Performance Scoreboard

How MCH ratings have performed — measured as alpha (return minus benchmark), not raw price moves. SPY + GICS sector SPDR (alpha = stock return − benchmark, same window).

1 · Pre-registered out-of-sample anchor

The honest track record. The REG-004 release is frozen as the benchmark: 480 names pre-registered on 2026-07-08.

Status

scored · 0 scored yet — the forward record begins accruing from 2026-07-08 as monthly prices are appended. Check back after 2026-07.

Sector-alpha (mean)

—

once ≥21d horizon is reached

Rating	N	Avg realized	Alpha vs SPY	Alpha vs sector	95% CI (sector α)	Hit-rate
No forecasts at the ≥21-day horizon yet.

2 · Legacy calibration set

Genuinely out-of-sample but thin: 22 names, window 2026-04-24→2026-07-08, median horizon 75 days. Mean sector-alpha +7.7% (95% CI [-2.8%, +21.1%]).

Rating	N	Avg realized	Alpha vs SPY	Alpha vs sector	95% CI (sector α)	Hit-rate
BUY	17	+14.7%	+9.9%	+9.3%	[-4.4, +26.8]	43.8%
HOLD	3	+15.2%	+10.5%	+8.6%	[-1.9, +18.8]	66.7%
SPECULATIVE BUY	2	-2.0%	-6.8%	-6.0%	—	50.0%

Method. One forecast per name (earliest dated snapshot → latest observed price; ≥21-day horizon), so no name is double-counted across overlapping snapshots — N is distinct names. Alpha = the name's realized return minus its benchmark's return over the identical window (raw close vs raw close). Benchmark hit: a BUY beats its sector, a SELL trails it, a HOLD tracks it (±10%). The 95% CI is a 2,000-sample bootstrap over names — with a small, correlated, single-window sample it is wide by construction and honest about it. Retired/delisted names are kept at their last observed price (no survivorship scrub), except that a move below -85% is treated as a corrupted price print and excluded — consistent with the publish-time price anomaly gate. 1 name(s) excluded on a price-data anomaly (NPN). A genuine track record requires forward out-of-sample time across multiple regimes; that is what section 1 accrues. Generated 2026-07-08 14:56.