital spending for 2026 revised up to the $195-205bn area, with quarterly free cash flow turning negative for the first time; a $25bn bond placement to fund data centres; and a widening regulatory front, between the complaint filed by nearly three hundred French publishers over AI-generated summaries and the federal prosecutors' appeal on the search antitrust remedy. It is the paradox of this phase: the spending meant to secure leadership gets punished in the short run because it compresses margins.

The rest of software followed. AppLovinShort shed 5.99% under a wave of downgrades from three large brokers, arriving after a quarter that missed on revenue — its first guidance stumble since listing. DatadogLong lost another 5.37%, extending its post-earnings weakness: guidance was raised, but concerns remain over slowing growth and a compressing cash margin. AdobeLong gave up 3.39%. The message is the same for all three: on high-growth names with stretched valuations, numbers that are good but not exceptional get sold.

Among semiconductor stocks the session held: NVIDIALong essentially flat at minus 0.02%, AMDLong up 1.01%, MicronShort up 0.87%. The rotation hit software and advertising rather than silicon.

Beneath indices giving up three tenths, money moved in an orderly and recognisable way: leading the session, the US energy ETF at plus 1.25% and the utilities ETF at plus 1.16%; at the bottom, communications at minus 0.50% and real estate at minus 0.72%. Utilities and energy rising together while bond yields run is the classic signature of a defensive rotation, and it explains why US small caps added 0.43% while the large indices slipped.

On individual names the move was broad. Edison InternationalShort gained 3.04%, Constellation EnergyLong 2.93% and PG&ELong 1.53%; on the energy side Marathon PetroleumLong up 5.03% and ValeroLong up 2.85% led the sector, moving more than twice as far as their own basket.

Our model is split here, and it is worth explaining why. On Constellation Energy and PG&E the weekly signal is a buy and the session rewarded it. On Edison International it is a sell, and the stock rose 3%: when the model stands against the move, the correct read is not that the market will come back, it is that the name should be kept small until one of the two gives way. On utilities there is also a context caveat — a sector rising because yields are rising has a fragile relationship with rates, and the data that moves them lands today.

The heaviest note came from São Paulo. JPMorgan cut Brazilian equities to neutral from overweight, trimming its year-end Ibovespa target to 185,000 points from 190,000. Four reasons were given: a rate cycle close to its end, with only one further 25-basis-point cut expected in September; slowing economic activity; a deteriorating credit cycle; and the historical underperformance of Brazilian assets in the six months before an election. The index closed down 2.5%, a sixth consecutive losing session.

Selling was broad and orderly by sector: banks with Itaú UnibancoShort and Banco do BrasilShort around minus 2.5%, steel with GerdauLong at minus 4.32%, oil with PetrobrasLong at minus 2.18%, and across our universe ItausaShort at minus 3.7% and ValeShort at minus 2.02%. Two of the day's six outsized moves were Brazilian, which is the signature of a market falling as a block: when the information is the country, individual names stop telling different stories.

Something worth isolating happened to our model. The weekly signal on Itaú UnibancoShort, Banco do BrasilShort, ItausaShort, BradescoShort, CosanShort and AmbevShort turned to sell on this week's candle, at the very moment the downgrade reached the market. That coincidence needs stating precisely: those signals are still forming and only confirm at Friday's close. On ValeShort the sell has instead been active and confirmed since May.

In Milan the main basket closed soft but above par, held up by two themes. The first is energy: EniLong was the best on the board at plus 2.01%, hooked directly to the surge in crude. The second is defence, with LeonardoLong up 0.75%. PrysmianShort also did well at plus 1.35%, on the grids and data-centre theme.

On the other side, decisive selling on StellantisShort at minus 2.60%, and a banking sector slightly lower: UniCreditLong and Intesa SanpaoloLong both around half a point down, Banco BPMLong at minus 0.88%. EnelLong closed essentially unchanged at minus 0.91%. Worth noting on the bond side: the Bund at a fifteen-year high and broad selling across European government paper, consistent with what was happening to US Treasuries.

A technical note on the banks, because that is where our position is most exposed. The weekly signal on Intesa Sanpaolo, UniCredit and Banco BPM has been a buy since April, and the three positions are up 20.86%, 24.66% and 33.84% respectively from the signal. The relative strength reading on the weekly is above 70 on two of the three, though: these are mature trades, where the useful work is protecting the result rather than adding to it.

Four baskets defined the session, and they are worth looking at as instruments rather than as thermometers. The US energy ETFLong on US energy gained 1.25% and is the most direct way to hold the crude theme without choosing among eight refiners; over the month it carries plus 10.62%, the widest move on the whole sector map. The European oil and gas ETFLong on European oil and gas did even better on the day at plus 1.63%.

On the defensive side, the US utilities ETF added 1.16% on a day of rising yields — a combination you do not normally see, and one that signals rotation more than conviction. At the other end of the map, the US real estate ETFLong on US real estate lost 0.72% and is the only sector negative on the day, the week and the month alike.

Two baskets stay central even without being the day's news. Gold ETFs are how you take the metal theme while paying less risk than the miners carry: SPDR Gold SharesLong sits among this week's new buy signals and has the lowest typical loss of the group at 2.3%. And the semiconductor ETF remains the way to hold silicon without depending on the single regulatory constraint that kept NVIDIA stock flat on Tuesday: US technology ETFLong on US technology closed at minus 0.12%, with its weekly signal turning to buy on this very candle.

The calendar for the coming days is dense and concentrated in two blocks. The first is Brazilian, and it lands in the worst possible week: banks, utilities and steel report between today and Friday while the index falls for a sixth session and the country has just been downgraded. The second is American and closes next Tuesday with retail distribution.

Wednesday 12 · Banco do BrasilShort · BraskemShort · E.ONLong · UltraparLong · Virgin GalacticShort

Thursday 13 · JD.comLong · Intuitive MachinesShort · CPFL EnergiaShort · BradesparShort · OndasLong

Friday 14 · CosanShort · PDG RealtyLong

Tuesday 18 · Home DepotLong · BaiduShort

The name worth pausing on is JD.comLong, reporting tomorrow. It is one of Tuesday's six outsized moves, minus 4.6%, more than twice its typical swing, and at the same time our pre-earnings lens assigns it five active clues: improving money flow, rising buy volume, lows that have stopped falling, compressing volatility and a pattern of higher lows. Five favourable clues and one heavy bar on the same day do not cancel out: they make the name interesting and unsuitable for an entry before the print.

One point of method we make every time. Our measurement of the July season says entering with an active signal ahead of results returns 2.12% on average against 2.03% for the control group: earnings do not move the average return, they widen the dispersion. They are therefore not a reason to enter, nor a reason to stay out — they are a reason to enter smaller.

The four European baskets we follow all have an active weekly buy signal dating from the first half of April, so seventeen weeks, and all four sit above their own weekly Point of Inversion. That is where the similarity ends. The difference lies in what they have produced and in how far they are from their highs.

FTSE MIBLong is the best of the four at plus 15.78% from the signal, and stands 0.58% below its record. CAC 40Long follows with plus 7.76% and minus 0.40% from its high. DAXLong carries plus 6.48% and is the closest of all to its peak at minus 0.33%. Three baskets glued to their records, with distance from the weekly Point of Inversion between 2.52% and 3.12%: a thin but even cushion.

The fourth is the one that says something. FTSE 100Long on the UK basket has produced plus 1.79% in seventeen weeks, a fifth of the French figure and a ninth of the Italian, and sits 7.91% below its record — twenty times the distance of the other three. That is the divergence worth watching: when three indices on one continent are at their highs and the fourth is almost 8% below, the fourth is not taking part in the same move, and its distance from the Point of Inversion, 1.87%, is also the thinnest of the group.

On Tuesday SPYLong, the S&P 500 ETF, and QQQLong, the Nasdaq 100 ETF, lost almost the same — minus 0.32% and minus 0.34% — and sit in two completely different technical positions.

SPY is 0.81% below its record, with the buy signal active since 6 April, so in its eighteenth week, and plus 13.41% from the level where it fired. All three of our system's profit-taking windows have been reached, the last in week seventeen: there are no targets left to publish ahead, and what governs the trade is the rising protection alone. The day's technical fact is something else, though: for four sessions price had stayed inside its congestion band, and on Tuesday it left it from below. The zone went from congestion to resistance, and buy volume fell to 25.14% of the total.

QQQ sits at the opposite end. It is 4.03% below its record, the third profit-taking window was never reached, and on the forming weekly candle our model has printed a buy signal after four weeks in which the declared signal was a sell. The weekly Point of Inversion dropped thirty-six points at once, which means it is the model changing side rather than price having moved. It is provisional and closes on Friday.

The reason for the divergence is in the basket. QQQ weighs over 50% in technology and just 0.24% in financials; SPY 32.91% and 12.59%. On Tuesday energy and utilities gained over 1%: together those two boxes are worth 6.56% in SPY and 2.20% in QQQ. When money rotates into defensives, the Nasdaq has nowhere to park it.