Post-mortem: the $17.00 that wasn't

For nine weeks this spring, our row for Great Lakes Dredge & Dock (GLDD) showed the same price: $17.00. Exactly $17.00, every day, April through early June - while the scores beside it kept updating as if nothing were wrong. The reason is straightforward in the way data-pipeline failures are: the company had ceased to exist as a public stock. It was acquired for $17.00 a share in cash, and trading ended April 1. Our pipeline never noticed. This is the post-mortem of a zombie ticker.

Episode window: March 30 - June 11, 2026.

Date	What our platform showed	What was true
Mar 30	price $16.97, composite 79, momentum 97	tender offer at $17.00 closing; the "momentum" was the price pinned to the deal
Apr 1	-	merger completed; shares delisted at $17.00 cash
Apr 1 - Jun 5	price $17.00 frozen; composite ~62; momentum ~75	no public stock existed
Jun 11	price $8.87, model fair value $16.23, "MoS +83%"	a meaningless data-vendor print on a dead symbol

Anatomy of the failure

It compounded in three stages. First, the momentum illusion: in late March the pillar read 97-99 - but the "trend" it measured was a stock glued to a $17.00 cash tender price. Deal arbitrage pins a price; a trend signal cannot tell devotion from momentum. Second, the silent death: after the April 1 closing, our market-data vendor kept serving the symbol with its final print, $17.00, rather than an error. Nothing in our pipeline asked "is this security still alive?" - so the row kept scoring a company that no longer traded. Third, the garbage resurrection: in early June the vendor's feed for the dead symbol began returning anomalous prints near $9, and on June 11 our batch ingested $8.87 - which, divided into a $16.23 model fair value computed from the acquired company's final fundamentals, displayed as one of the largest "margin of safety" readings on the platform. Every input was confidently wrong.

What we changed

On June 11 we removed GLDD from the scored universe and purged its cached rows - along with five other tickers (ODP, HBI, REVG, ZEUS, HTBK) the same investigation caught via the identical signature: a price frozen for 100+ consecutive rows because a corporate action had ended trading while the data vendor kept serving the last print. We added a permanent manual-exclusion control that survives universe rebuilds (the vendor still lists these symbols as active, so without it they would re-enter). Automated staleness detection - flagging any ticker whose price stops moving for implausibly long - is the durable fix and is tracked on our public roadmap of model improvements.

What to take from it

For reading this platform: the "last updated" stamp and the model-confidence readout exist for exactly this reason, and an extreme margin-of-safety outlier deserves a "is this price even real?" check before any other question. For reading any quantitative platform: every screener you will ever use sits downstream of data vendors, and corporate actions - mergers, delistings, ticker recycling - are where data dies quietly. The failure mode is not exotic; what varies is whether the platform tells you. We publish our misses because a model you can only trust when it is right is not one you can trust at all.

Reconstructed from archived platform score history and public merger filings (March-June 2026). Published as an education piece about data-pipeline failure modes; nothing here is investment advice.