Eight open predictions were carried into today's close and all eight could be scored against settled regular-session data — no gaps, nothing deferred. Five resolved in the predicted direction; three were killed by conditions written down in advance. The backlog is now empty.
| Setup | Outcome | Evidence | Δ-ATR | Actual |
|---|---|---|---|---|
| AF L USOmm-260729-AF-USO-L | FIRE | Crude closed up 7.32% at 129.31, nowhere near the 120.49 kill. Energy confirmed down at the single-name level (XOM +2.42%, CVX +2.28%, XLE +1.88%). Caveat honoured from build: the lean was already about +7% gapped premarket, so direction was right but timeliness was never tested. | — | 129.31 |
| AF S XLKmm-260729-AF-XLK-S | FIRE | Technology closed down 2.64% at 166.57 and never reclaimed 171.09. All three kills stayed dormant: semis did not turn green (SMH −4.80%) and no mega-cap bid led a green tape (SPY −1.54%). | — | 166.57 |
| AF S XLImm-260729-AF-XLI-S | FIRE | Industrials closed down 3.19% at 176.66. The stated mechanism strengthened rather than reverted: CAT closed −6.91% at 782.71, deeper than the −4.74% premarket print the lean was built on, extending the AI-capex de-rate into the industrial layer. | — | 176.66 |
| AF L XLVmm-260729-AF-XLV-L | VOID | The pre-registered kill fired: health care closed at 166.24, below the 167.26 line, down 0.61% with MRK −1.11%. XLV did beat the tape, but the kill was written as an absolute close, not as relative strength — scored VOID rather than rescued after the fact. | — | 166.24 |
| AF S XLYmm-260729-AF-XLY-S | FIRE | Consumer cyclicals closed down 0.77% at 111.61 and never reclaimed 112.48. The crude-reversal kill was the one to watch and it went the other way: USO +7.32% kept the pump-price headwind firmly in place. | — | 111.61 |
| AF S GLDmm-260729-AF-GLD-S | VOID | Two kills fired at once. Gold closed at 371.08, above the 369.37 line and up 0.46%, after trading as high as 377.62 — the safety bid the thesis said would keep failing to show up finally showed up. | — | 371.08 |
| AF S CPERmm-260729-AF-CPER-S | VOID | Copper closed 38.35 against a 38.33 kill — a two-cent, 0.05% margin, essentially flat on the day. Worth recording that the risk flagged at build did NOT fire: XLB closed −1.15%, so the lowest-conviction lean of the set broke on the leg nobody was watching. | — | 38.35 |
| AF L IGVmd-260729-1218-AF-IGV-L | FIRE | The midday software-versus-semis call landed and widened: IGV closed +0.64% at 92.37, above its 92.30 trigger, while semis were sold hard (SOXX −5.39%). The spread came in at 6.03 points against 5.54 at build. Caveat: the session low of 91.795 sits below the exec stop and a 12:18 emission cannot prove that low came before the trigger. | 0.03 | 92.37 |
All closes confirmed (Massive, regular-session close). Every referenced prior close matched the snapshot exactly — no stale references. Δ-ATR is shown only where a setup carried a single numeric level; the seven sector and commodity leans are direction calls without one.
Today's eight predictions hit 5 of 8, which nudged the hit rate up — and yet the Brier score (which grades how well-sized the confidence was, not just whether the call was right) got slightly worse. That combination is the finding of the day and it is explained below.
On the rolling last-50 window — the more responsive read — Brier moved from 0.2281 yesterday to 0.2322 today (+0.0041, worse) while the hit rate climbed from 64% to 66%. Across the week it is 0.2296 on 07-24 to 0.2322 now (+0.0026), with the hit rate up four points. Today's own eight-setup cohort scored 0.2365.
The reason is confidence bunching. All eight of today's calls were stated in a narrow 0.54–0.57 band (mean 0.551), so the model expressed almost no preference between its strongest thesis and its weakest. When three of those near-identical calls fail, the penalty on the misses outweighs the credit on the hits — being right more often does not help if every call is priced the same. The fix is spread, not accuracy.
Which setup families actually carry an edge, across the whole validated record. Calibration-eligible setups only. Diagnostic, never a trade plan.
| Pattern | Dir | n | Hit rate | Mean prob | Mean Brier | Evidence class |
|---|---|---|---|---|---|---|
| AF | L | 37 | 76% | 0.55 | 0.227 | lore_pending |
| AF | S | 34 | 65% | 0.55 | 0.232 | lore_pending |
| LRT | S | 33 | 42% | 0.39 | 0.275 | lore_pending |
| MS | L | 31 | 45% | 0.52 | 0.249 | lore_pending |
| SRB | L | 8 | 62% | 0.43 | 0.299 | lore_pending |
| GFD | L | 6 | 50% | 0.41 | 0.259 | practitioner_backtest |
| SBD | L | 6 | 0% | 0.37 | 0.138 | primary_source |
| SRT | S | 6 | 67% | 0.47 | 0.317 | lore_pending |
| MS | S | 5 | 80% | 0.46 | 0.295 | lore_pending |
| LRB | L | 4 | 50% | 0.46 | 0.215 | lore_pending |
| SBD | S | 3 | 0% | 0.28 | 0.084 | primary_source |
Six further single-observation slices (CGR-L, FBR-S, GFU-S, VAB-L, VSR-L at n=1) are omitted as directionally meaningless; GFU-S is classed retired_refuted and VSR-L retired_replaced_by_VBR.
Rolling last-50 Brier by validation date. The 07-16 to 07-21 stretch (0.221–0.224) remains the best of the record.
| Instrument | Dir | Target | Entry ref | Close | Realized | Result |
|---|---|---|---|---|---|---|
| ES (S&P Sep) SPX · resolve 07-03 | down | −1.5% | 7385.25 | 7537.43 | +2.06% | MISS |
| NQ (Nasdaq Sep) NDX · resolve 07-03 | down | −2.5% | 29346.75 | 29697.87 | +1.20% | MISS |
| NQ (Nasdaq Sep) NDX · resolve 07-03 | short | −3.0% | 29200.00 | 29697.87 | +1.71% | MISS |
| SPX (S&P 500) SPX · resolve 07-03 | up | +0.5% | 7369.00 | 7537.43 | +2.29% | HIT |
| BTC (Bitcoin) BTC-USD · resolve 07-04 | up | +4.0% | 60365.00 | 63086.45 | +4.51% | HIT |
| GC (Gold, spot) GOLD · resolve 07-10 | up | +2.0% | 4046.00 | 4104.10 | +1.44% | MISS |
Unresolved (score N/A, never guessed): XLK/XLU rotation pair and IWM-vs-QQQ pair (no price series — no series tag); BZ Brent Aug and USD/JPY, both resolve 07-10 (no series); ZW Wheat CBOT (no symbol, no series). Five of eleven cannot be scored from the price table.
est. (thinktank-v2 derived, asOf 2026-07-29) · read-only · DB not mutated.
Stocks were sold across the board, and the bigger the company the harder it was hit. The Dow proxy lost 2.18% and the Nasdaq 100 proxy 2.04%, while the equal-weighted version of the S&P — which treats every member the same size — gave up only 0.90%. Long-dated Treasuries fell 1.65%, meaning long yields rose on the day, so bonds offered no shelter either.
| Proxy | Close | Session % | Read |
|---|---|---|---|
| SPY (S&P 500) | 729.46 | −1.54% | Broad tape lower |
| QQQ (Nasdaq 100) | 661.73 | −2.04% | Growth led the decline |
| DIA (Dow 30) | 515.41 | −2.18% | Worst of the four — CAT weight |
| IWM (Russell 2000) | 288.57 | −1.64% | Small caps in line, not leading down |
| RSP (equal-weight S&P) | 215.73 | −0.90% | Beat cap-weight by 0.64pp |
| TLT (20y+ Treasuries) | 82.85 | −1.65% | Long yields up — no bond shelter |
Realized regime (how today actually resolved, not a forward call): risk-off but rotational rather than a broad liquidation. Money left the largest index constituents and the AI-buildout complex; it did not leave the market indiscriminately, and it found energy, staples and enterprise software on the way out.
A Federal Open Market Committee decision was scheduled for 2:00 PM ET with a press conference at 2:30 and no dot plot. The policy outcome was not independently confirmed from a primary source in this run and is therefore not asserted here; the observable facts are the bond and equity closes above. VIX is omitted — the same-day index level is not entitled on this feed and a prior-day close would be misleading.
The cross-asset picture argues against reading today as a fear event. A genuine flight to safety lifts the dollar and Treasuries together; instead the dollar eased 0.56%, long Treasuries fell, crude exploded 7.32% higher, and high-yield credit barely moved. That is the signature of a commodity supply shock working through the tape, not investors running for cover.
| Asset | Close | Session % | Confirms or diverges |
|---|---|---|---|
| USO (WTI crude) | 129.31 | +7.32% | The day's dominant move — supply premium |
| GLD (gold) | 371.08 | +0.46% | Diverged from the prior two sessions; took a bid intraday to 377.62 |
| UUP (US dollar) | 28.42 | −0.56% | Softer — argues against a safety bid |
| TLT (long Treasuries) | 82.85 | −1.65% | Diverged from equities — no duration hedge today |
| HYG (high-yield credit) | 79.24 | −0.23% | Barely moved — credit did NOT confirm equity stress |
Two sectors of eleven finished green — energy on the crude move and consumer staples on the defensive bid. Industrials were the worst performer at −3.19%, worse than technology, which is the notable ranking of the day: the pain is no longer confined to the chip and software complex. Real estate and communication services were nearly unchanged, so the red column is shallower than the headline index losses suggest.
On breadth, the equal-weight versus cap-weight spread is the usable measure today: RSP −0.90% against SPY −1.54% means the average S&P member held up markedly better than the index. The dedicated breadth indices (S5FI, S5TH, ADRN, TICK, TRIN) were not pulled this run — no open setup was gated on a breadth condition, so the mandatory breadth fetch did not apply and none of the scoring above depends on it.
The internals were far more violent than the index prints, and they split cleanly along one line: companies that spend on artificial-intelligence infrastructure were sold, and companies that use it were bought. Enterprise software had an outright strong day inside a 2% index decline.
| Bought — software and energy | % | Sold — AI buildout and cyclicals | % |
|---|---|---|---|
| INTU (Intuit) | +6.44% | CAT (Caterpillar) | −6.91% |
| ADBE (Adobe) | +5.72% | SOXX (semis ETF) | −5.39% |
| WDAY (Workday) | +5.21% | SMH (semis ETF) | −4.80% |
| ACN (Accenture) | +5.17% | MRK (Merck) | −1.11% |
| NOW (ServiceNow) | +4.65% | GDX (gold miners) | −0.86% |
| SNOW (Snowflake) | +4.64% | Miners diverged from the metal, which closed green. | |
| CRM (Salesforce) | +3.79% | Illustrative of market state only — not trade recommendations. | |
| XOM / CVX (majors) | +2.42% / +2.28% | ||
All figures confirmed (Massive, regular-session close) from the same batched snapshot used for scoring. The closing-bell newsletter layer (Stocktwits Daily Rip close, Axios Closer) was not pulled this run, so the narrative colour behind these moves is absent; the numbers themselves are primary-source and stronger than a newsletter restatement would be. Microsoft and Meta report after today's close, which is where the buildout-versus-consumer question gets its next data point.
The morning lens held up well on the things it could see and failed on the one thing it had over-extrapolated. It correctly called energy strength off the overnight supply headlines, correctly called technology and industrial weakness, and correctly identified the AI-capex de-rate as the session's organising mechanism — CAT went on to close 6.91% lower, deeper than the premarket print the call was built on. The midday addition, that capital was rotating within technology rather than leaving it, was the sharpest read of the day and widened in its favour.
Where it broke, it broke on the same logical shape twice. Both the gold short and the copper short rested on an asset continuing to refuse a bid it had refused for two sessions. Gold took the bid today and closed above its kill; copper closed two cents above its kill on a day its own flagged risk condition never materialised. A thesis whose evidence is a two-session absence is a thesis with a two-sample base, and today collected the third sample. The health-care long is a separate lesson: it was right about relative strength and wrong about the absolute close it had itself nominated as the test.
One of today's eight setups carried a written entry, stop and target. This is the only place a brief shows those numbers, and it does so after the fact, on a setup that is already closed and scored.
| Setup | Profile | Entry | Stop | Target | Realized R |
|---|---|---|---|---|---|
| AF L md-260729-1218-AF-IGV-L | day | 92.30 | 92.08 | 92.63 | +0.32 |
Marked from the close, the software long returned +0.32R — about a third of the amount that was at risk. The honest caveat, already recorded against the setup: the session low of 91.795 sits below the 92.08 stop, and because the idea was published at 12:18 there is no way to establish from daily data whether that low came before or after the 92.30 trigger. Treat +0.32R as the optimistic bound on this trade, not a verified clean result.
Across the whole validated record, 52 setups now carry a realized R: mean +0.19R, win rate 59.6%, stop rate 13.5%. Realized execution on closed setups — not a forward trade plan.