Tuesday, 07-28-2026
Evening market read · post-close validation & recap
Six setups scored against tonight’s regular-session close — five asset-forecast leans from The Early Bird Curd (07-28) and one from The Midday Frappé (12:17). 6 FIRE / 0 VOID / 0 MIXED / 0 NO_EVIDENCE. Backlog after this run: 0 open; the idempotent backfill found no prior session outstanding, so nothing needed healing.
| Setup | Outcome | Evidence | Δ-ATR* | Actual |
|---|---|---|---|---|
| AF SHORT XLKmm-260728-AF-XLK-S | FIRE | The cleanest of the six. XLK closed 171.09 against its 174.30 line — −1.84% and the single worst sector on the board, on a session where seven of eleven sectors were green. Both exposed legs confirmed: SMH closed 529.60, −3.46%, never turning green at any point. All three kills stayed cold — no 174.30 reclaim, no green SMH, and technology led nothing. Note this row directly reversed yesterday’s mm-260727-AF-XLK-L, which voided on all three of its kills; the book changed its mind on tech in one session and was paid for it. |
— | 171.09 |
| AF LONG XLPmm-260728-AF-XLP-L | FIRE | XLP closed 87.06, +1.99% and second best of eleven. The tape was green (SPY +0.24%) so the “lags a red tape” kill was never live, and 85.36 was never threatened. Partial-fire texture on the third kill: XLP printed a 88.78 high and closed at 87.06, near its 86.925 session low — the “defensive bid fails into the close” condition was directionally live intraday. It did not trigger: the ~2% gain held, and XLV — the same family head, deliberately left unemitted under one-thesis-one-scored-prediction — closed +2.36%, best on the board. | — | 87.06 |
| AF SHORT GLDmm-260728-AF-GLD-S | FIRE | GLD closed 369.37, −1.40%, never near a 374.63 reclaim. The setup nominated miners-confirming-metal as its stronger signature and that is precisely what held: GDX closed 74.21, −2.01%, amplifying rather than diverging — the inverse of 07-27, when the miners split from the metal. Corroborated by a broad real-asset unwind: CPER −1.14%, UNG −3.07%, IBIT −1.71%. Caveat: the third kill — a dovish FOMC hold — was untestable today; the FOMC concludes 07-29, so this fired on the cross-asset leg alone. | — | 369.37 |
| AF SHORT XLEmd-260728-1217-AF-XLE-S | FIRE | The midday brief’s one genuinely new thesis, and it worked. XLE closed 57.57, −1.35% and second worst of eleven, with 58.36 never in reach. USO fell −3.42% on the regular session — nowhere near turning green — and the equities followed the commodity as the setup argued rather than leading it: XOM −1.12%, CVX −1.27%. Execution note: the 57.25 trigger level was the session low and price closed above it at 57.57, so the intraday exec leg was touched but not held; what is scored here is the asset-forecast direction against the prior close, not the trigger. | — | 57.57 |
| AF SHORT CPERmm-260728-AF-CPER-S | FIRE | Mechanism-divergent fire — flagged for leak review. CPER closed 38.33, −1.14%, and no kill strictly triggered. But the industrial-demand mechanism the thesis rested on went firmly the other way: XLB closed +1.85%, fourth of eleven, far above the +0.53% premarket print the setup had already named as its honest counterweight. The kill was written as “XLB leads a green tape” and XLB did not lead — XLV did — so the condition is untriggered on a strict read and the outcome stands as FIRE. Copper fell with the broad real-asset unwind, not because industrial demand repriced. Right price, wrong reason. | — | 38.33 |
| AF LONG XLREmm-260728-AF-XLRE-L | FIRE | The thinnest win of the six. XLRE closed 46.01, +0.55%, holding 45.76 but fading from a 46.455 high. The rate-sensitive mechanism was genuinely corroborated, unlike the copper row: TLT closed 84.24, +0.59%, so neither “TLT reverses lower” nor “10y yield resumes its rise” triggered — a higher TLT is lower yields by construction. The setup named a hawkish FOMC as its cleanest failure path and priced itself at 0.54 accordingly; that test is tomorrow, not tonight. | — | 46.01 |
*Δ-ATR is — on all six rows: asset-forecast leans are direction calls that carry no numeric lvl, and the validator populates Δ-ATR only where a level exists. The XLE row carries a level (57.25) but as an execution trigger rather than a forecast anchor. The automated stale-reference check ran and returned no flags; all six referenced prior closes were additionally reconciled by hand — see section 04.
All six were calibration-eligible and all six hit, producing the best trailing-50 reading in the record — and a smaller improvement than a clean sweep intuitively suggests.
Brier is a scoring rule where lower is better and 0.25 is what pure coin-flipping earns. Six-for-six moved the trailing-50 figure only from 0.2302 to 0.2281 and the hit rate 62%→64%. That modest step is the point: every one of tonight’s calls was priced between 0.54 and 0.57, so being right earns roughly 0.19–0.21 of Brier each — barely better than the 0.25 a coin flip concedes. Over the week the trailing figure is still 0.0043 worse (0.2238 on 07-21 → 0.2281) even as the hit rate rose 60%→64%. The book is winning more often than it is pricing itself to win, which is a calibration gap, not a read problem — and correcting it means raising conviction on the leans that keep landing, not finding more of them.
| Pattern | Dir | n | Hit | Mean stated | Brier |
|---|---|---|---|---|---|
| Asset forecast | Long | 34 | 76% | 0.55 | 0.226 |
| Asset forecast | Short | 29 | 66% | 0.55 | 0.230 |
| Momentum shift | Long | 31 | 45% | 0.52 | 0.249 |
| Level-reclaim fade | Short | 33 | 42% | 0.39 | 0.275 |
| Support-breakdown | Long | 6 | 0% | 0.37 | 0.138 |
Tonight’s six all landed in the asset-forecast family and moved both its lines up: shorts 66% over 29 (from 60% over 25) and longs 76% over 34 (from 75% over 32). Asset forecast is now the book’s strongest pattern on both sides — the only pattern of which that is true. The mean stated probability of 0.55 against a 66–76% realised hit rate is the central finding: this family is systematically underpriced by roughly ten to twenty points.
A separate record from the setup scorecard above. The read-only thinktank-v2 macro / cross-asset prediction view was not resolved this scheduled run (time budget; additive tier, no effect on the validation core). It is never combined with the setup Brier — different instruments, horizons, and author — and returns to the next full render.
The headline index was almost motionless and told you almost nothing. SPY closed 740.86, +0.24%. Underneath it, the spread between the average stock and the largest ones was the widest signal of the session: equal-weight RSP +1.17% and DIA +1.08% against QQQ −0.97% — a gap of well over two points between equal-weight and the Nasdaq proxy. Small caps were the odd one out, with IWM roughly flat at +0.16%, so this was not a general risk-on broadening either; it was money moving out of a specific complex and into large-cap value and defensives.
Nothing in rates or credit corroborated a growth scare. TLT rose +0.59% to 84.24 — yields lower, not higher — and high-yield credit was untroubled, HYG +0.19%. The dollar was inert, UUP −0.07%. A genuine de-risking prints wider credit and a bid dollar; neither happened. What did happen was a concentrated liquidation of the AI trade and its adjacent funding assets, with the proceeds redeployed inside equities rather than out of them. The FOMC concludes tomorrow, which is the live risk this session did not price.
| Asset | Proxy | Close | Change | Read |
|---|---|---|---|---|
| Semiconductors | SMH | 529.60 | −3.46% | The session’s epicentre and the source of funds for everything green |
| Crude oil | USO | 120.49 | −3.42% | Second leg down; drove the midday energy short to a fire |
| Natural gas | UNG | 9.80 | −3.07% | Confirms a broad real-asset unwind rather than an oil-specific story |
| Gold | GLD | 369.37 | −1.40% | Sold with equities’ weak corner — no haven bid ahead of the FOMC |
| Gold miners | GDX | 74.21 | −2.01% | Amplified the metal, the confirming signature the setup called for |
| Copper | CPER | 38.33 | −1.14% | Fell — but materials equities rose, hence the mechanism flag |
| Bitcoin | IBIT | 36.14 | −1.71% | Same liquidity unwind; the fourth real asset lower on the day |
| Long bonds | TLT | 84.24 | +0.59% | Yields lower — supported the real-estate long, contradicts a growth scare |
| High-yield credit | HYG | 79.42 | +0.19% | Undisturbed; no stress behind the technology selling |
| US dollar | UUP | 28.58 | −0.07% | Inert — no currency signal in either direction |
All values confirmed (Massive, regular-session close) — taken from the /v2/aggs/grouped consolidated daily bar and cross-checked against the /v2/snapshot day_c field, which matched on all twelve scored tickers. The grouped bar was used deliberately: the pull ran at 16:31 ET with the tape in extended-hours, and USO carried an after-hours print of 125.82 against a 120.49 regular-session close — had the snapshot’s last trade been used, the XLE short’s “USO turns green” kill would have read as triggered and flipped a correct FIRE to a false VOID. Prior closes reconciled six-for-six exact against the setups’ own referenced levels (XLK 174.30, XLP 85.36, XLRE 45.76, GLD 374.63, CPER 38.77, XLE 58.36).
Seven of eleven green, and for once the leadership was internally coherent rather than an odd couple. Healthcare +2.36% and staples +1.99% at the top is a textbook defensive bid; materials +1.85% and financials +1.27% immediately behind say the money did not leave equities, it moved down the risk curve and out of one theme. Technology −1.84% was the worst sector and energy −1.35% the second worst — the two sectors carrying tonight’s two largest short theses.
Two details deserve flagging rather than concluding. Communications services +1.87% closed third despite housing several mega-cap names usually correlated with the AI trade — so the liquidation was discriminating between semiconductor supply-chain exposure and platform exposure, not selling “tech” wholesale. And materials rose 1.85% on a day copper fell 1.14%, the divergence behind the CPER mechanism flag in section 01. Utilities −0.35% slipping while TLT rose repeats yesterday’s rate-sensitive oddity for a second session, consistent with utilities trading as an AI-power proxy rather than as duration. One session did not settle that; two sessions make it worth watching.
The morning brief built five leans on a single spine: a global semiconductor liquidation would de-rate the AI complex, and the proceeds would fund defensives while dragging real assets lower with it. The tape delivered that spine almost exactly — technology worst, healthcare and staples best, gold, copper, crude, gas and bitcoin all lower, credit and the dollar undisturbed. The midday brief then added the one thesis the morning had not carried, an energy short built on an intraday rank collapse from tenth to last, and that fired too.
The discipline worth preserving is what the briefs did not do. Every one of the six was capped at 0.54–0.57 by the calibration prior rather than talked up on conviction, and each carried an explicit counterweight — the copper row named XLB’s green premarket, the real-estate row named the FOMC, the energy row named the fact that energy owns the best year on the board and that a 4.5% commodity move invites mean reversion. Tonight vindicated the reads. It did not vindicate the prices, and section 02 argues the prices were the weaker half.
Set against that, one genuine caution: a 6–0 session is far less information than it appears. Five of the six were expressions of one macro driver, so this is closer to one thesis confirmed five times than six independent confirmations — the dedup logic that governs calibration suppresses same-thesis duplicates for exactly this reason, and these cleared it only because they were argued as separate families. And the copper row was right for materially the wrong reason. The single largest open question is unchanged from this morning: the FOMC concludes tomorrow, and it is the named kill leg on two of tonight’s six fires that today could not test.