<!-- Generated from src/ data by scripts/generate-llm-files.ts. Do not edit by hand. -->
# The week in signals: Jul 27–31

> The standing Friday scorecard, in the loudest earnings week of the season. The AI-infrastructure names the credible rooms have been long all month fell as much as 21% by Wednesday, then Microsoft's blowout print snapped the whole trade back on Thursday. The honest part: one earnings report whipsawed the entire tape, and the biggest mover of the week, Microsoft itself, walked in with a quiet, undecided room the score never called.

By Maya Koeva · 2026-07-31

![A glossy chrome desk calendar page with a glowing signal waveform running across its face, marking a week of calls against outcomes.](/blog/week-in-signals-jul-27-31.png)

The week-end scorecard, the same way we did it for [June](/blog/june-signal-scorecard) and every
Friday since, with the same rules. The score does not try to time the market. It grades who is
talking about a name and which way they lean, weighted by how
[credible](/learn/what-is-a-credibility-score) those voices have been. The Friday question is
narrower: where the score and the crowd genuinely disagreed this week, who was right so far?

This was the densest earnings week of the season. Tesla and Alphabet reported the week before,
Microsoft and Meta on Wednesday night, Apple and Amazon last night, with a Fed decision wedged in
on Wednesday afternoon. Two things about the data before we grade. First, this was really two
weeks in one: a brutal three-day selloff in the AI-infrastructure trade Monday through Wednesday,
then a violent reversal on Thursday after Microsoft reported. Second, of the four that reported
this week, only Microsoft's post-print tape has settled in our set so far, so we grade Microsoft
below and hold Meta, plus Apple and Amazon from last night, for next Friday. Every price below
runs from the July 27 close to the July 30 close, the last completed session before we pulled the
data.

## What the signal caught

All month the credible rooms in the accounts we track have leaned one way: not on the megacap
spenders, but on the [AI-infrastructure names those budgets get spent with](/blog/microsoft-meta-fed-day).
This was the week that thesis got stress-tested to the breaking point and then rewarded inside
three sessions.

By Wednesday's close the whole trade was in free fall. Nebius, the single loudest name on our
board this week at 221 mentions, was down 21% from Monday. Micron down 18%. MaxLinear, the
[autopsy we ran on Monday](/blog/signal-autopsy-maxlinear), down 12%. One account we track called
it the worst drawdown on record for Goldman's high-beta momentum basket, worse than 2000 or 2008.
Through all of it the credible rooms did not flip: Nebius held a score of 79 at 74% bullish,
Micron 84 at 63% bullish, both still net long into the teeth of the drop.

Then Microsoft reported Wednesday night. Azure grew 43%, cloud crossed $100 billion in annual
revenue, and enterprise backlog jumped 84%, the clearest proof yet that the AI spending is
turning into AI revenue. That was the demand signal the entire complex trades on, and Thursday it
snapped back: Nebius round-tripped 27% off its Wednesday low to close the week almost exactly
where it started, Micron bounced 18%, MaxLinear 17%. The names ended roughly flat on the week,
but the path is the point. The score pointed at where the credible conviction sat, that
conviction held through a 20% drawdown, and Thursday it got the last word.

| Company | Score | Mentions | Trusted | Week move |
|---------|-----:|:--------|-------:|--------:|
| Nebius (NBIS) | 79 | 221 (bull 74% / bear 14%) | 26% | +0% |
| Micron (MU) | 84 | 107 (bull 63% / bear 24%) | 11% | -3% |
| NVIDIA (NVDA) | 78 | 137 (bull 50% / bear 35%) | 20% | -1% |

*[Chart: Nebius daily close: down 21% to Wednesday's low while the credible room stayed 74% bullish, then a 27% snap-back on Thursday after Microsoft's print. A full round trip in five sessions.]*

## Where the volume was hollow

Apple had its loudest week in our entire coverage: 107 mentions, nearly five times its count from
early in the week. And the score went the other way. It read 81 on Monday, cooled to 45 by
Thursday, and sits at 24 this morning, [as we walked through yesterday](/blog/apple-amazon-close-the-gauntlet).
The reason is in what the extra volume was made of. The crowd that arrived to watch Apple pass
Nvidia and touch a $5 trillion market cap brought milestone chatter, not directional conviction,
and the bullish share fell as the room grew. This is the cleanest [volume-is-not-signal](/learn/volume-vs-signal)
case of the week: the headline count tripled while the credible read got weaker, not stronger.
Apple reported last night, so the price verdict lands in next Friday's scorecard, but the score's
job this week was to tell you the loud new crowd was not a convinced one, and it did.

## The honest part

Now the section this format exists for, and this week it earns its keep. A single earnings print
whipsawed the entire tape, and that makes any one-Friday read noisier than usual. Three things we
will not dress up:

| Company | Score | Mentions | Trusted | Week move |
|---------|-----:|:--------|-------:|--------:|
| Microsoft (MSFT) | 83 | 126 (bull 56% / bear 20%) | 13% | +15% |
| Oracle (ORCL) | 58 | 44 (bull 18% / bear 61%) | 11% | +6% |
| SanDisk (SNDK) | 70 | 118 (bull 32% / bear 50%) | 17% | +0% |

First, Microsoft. It rose 15% on the week, the biggest winner on our board, and the score did not
call it. Going into Wednesday its room was quiet and thin: a score of 68 the morning of the print,
with just 15% of its mentions from trusted accounts and the crowd split. The score firmed to 83 *after* the
blowout, which is it reacting to a great quarter, not predicting one. We take no credit for
Microsoft. Meta is the mirror image: its score fell from 61 to 35 as the room turned 51% bearish,
again the score moving with the news rather than ahead of it.

Second, Thursday's reversal lifted almost everything, including names the score is cold on. Oracle
carries a 58 and a 61% bearish room, and it rose 6% on the week anyway, most of it in Thursday's
melt-up. A rally that broad is exactly the kind of week where a single Friday snapshot proves
least.

Third, the catch we led with hung entirely on one print. The AI-infrastructure names were down
20% on Wednesday and *needed* Microsoft to deliver. It did, this time. SanDisk is the reminder in
the other direction: 118 mentions but a room that turned net bearish this week, 50% to 32%, its
score cooled to 70, and it round-tripped to flat like the rest. The score reads where conviction
sits. It does not promise the conviction is right, and a week that turns on one earnings call is
the honest proof of that.

## The takeaway

Strip out the whipsaw and the week comes down to one contrast. Where the score pointed at credible
conviction, the AI-infrastructure receivers, that conviction survived the worst drawdown the trade
has had and got Thursday's last word. Where the crowd got loud without conviction, Apple at its
$5 trillion milestone, the score cooled to tell you so. Everything in between, Microsoft's own
blowout, Oracle's broad-market bounce, was the tape reacting to a single print, and we grade those
reactions honestly: the score followed them, it did not foresee them. The reads that will actually
settle this week, Meta, Apple, and Amazon's own post-print tape, we grade next Friday.

For the method behind these calls, see [how a track record is graded](/learn/how-a-track-record-is-graded)
and [volume versus signal](/learn/volume-vs-signal).

---

*Quantral surfaces signals and context from public sources to support your own research.
Nothing here is financial advice or a recommendation to buy or sell. Past signals are not
indicative of future results.*
