My journal / log your own trades — Vig grades every one
against price AND against the S&P over the same window, then coaches
OPEN — marked to last close
Trade
Entered
Size
Entry
Last
P&L
vs S&P
Model @ entry
Model now
CLOSED — the record
Trade
Held
Size
Entry→Exit
P&L $
P&L %
vs S&P
Grade
Model @ entry
The coach
— rule-based, recomputed on every trade; sharpens as your sample grows
copy record for Claudepaste into claude.ai for a
deep coaching read — free with your subscription
Vig’s record / weekly top-3, graded after each 5-day horizon
— the machine is held to the same standard as you
Relative hit rate
49% of 140 calls
beat the S&P median over 5d (the model's actual claim)
Avg edge per call
+0 bps / 5d, vs median
absolute P&L basis: +16 bps, hit 53%
Coverage
47 weeks
longest losing streak: 4 weeks
Cumulative edge / mean top-3 return vs S&P median,
summed arithmetically week by week — not compounded, not a P&L
final: -0.1 percentage points of
cumulative relative edge over 47 weeks (gross, before costs —
a yardstick of pick quality)
Where the edge lives
Side
N
Hit
Avg edge
LONG
94
50%
+30 bps
SHORT
46
48%
-61 bps
Conviction
N
Hit
Avg edge
low
47
51%
-4 bps
mid
46
46%
+43 bps
high
47
51%
-38 bps
Self-diagnosis / computed from the record, regenerated daily
Relative hit rate 49% over 140 calls is not distinguishable from 50/50 (t≈-0.2) — judge the process, not short-run outcomes.
Context: over the full 166-week record the same rule averages +12 bps/call (hit 51%); this window runs -12 bps vs that baseline — consistent with the longer record. Four alternative selection rules were tested on the full record (2026-07, see RESEARCH_NOTES): none separable from noise, none adopted.
High-conviction calls do NOT hit more often than low-conviction ones in this window — treat the probability as a ranking device, size picks equally.
Longest weekly losing streak: 4 weeks. An edge this size guarantees streaks like this — the failure mode to avoid is abandoning the process inside one.