Track record · first-party panel 2026-05-15 → 2026-08-31

What happened after we flagged them.

Since 2026-05-15 we have recorded every nightly scan of the full Broken List into an append-only panel — 69 scans, no backfill, no edits. That makes the list testable: once a stock is flagged, does it actually stay broken? This page is the observed answer over the first 108 days, published as fact, not forecast.

Market context: over the same window the S&P 500 (SPY) went from 739.17 (2026-05-15) to 767.05 (2026-08-31), +3.8%. A rising market, and three in four flagged names still hadn't recovered — persistence here isn't a market artifact.

The severity gradient holds

A warning taxonomy is only honest if the tiers mean something. They do: over the observed window, recovery ran Watch → Amber → Red exactly as the tiering intends. The deeper the tier at first capture, the less likely the name had left the list.

Seed cohort: every ticker on the list at the first capture (2026-05-15), measured through 2026-08-31. “Sustained Red stint” = a name that entered below Red and later held the Red List for 3+ consecutive scans.

Names flagged while we watched

The seed names were already broken when tracking began. The cleaner test is the 957 names newly flagged since, each observed for at least 60 days. Nearly all enter at the Watch tier — the early-warning rung — and 51% durably recovered, most within about two months (median 62 observed days on the list before exit). But 62% went on to hold a worse tier for 3+ consecutive scans, and 31% went on to a sustained Red List stint. The two groups overlap — a name can slide to Amber, hold there, and still recover later — so the figures don't sum to 100. When a Watch flag escalates, the flag was early — that is the job.

New names almost never enter at Red or Amber directly — a stock crosses the −20% Watch floor first — so those columns are tiny cells; read All and Watch.

How these numbers are computed

Every night the full universe — ticker, tier, decline depth, decline sigma, time-frame continuity, price — is appended to a first-party panel that is never edited or backfilled. “Still broken” means present in the latest scan. “Durably recovered” means absent from the last 5 scans (about a week) — names hovering at the −20% line flicker on and off the list day to day, and a durable exit is the honest bar. Caveats: a delisted stock leaves the panel the same way a recovered one does, and delistings are the worst outcomes — so true recovery rates are, if anything, lower than shown. This covers one observation window in one market regime, and it is descriptive history, not a forecast. The full report and definitions ship in the repository as data/track_record.md; figures regenerate with the nightly refresh.