jev shadow — TypeSafe Jev vs LLM direction call
paper only · realised YES-price move is the judge · refreshed 0s agoexperiment
day 14 of 14
pairs scored
84,020 pairs
priced (24h+ old)
84,009 marked
agree with LLM
83.5%
jev p50 latency · cost
237 ms · $0.94
realised move — hit-rate and $10-ticket paper PnL
A call is a hit when the YES price moved ≥ 1¢ in the called direction within the horizon, a miss when it moved against it; flat rows are shown but excluded from the rate. UNCLEAR calls never trade.
| LLM (Opus/Sonnet) | 1h | 6h | 24h |
|---|---|---|---|
| all | 49% (1366/2781 · 10455 flat) -6819.03$ mid +1712.35$ | 51% (2708/5329 · 7564 flat) -6172.68$ mid +3523.62$ | 50% (3555/7169 · 5374 flat) -7576.38$ mid +856.33$ |
| conf>=60 | 48% (483/997 · 3429 flat) -1635.76$ mid +1175.82$ | 50% (956/1905 · 2292 flat) -1743.36$ mid +1156.93$ | 52% (1269/2429 · 1617 flat) -2198.87$ mid -269.63$ |
| conf>=70 | 51% (228/443 · 1819 flat) -615.64$ mid -3.85$ | 50% (434/868 · 1195 flat) -806.17$ mid +208.06$ | 53% (577/1087 · 898 flat) -986.90$ mid -390.32$ |
| conf>=80 | 62% (84/135 · 819 flat) -130.41$ mid +18.48$ | 55% (162/294 · 487 flat) -174.93$ mid -49.02$ | 55% (204/369 · 382 flat) -289.90$ mid -228.05$ |
| Jev (TypeSafe) | 1h | 6h | 24h |
|---|---|---|---|
| all | 51% (1704/3339 · 13480 flat) -6195.77$ mid +1696.38$ | 53% (3356/6354 · 10223 flat) -5344.20$ mid +3069.54$ | 54% (4568/8517 · 7688 flat) -5898.28$ mid +1412.13$ |
| conf>=0.5 | 52% (1066/2061 · 7128 flat) -3054.61$ mid +548.42$ | 53% (2024/3805 · 5248 flat) -2831.50$ mid +1034.29$ | 55% (2717/4930 · 3932 flat) -3574.23$ mid -145.79$ |
| conf>=0.7 | 50% (609/1223 · 4234 flat) -1839.12$ mid +151.59$ | 54% (1272/2338 · 3036 flat) -1735.07$ mid +173.36$ | 57% (1692/2994 · 2263 flat) -2079.25$ mid -239.78$ |
| conf>=0.85 | 49% (267/546 · 2167 flat) -865.83$ mid -51.52$ | 54% (601/1122 · 1542 flat) -875.80$ mid -82.48$ | 58% (830/1433 · 1162 flat) -941.09$ mid -258.45$ |
where the LLM abstained
Pairs the LLM returned as UNCLEAR (incl. the “not mentioned” filler rows) but Jev called UP or DOWN. This is the signal Jev would add on top of the current setup.
| Jev only, LLM UNCLEAR | 1h | 6h | 24h |
|---|---|---|---|
| all | 52% (714/1380 · 6839 flat) -2602.86$ mid +378.74$ | 53% (1406/2673 · 5467 flat) -2293.54$ mid +1377.91$ | 55% (2017/3698 · 4270 flat) -1888.33$ mid +1516.58$ |
| conf>=0.5 | 54% (391/718 · 2873 flat) -1003.33$ mid +196.24$ | 53% (714/1341 · 2210 flat) -948.09$ mid +205.21$ | 57% (1000/1748 · 1736 flat) -1029.18$ mid +199.00$ |
| conf>=0.7 | 53% (180/337 · 1345 flat) -471.04$ mid +90.71$ | 54% (372/686 · 978 flat) -427.61$ mid +94.55$ | 57% (499/870 · 752 flat) -558.47$ mid +7.01$ |
| conf>=0.85 | 56% (74/133 · 534 flat) -182.09$ mid +27.35$ | 55% (161/293 · 361 flat) -183.33$ mid +0.39$ | 56% (200/356 · 273 flat) -237.21$ mid -49.28$ |
agreement — LLM (rows) × Jev (columns)
| UP | DOWN | UNCLEAR | |
|---|---|---|---|
| UP | 3,146 | 671 | 3,037 |
| DOWN | 238 | 4,584 | 1,631 |
| UNCLEAR | 2,205 | 6,053 | 62,455 |
83.5% agreement over 84,020 pairs · 909 outright flips (UP↔DOWN)
cumulative net paper PnL — 24h horizon
LLM: every UP/DOWN call · Jev: calls with confidence ≥ 0.7. Same $10 ticket, same horizon, same rows. Net: taker pays a 2¢ round-trip spread; calls priced outside 0.05–0.95 are left out.
LLM
Jev
latest disagreements
Pairs where both models committed and disagreed. The 24h move says who was right.
| when | headline | market | LLM | Jev | YES t0 → 24h | right |
|---|---|---|---|---|---|---|
| 03 Oct 23:22 UTC | US Military Buildup Raises Iran Escalation Risk (bloomberg_markets) | Will Trump speak to Reza Pahlavi by October 31? | UP 48 | DOWN 0.19 | 0.04 → 0.02 (-0.02) | Jev |
| 03 Oct 21:42 UTC | US Military Buildup Raises Iran Escalation Risk (bloomberg_markets) | Will the next US-Iran senior diplomatic meeting take place in Switzerland by December 31, 2026? | UP 45 | DOWN 0.83 | 0.04 → 0.04 (+0.00) | — |
| 03 Oct 17:22 UTC | G7 to release 100 million barrels of oil and diesel after Trump export ban threat (bbc_biz) | Will EU emergency diesel stocks be at least 34M tonnes at end of November 2026? | DOWN 65 | UP 0.50 | 0.34 → 0.28 (-0.06) | LLM |
| 03 Oct 17:03 UTC | Pressure Builds on Iran as Trump Expands US Military Presence (bloomberg_politics) | Will Mohammed bin Abdulrahman Al Thani attend a US x Iran diplomatic meeting by December 31, 2026? | UP 45 | DOWN 0.57 | 0.33 → 0.33 (+0.00) | — |
| 03 Oct 17:03 UTC | Justice Department Urges Judge to Let Trump’s Media Ban Proceed (bloomberg_politics) | Will Donald Trump publicly insult someone on October 23, 2026? | UP 62 | DOWN 0.44 | 0.90 → 0.90 (+0.00) | — |
| 03 Oct 17:03 UTC | G7 to release 100 million barrels of oil and diesel after Trump export ban threat (bbc_biz) | Will EU emergency diesel stocks be at least 35M tonnes at end of November 2026? | DOWN 68 | UP 0.49 | 0.28 → 0.17 (-0.11) | LLM |
| 03 Oct 17:03 UTC | G7 to release 100 million barrels of oil and diesel after Trump export ban threat (bbc_biz) | Will EU emergency diesel stocks be at least 33M tonnes at end of November 2026? | DOWN 60 | UP 0.61 | 0.39 → 0.45 (+0.07) | Jev |
| 03 Oct 17:03 UTC | G7 to release 100 million barrels of oil and diesel after Trump export ban threat (bbc_biz) | Will EU emergency diesel stocks be at least 37M tonnes at end of November 2026? | DOWN 72 | UP 0.27 | 0.06 → 0.05 (-0.01) | — |
| 03 Oct 16:42 UTC | US Military Buildup Raises Iran Escalation Risk (bloomberg_markets) | Will USD be between 2.3M and 2.4M Iranian rials on October 31? | UP 52 | DOWN 0.16 | 0.03 → 0.04 (+0.01) | LLM |
| 03 Oct 14:43 UTC | Houthis blame Saudi air strikes as loud blasts rock Yemen's capital Sanaa (france24) | Saudi troops enter Sanaa by October 31, 2026? | UP 50 | DOWN 0.32 | 0.07 → 0.04 (-0.04) | Jev |
| 03 Oct 14:43 UTC | New aircraft carrier, 10,000 US troops: Is the Iran war about to escalate? (aljazeera) | Will fewer than 20 ships transit the Strait of Hormuz between October 5-October 11? | UP 55 | DOWN 0.23 | 0.09 → 0.21 (+0.12) | LLM |
| 03 Oct 13:41 UTC | G7 to release up to 100m barrels of emergency oil and diesel reserves (guardian_world) | Will US crude oil reserves fall to 260M by October 30, 2026? | UP 72 | DOWN 0.39 | 0.10 → 0.04 (-0.06) | Jev |
| 03 Oct 10:42 UTC | Manchester City’s guilty verdict brings football’s great sell-off crashing to earth (guardian_world) | Will Manchester City be expelled from the Premier League as a sanction? | UP 50 | DOWN 0.52 | 0.23 → 0.21 (-0.02) | Jev |
| 03 Oct 09:41 UTC | Democrats May Have Found the Recipe for Flipping Red-State Senate Seats (nyt_politics) | Republican Senate odds hit 55% by October 31? | UP 55 | DOWN 0.68 | 0.01 → 0.01 (+0.00) | — |
| 03 Oct 09:41 UTC | Democrats May Have Found the Recipe for Flipping Red-State Senate Seats (nyt_politics) | Republicans flip Democrats for Senate by November 2? | UP 50 | DOWN 0.82 | 0.12 → 0.12 (+0.00) | — |
| 03 Oct 09:41 UTC | Democrats May Have Found the Recipe for Flipping Red-State Senate Seats (nyt_politics) | Republican Senate odds hit 50% by October 31? | UP 50 | DOWN 0.45 | 0.12 → 0.12 (+0.00) | — |
| 03 Oct 09:41 UTC | Democrats May Have Found the Recipe for Flipping Red-State Senate Seats (nyt_politics) | Republican Senate odds hit 40% by October 31? | UP 45 | DOWN 0.38 | 0.35 → 0.35 (+0.00) | — |
| 03 Oct 09:41 UTC | India Renames 28 Ladakh Sites, Risking Fresh China Friction (bloomberg_politics) | Will the next Russia x Ukraine peace talks take place in India by December 31, 2027? | UP 30 | DOWN 0.35 | 0.01 → 0.01 (+0.00) | — |
| 03 Oct 09:03 UTC | India beat arch-rivals Pakistan to win Asian Games T20 cricket gold (aljazeera) | Asian Games Men: Pakistan vs India | DOWN 90 | UP 0.13 | pending | — |
| 03 Oct 07:23 UTC | Pressure Builds on Iran as Trump Expands US Military Presence (bloomberg_politics) | Hormuz Island no longer under Iranian control by October 31? | UP 45 | DOWN 0.37 | 0.02 → 0.01 (-0.00) | — |