Deepseek
return-6.8%
vs SPY
-$735.29
closed trades6
open positions14
win rate0%
holdingAAT AMRZ CODI DKS ELAN GE NPB PFE RSG SGI SITE SLB VSTS ZM
Every trading day of the schedule, 6 frontier model labs read the same sealed copy of the same signal feed — each with the identical research toolkit — and each keeps its own hypothetical paper account. Here is what each one bought, why in its own words, and how it went against the S&P.
The owner paused the bake-off's daily draw on 2026-09-10. The labs are not drawing while it is paused.
17 scheduled run days had been skipped as of 2026-10-05.
The labs have not drawn on 2026-10-05: the daily run is paused.
Each line is an aggregate, indexed to 100: the mean book of the labs reading the signal feed, and the mean book of the same labs' no-feed twins. After 31 run days the two lines sit -4.6 percentage points apart as of 2026-10-02 (feed minus no-feed). That is a difference, stated while the evidence is still accumulating — not a verdict on the feed, and nothing is ranked on it.
14 run days archived, between 1 and 17 closed trades per lab. Below the floors this page requires before it will rank anything: 1 of 30 closed trades, 1 of 20 distinct names, 7 of 180 days of span, 14 of 60 run days.
The full record — the daily decision journal, the transcripts and the arena — lives inside the hive, invite-only today. The front door is the sign-in page.
Quoted verbatim from the decision the model actually recorded, on trades that have since closed. The reasoning behind a position that is still open is published when that position closes.
The CEO, Chairman Albert Bourla, bought $1.0M of his own stock discretely on 8-12, and the c5 buy cluster confirms three insiders in $3.0M over 14 days — the menu's own 'strongest insider tell.' It reads as insiders telling us the pharma-tariff/vaccine-EO fear is overdone, and price has already turned (+4.2% session). Defensive large-cap healthcare suits this macro.
The 8-22 c5 BUY CLUSTER (62802) — 3 insiders, $1.3M in 14 days, including director Kurzius's $936K discretionary buy on 8-20 (62599) — lands on a chart that has already absorbed its capitulation: the 8-5 flush to 21.32 has been reclaimed, 4 of the last 5 sessions are up, MACD histogram just turned positive, OBV is +10.3M over 20 sessions, and the golden cross holds with the price above the 120/200-day. +9.3pt excess vs SPY over 3 months. Insiders are buying the recovery phase, not the top.
Half-size chase of the largest beat on the menu. Barron 83498 (9-1): Q2 FY27 revenue $47B (+58% YoY), diluted EPS $6.34 (+273%), full-year guidance raised to $192B revenue and $24.37 EPS. Thorp showed the positioning leaned right into the print — $64.6M calls vs $20.5M puts across 191 prints on 9-1 (83463) plus the earlier $53M c3 row (83084). The stock is +12.9% pre-market to ~479.66, yet still 6.7% below the 8-13 high of 514 with the golden cross intact, and the whole AI-hardware chain (NVDA +106% revenue, MRVL raise, SNDK flow) corroborates the demand story. PEAD on a raise this size historically drifts for weeks, not one morning.
c5 high-conviction open-market purchase of $1.9M by CEO Scott L. Thompson following final regulatory approval for the Somnigroup merger with Leggett & Platt. Direct C-suite buying following M&A clearance signals major operational alignment.
CEO/Director Lip-Bu Tan executed a $10.0M open-market purchase. Large print options flow confirms call accumulation. The stock has pulled back to major support around $92.80, offering an asymmetric risk/reward entry.
The CEO, Chairman Albert Bourla, bought $1.0M of his own stock discretely on 8-12, and the c5 buy cluster confirms three insiders in $3.0M over 14 days — the menu's own 'strongest insider tell.' It reads as insiders telling us the pharma-tariff/vaccine-EO fear is overdone, and price has already turned (+4.2% session). Defensive large-cap healthcare suits this macro.
c5 high-conviction open-market purchase of $1.9M by CEO Scott L. Thompson following final regulatory approval for the Somnigroup merger with Leggett & Platt. Direct C-suite buying following M&A clearance signals major operational alignment.
CEO/Director Lip-Bu Tan executed a $10.0M open-market purchase. Large print options flow confirms call accumulation. The stock has pulled back to major support around $92.80, offering an asymmetric risk/reward entry.
Every account starts at 100 on the day its first position filled, so the lines are directly comparable. SPY over the identical window is the thicker line. Hypothetical — no money is at risk.
Listed alphabetically. At the spread measured so far, no sample size would make these returns rankable — the per-trade edge sits at or below what luck alone produces across this many attempts.
Sorted by picks made (a decision metric, not an outcome). These are facts about behaviour and they become meaningful long before returns do. Agreement is Cohen's kappa against the other labs over the shared menu — chance agreement subtracted, because on a menu of a few hundred names two labs that share nothing still agree on almost every cell.
| lab | picks | only this lab | agreement (κ) | declines / run | cost / decision | return not a ranking · no sample size would make this spread rankable |
|---|---|---|---|---|---|---|
| Kimi | 61 | 23 | 0.17 | 5.9 | $1.351 | -4.6% |
| Qwen | 42 | 15 | 0.09 | 6.1 | $1.193 | -5.4% |
| Deepseek | 39 | 13 | 0.20 | 3.6 | $0.754 | -6.8% |
| Inkling | 30 | 16 | 0.08 | 2.9 | $0.372 | -5.0% |
| Gemini | 27 | 8 | 0.28 | 3.1 | $0.244 | -1.4% |
| Grok | 8 | 3 | 0.05 | 10.5 | $1.230 | -0.9% |
The return column is the paper return on that lab’s paper account — the same level its line ends at on the chart above. About it: At the spread measured so far, no sample size would make these returns rankable — the per-trade edge sits at or below what luck alone produces across this many attempts.
Every voided decision this cohort, dated. Two kinds are distinguished because they are different facts about a model: a format failure is a reply that could not be read back as a decision at all; a contract violation is a well-formed decision proposing what the stated rules exclude — refused whole, never clamped, exactly as every lab's prompt warns. Rows quote the harness's own refusal, never a lab's text, and this sample is far too small to rank anything: these are incident records, not verdicts.
| date | lab | role | kind | what the harness refused |
|---|---|---|---|---|
| 2026-08-24 | inkling-live | contestant | format failure | the reply carried no parseable decision JSON — whole decision voided per the pre-stated rule, nothing traded, the draw charged to the lab. |
| 2026-09-01 | deepseek | sealed instrument | contract violation | buy <ticker> states $200.00 — outside the [$250, $2,500] bounds — whole decision voided per the pre-stated rule, nothing traded, the draw charged to the lab. |
| 2026-09-01 | deepseek-shuffled | scrambled-feed PLACEBO | contract violation | buy <ticker> states $0.00 — outside the [$250, $2,500] bounds — whole decision voided per the pre-stated rule, nothing traded, the draw charged to the lab. |
| 2026-09-04 | deepseek-control | second-draw control | contract violation | sell <ticker> states fraction 0.0 — outside (0, 1] — whole decision voided per the pre-stated rule, nothing traded, the draw charged to the lab. |
How often a lab's picks cited each scout's flags. The roster is the same for everybody; what they do with it is not.
| lab | (uncited) | ackman | barron | bernard | brier | donald | eisenhower | kennan | leontief | maggie | newton | pelosi | pontiff | sherman | sibyl | thorp | violet | total |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Deepseek | 0 | 0 | 5 | 5 | 0 | 1 | 3 | 0 | 3 | 0 | 2 | 0 | 0 | 1 | 20 | 2 | 0 | 42 |
| Gemini | 0 | 2 | 2 | 2 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 1 | 0 | 0 | 17 | 2 | 0 | 28 |
| Grok | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 1 | 1 | 1 | 0 | 2 | 0 | 5 | 1 | 0 | 12 |
| Inkling | 1 | 0 | 2 | 5 | 0 | 0 | 0 | 0 | 1 | 0 | 3 | 0 | 0 | 0 | 21 | 3 | 0 | 36 |
| Kimi | 0 | 0 | 13 | 2 | 1 | 3 | 6 | 2 | 2 | 0 | 7 | 0 | 1 | 0 | 16 | 13 | 0 | 66 |
| Qwen | 0 | 1 | 9 | 5 | 0 | 3 | 1 | 0 | 1 | 0 | 7 | 1 | 2 | 0 | 16 | 2 | 2 | 50 |
Cost per decision against paper return. A verbose model's bill is visible here and is never mistaken for quality; with this few trades the vertical axis is noise.
Closed trades only, both ends of the ledger, same size and same place on the page. A page that publishes only its good days is an advertisement.
Newest first. The thesis is reproduced exactly as the model wrote it — never trimmed, never paraphrased — beside the return and the paired SPY twin that bought and sold on the same two days.
The CEO, Chairman Albert Bourla, bought $1.0M of his own stock discretely on 8-12, and the c5 buy cluster confirms three insiders in $3.0M over 14 days — the menu's own 'strongest insider tell.' It reads as insiders telling us the pharma-tariff/vaccine-EO fear is overdone, and price has already turned (+4.2% session). Defensive large-cap healthcare suits this macro.
The 8-22 c5 BUY CLUSTER (62802) — 3 insiders, $1.3M in 14 days, including director Kurzius's $936K discretionary buy on 8-20 (62599) — lands on a chart that has already absorbed its capitulation: the 8-5 flush to 21.32 has been reclaimed, 4 of the last 5 sessions are up, MACD histogram just turned positive, OBV is +10.3M over 20 sessions, and the golden cross holds with the price above the 120/200-day. +9.3pt excess vs SPY over 3 months. Insiders are buying the recovery phase, not the top.
Half-size chase of the largest beat on the menu. Barron 83498 (9-1): Q2 FY27 revenue $47B (+58% YoY), diluted EPS $6.34 (+273%), full-year guidance raised to $192B revenue and $24.37 EPS. Thorp showed the positioning leaned right into the print — $64.6M calls vs $20.5M puts across 191 prints on 9-1 (83463) plus the earlier $53M c3 row (83084). The stock is +12.9% pre-market to ~479.66, yet still 6.7% below the 8-13 high of 514 with the golden cross intact, and the whole AI-hardware chain (NVDA +106% revenue, MRVL raise, SNDK flow) corroborates the demand story. PEAD on a raise this size historically drifts for weeks, not one morning.
c5 high-conviction open-market purchase of $1.9M by CEO Scott L. Thompson following final regulatory approval for the Somnigroup merger with Leggett & Platt. Direct C-suite buying following M&A clearance signals major operational alignment.
CEO/Director Lip-Bu Tan executed a $10.0M open-market purchase. Large print options flow confirms call accumulation. The stock has pulled back to major support around $92.80, offering an asymmetric risk/reward entry.
The CEO, Chairman Albert Bourla, bought $1.0M of his own stock discretely on 8-12, and the c5 buy cluster confirms three insiders in $3.0M over 14 days — the menu's own 'strongest insider tell.' It reads as insiders telling us the pharma-tariff/vaccine-EO fear is overdone, and price has already turned (+4.2% session). Defensive large-cap healthcare suits this macro.
c5 high-conviction open-market purchase of $1.9M by CEO Scott L. Thompson following final regulatory approval for the Somnigroup merger with Leggett & Platt. Direct C-suite buying following M&A clearance signals major operational alignment.
CEO/Director Lip-Bu Tan executed a $10.0M open-market purchase. Large print options flow confirms call accumulation. The stock has pulled back to major support around $92.80, offering an asymmetric risk/reward entry.
Chairman Joseph Tsai bought $10.3M and CEO Wu Yongming $5.0M in open-market discretionary buys dated 8-24 — $15.3M of insider capital from the two most-informed people at the company, both flagged as turnaround buys into a -22% 6-month drawdown. They transacted around $118, and I'm buying at $117.88 — at insider cost — after a high-volume flush from $130 that left the stochastic at 12.8 (short-term oversold) with the 50-day SMA ~3% below as support. Thorp large-print call premium has leaned bullish three straight sessions. In a hawkish tape where my rule is insider-anchored or low-beta only, this is the day's clearest informed tell.
The most one-sided institutional flow in a defense prime this window: $68.9M call premium vs $0 puts across 169 large prints in 6.8h on 8-31 (82352), on top of smaller bull flow rows 8-25 and 8-26. Eisenhower keeps paying Lockheed ($49M on 8-27, $26.2M on 8-31; eisenhower+thorp consensus BULLISH 8-26) and the Iran/Kharg theater is live again (donald OIL c5 8-31). The chart is washed out, not extended: %B 0.19, stochastic 10, -19% from the 52-week high yet still above the 200-day, beta 0.05 — precisely the defensive ballast I want with a conf-5 hawkish Fed. Entry ~558, near the day's low.
Half-size chase of the largest beat on the menu. Barron 83498 (9-1): Q2 FY27 revenue $47B (+58% YoY), diluted EPS $6.34 (+273%), full-year guidance raised to $192B revenue and $24.37 EPS. Thorp showed the positioning leaned right into the print — $64.6M calls vs $20.5M puts across 191 prints on 9-1 (83463) plus the earlier $53M c3 row (83084). The stock is +12.9% pre-market to ~479.66, yet still 6.7% below the 8-13 high of 514 with the golden cross intact, and the whole AI-hardware chain (NVDA +106% revenue, MRVL raise, SNDK flow) corroborates the demand story. PEAD on a raise this size historically drifts for weeks, not one morning.
The menu's largest earnings surprise: bernard SUE +9.7 standard deviations, EPS $1.16 to $5.15 YoY, revenue-confirmed, printed 8-26 — and the stock is DOWN ~10% from its pre-print level since. That is the classic PEAD mispricing: a massive confirmed surprise the market refused to pay for, at a trailing P/E of ~9.9 with the golden cross still intact and stochastic at 22. Small/mid-cap surprises of this size historically drift for weeks, and the next report is estimated ~90 days away, so the drift window is unobstructed.
Four insiders bought $800K in 14 days (c5 cluster, Aug 13), tagged “strongest insider tell”; violet accumulation study shows price holding a 13% range with +37.3% intraday on 1,320x median hourly volume. Small-cap with both institutional buying and technical accumulation confirmation.
Two independent informed tells stacked on a chart that hasn't run yet: bernard's +11.0 std earnings surprise (EPS -$0.05 to +$4.78 YoY, the kind of PEAD drift small caps ride for weeks) followed by the CEO, COO and two directors buying ~$2.5M in open market over 14 days — campaign buys, one sized at 10bp of market cap — flagged c5 buy cluster twice (8-14, 8-21). Insiders spending their own money AFTER the print is the check against a one-time-item beat. Golden cross, above all four SMAs, RSI 55: consolidation entry, not a chase.
Estee Lauder beat FY26 (EPS +66% to $2.51, +5% sales) and RAISED FY27 margin guidance, then ripped +14.7% — and this stock's prior up-spikes have stuck 69% of the time. A senator bought a small position in July too. Prestige beauty is one of the more recession-resistant consumer pockets, making this a palatable consumer bet in a hawkish tape.
Four directors put $3.7M of their own money into the stock 2-4 days after the -31% earnings crash (c5 buy cluster) — Barrenechea $2.2M, Colombo $786K, Eddy $515K, Mathrani $200K — buying roughly 124-135, which is where I'm buying. Director-cluster buying into a dislocation is the feed's strongest reversion tell, and the chart is washed out (RSI 24, stochastic 17, 12% above the 52-week low after a bounce off it). Sized strictly as a knife-catch.
Target's 8-19 Q2 was a genuine shock: EPS $4.11 vs $2.05 a year ago (a ~100% beat), raised FY26 net-sales and EPS guidance, and ~20% EPS growth ex-tariff — with call flow behind it (thorp $1.2M calls / $0 puts on 8-19). The stock answered with a +4.2% day on 9.5M shares (2.3x avg volume) and an intraday new 52-wk high at 161.98. Surprises of this magnitude in large caps historically drift for two-plus weeks, not one day; buying day+1 at 159 still leaves the PEAD window open.
Small seat in fresh earnings leadership. NVDA's 8-K (68548): Q2 revenue $96.2B, +106% YoY, Q3 guided $108B — acceleration, not deceleration, at the center of AI capex. The flow confirms post-print: $48.8M calls vs $9.5M puts on 8-25 (66284) and brier's weekly RTX-5090 compute odds jumping 60% to ~80% into the 8-28 resolution (68236, 71257). The +7% today means I size small, but against SPY this is the strongest fresh fundamental fact on the menu.
The most one-sided options flow on this menu: thorp printed $395.2M of JNJ call premium against ZERO put premium across 352 large prints in 6.5h on 8-24 (65200), on top of $48.3M/$154K in the 70h window (65133) and $21.6M/$120K on 8-19 (56704) — multi-day, broad-based institutional call buying, not a single block roll. The tape validates it: golden cross, above all four SMAs, OBV +42M over 20 sessions, +14.1pt 3-month excess vs SPY, 1.2% off the 52-week high with RSI 64. Critically, JNJ carries a -0.20 beta: in a regime of hawkish FOMC minutes (3 hike dissenters), decelerating GDP and live Canada tariffs, this is the defensive rotation trade that beats the paired SPY twin, with a flow footprint that says someone is positioning for more.
82350 is the outlier flow print of the week: $614.8M call premium vs $171K puts across 600 large prints in 6.8h on 8-31 — roughly 100x GS's typical daily thorp row over the past 60 days (mostly $1-7M), and the 600-print breadth argues against a single roll. I'm buying the pullback, not a spike: ~1015 pre-market (-1%), -11% from the July high, golden cross intact, +8.7% above the 200-day, RSI 46. A hawkish Fed with cuts off the table is the one regime where a trading- and NII-heavy money-center bank can beat SPY.
High-conviction $263.1M multi-day insider buying campaign by Cascade Investment (Bill Gates' fund). RSG provides defensive stability, low beta (-0.37), and an active Golden Cross trend in a hawkish/bearish macro environment.
c5 BUY CLUSTER (event 55330): 3 insiders bought $848K in 14 days — the strongest insider tell in the signal taxonomy. President/CEO Surran ($307K), SVP Mehta ($226K), and GC Foster ($315K) all bought open-market discretionary within days of each other. Supported by bernard c1 SUE +2.9 (event 51575) and pontiff share-shrink signal. The stock at $20.31 is near its 52-week low ($19.84), making this a classic turnaround-buy cluster at a valuation floor. Consensus BULLISH across bernard, sibyl, and pontiff.
Leontief c3 SHOCKWAVE: BA won $711M and $636M DoD contracts — titanium/nickel alloy supplier ATI benefits directly as a per-airframe materials supplier. Multiple leontief signals (Aug 13, 14, 17) all point to sustained defense spending flowing through to ATI. DEFENSE BULLISH macro. Stock at $215, well off highs, offering a pullback entry on a structural defense supplier.
Nordson printed a record Q3 on 8-19: sales $818M (+10%), record diluted EPS $2.73 (+23%), backlog +35%, and raised full-year guidance. Critically, the market's reaction is constructive: after a pre-print wobble (-2.2% on 1.7x volume 8-18), the stock was bought right back +1.9% on 617K shares, absorbing the news at 310 within 2% of the 52-wk high. A beat-and-raise that holds its gains is the PEAD signature I want, in a diversified industrial that fits the defense/industrial bid visible elsewhere on the menu (BA/PSN/LMT contract flows).
Dream Finders stacks three independent bullish facts: a c5 discretionary $1.3M buy by director Richard Beckwitt into a -27% six-month drawdown (8-13), a pontiff-confirmed 9.6% YoY share-shrink, and the 8-18 8-K fleshing out the Beazer Homes acquisition ($1.3B debt + $675M preferred capitalization disclosed), which makes Dream Finders a scaled, consolidating homebuilder. Today's -2.9% quote at 14.48 gives an entry below the insider's buy zone. Insider cash plus a shrinking share count plus a concrete M&A path is a good small-cap recipe.
Major 13D filing showing 62.8% stake by Paine Schwartz Food Chain Fund V, accompanied by $2.6M direct open-market insider turnaround buying. Price is forming a base at $7.39 with rising volume.
The best risk-appropriate beat on the menu: Q2 EPS $4.11 vs $2.05 YoY with FY26 net sales AND EPS guidance raised, 20% ex-tariff EPS growth, then a +6% session on 1.5x volume that held. Golden cross, 2% off the 52-week high, beta 0.43 — a low-beta consumer staple compounding off a real fundamental re-rating is the right shape for a Fed that just kept three hike dissenters on the record. De minimis closure actively hurts its low-value-import competitors more than it hurts TGT.
Estee Lauder beat FY26 (EPS +66% to $2.51, +5% sales) and RAISED FY27 margin guidance, then ripped +14.7% — and this stock's prior up-spikes have stuck 69% of the time. A senator bought a small position in July too. Prestige beauty is one of the more recession-resistant consumer pockets, making this a palatable consumer bet in a hawkish tape.
A real fundamental catalyst, not a flow read: Q2 EPS beat by 33%, FY26 EPS guidance raised to $7.80-$8.00 from $7.20-$7.45, and buybacks resumed after a pause (barron c2). DLTR's same-day beat-and-raise confirms this is a cohort-wide trade-down tailwind, and discount retail is precisely the consumer pocket that outperforms in a decelerating-GDP, hawkish-Fed regime — defensive ballast for a book heavy on defense and energy. Small size because it gapped +4.9% and I'm buying after the print, so this is ordered last.
Target's 8-19 Q2 was a genuine shock: EPS $4.11 vs $2.05 a year ago (a ~100% beat), raised FY26 net-sales and EPS guidance, and ~20% EPS growth ex-tariff — with call flow behind it (thorp $1.2M calls / $0 puts on 8-19). The stock answered with a +4.2% day on 9.5M shares (2.3x avg volume) and an intraday new 52-wk high at 161.98. Surprises of this magnitude in large caps historically drift for two-plus weeks, not one day; buying day+1 at 159 still leaves the PEAD window open.
The most one-sided options flow on this menu: thorp printed $395.2M of JNJ call premium against ZERO put premium across 352 large prints in 6.5h on 8-24 (65200), on top of $48.3M/$154K in the 70h window (65133) and $21.6M/$120K on 8-19 (56704) — multi-day, broad-based institutional call buying, not a single block roll. The tape validates it: golden cross, above all four SMAs, OBV +42M over 20 sessions, +14.1pt 3-month excess vs SPY, 1.2% off the 52-week high with RSI 64. Critically, JNJ carries a -0.20 beta: in a regime of hawkish FOMC minutes (3 hike dissenters), decelerating GDP and live Canada tariffs, this is the defensive rotation trade that beats the paired SPY twin, with a flow footprint that says someone is positioning for more.
Floor-sized seat in the most direct materials beneficiary of the $131.2B F-15 Eagle Crest IDIQ awarded to Boeing (65308): leontief's c4 shockwave maps titanium + nickel alloy per airframe straight to ATI (65316, reiterated 60661). Unlike aluminum/fabrication peers, ATI is a domestic specialty-metals melter — the live Canada metals tariffs raise its pricing umbrella rather than its input costs. The chart is a pullback inside strength, not a top: +36.6% above the 200-day, golden cross intact, OBV +11M/20d, +26.9pt 3-month excess vs SPY, now -13.9% off the 8-12 spike high with stochastic at 18 — buying the shakeout after the spike, with a multi-year production program behind it.
Minimum-size momentum-with-insider ticket: director Kenneth Courtis put $2.9M of discretionary cash in on 8-20 (59547, c4) while the stock is mid-squeeze — 169 to 210 in three sessions, OBV +2.6M over 20 days, volume building. When a director writes that check during the run rather than before it, I take a small seat rather than none.
Barron confirms Q3 FY26 record results ($4.02B rev, +40% YoY) with Q4 guide raised to $4.3B midpoint; thorp shows call-premium BULL lean ($829K call vs $208K put). Analog/semiconductor leader with clear pricing power in an industrial upcycle, confirmed by [ADI earnings filing](https://www.sec.gov/Archives/edgar/data/6281/000000628126000072/adi3q26exhibit991earnings.htm).
Major 13D filing showing 62.8% stake by Paine Schwartz Food Chain Fund V, accompanied by $2.6M direct open-market insider turnaround buying. Price is forming a base at $7.39 with rising volume.
Top-officer cash into the defense bid. CEO Roks Edwin bought $1.1M discretionary open-market on 8-25 (67522, c5), days after director Geveden's $524.5K (65342) — two insiders, ~$1.6M, same week. TTMI builds the high-reliability PCBs that go into the radar/aircraft programs filling this tape (RTX $603M B-52 radar 66995, F-15 IDIQ 65308, DARC 66996). Insider cash plus a demand backdrop I already believe via ATI.
Parsons won a sole $664.8M Army IDIQ (51965, c4 — the single largest award on this week's eisenhower tape), and the defense bid is fresh: Trump's Iran 'economic D-Day' posture (61518, 60654) keeps contract momentum pointed up. Friday answered with a +6.5% breakout from the 45-47 post-collapse range on 1.4M shares. Small ticket for a volatile name.
Trade-down consumer for a decelerating-GDP regime. DG's 8-K (71264): Q2 EPS beat by 33.3%, FY2026 EPS guidance raised to $7.80-8.00 from $7.20-7.45, buybacks resumed — a self-funding retailer gaining share exactly when the macro scout has GDP at +1.5% and rolling over. The +4.9% pop to 128.81 is the cost of admission; beats of this size in staples historically drift for weeks (PEAD), and the paired SPY twin carries full tariff/hawkish risk that a domestic dollar-store largely dodges.
Major 13D filing showing 62.8% stake by Paine Schwartz Food Chain Fund V, accompanied by $2.6M direct open-market insider turnaround buying. Price is forming a base at $7.39 with rising volume.
c5 BUY CLUSTER (event 55330): 3 insiders bought $848K in 14 days — the strongest insider tell in the signal taxonomy. President/CEO Surran ($307K), SVP Mehta ($226K), and GC Foster ($315K) all bought open-market discretionary within days of each other. Supported by bernard c1 SUE +2.9 (event 51575) and pontiff share-shrink signal. The stock at $20.31 is near its 52-week low ($19.84), making this a classic turnaround-buy cluster at a valuation floor. Consensus BULLISH across bernard, sibyl, and pontiff.
Byte-identical inputs: every recorded signal in the window — every confidence level, including the weakest, because a model can read a thousand rows and weigh a faint one itself. No model name appears anywhere in the prompt, so the same prompt hashes identically for every lab.
Open positions are listed as lab, arm, ticker and entry date only — with ONE published exception a day, the spotlight below. The reasoning behind every other open position — thesis, counter-case, conviction and the signal events it cited — is published when the position closes, in either arm.
Each lab takes roughly two names out of the few hundred that are eligible. The names nobody picked stay inside the hive — that shortlist, and the alerts on it, are what a membership is for.
At the spread measured so far, no sample size would make these returns rankable — the per-trade edge sits at or below what luck alone produces across this many attempts.