"Language drift" sounds abstract. What does it mean in practice? This post shows four real companies whose filings scored above the distress ceiling — what they wrote, why it scored high, and what happened next.
The score is high when a company writes things that are simultaneously unusual for itself (new or escalating language) and unusual across the whole corpus (not something every company was saying that year). Both conditions have to hold. Corpus-wide boilerplate moves no scores. (This is the high end of the same score whose low end is a language-stability factor.)
SVB's 2022 annual report — filed February 24, 2023 — contained language about unrealized losses in its securities portfolio that had escalated sharply from prior filings. Phrases around the scale of losses on held-to-maturity and available-for-sale securities, the sensitivity of that position to rising rates, and the reliance on short-term borrowing to manage the mismatch appeared at a frequency and corpus-wide rarity that few other companies were matching that year.
Why it scored high
The language about portfolio losses wasn't new that year — rising rates affected everyone, so that language was downweighted as corpus-wide. What made SVB unusual: the specificity and scale of its exposure language was rare across the corpus. JPMorgan mentioned rate risk. SVB disclosed a position that, read carefully, described a structural liquidity problem. The corpus-wide normalization is what surfaced it.
At the time of the filing, analyst consensus was 12 Buy / 11 Hold / 1 Sell. Forbes had named SVB one of America's best banks weeks earlier. The stock had not yet moved. The FDIC arrived March 10 — 14 days after the filing date.
At 138.5, BBBY's score was 2.7× the distress ceiling — the highest in our labeled set. The April 2021 filing contained dramatically escalating language around inventory management failures, the strategic reset underway, and accelerating store closure obligations. The language specifically around liquidity, vendor payment terms, and the pace of the business transformation had shifted significantly from prior filings and was unusual across the corpus.
The context at the time
When this filing appeared, BBBY stock was still up ~25% in the meme stock rally. Retail sentiment was bullish. The filing language told a different story — two years before the April 2023 bankruptcy filing, the semantic drift was already extreme.
Party City went bankrupt in April 2023, so it belongs in any honest accounting — and on the full corpus, FilingDrift misses it. Its peak pre-event score (46.3) never clears the 51.5 ceiling. This is worth understanding because it's the same property that makes the long-side factor work.
Why it's a miss
Party City's distress language was common-distress vocabulary — goodwill impairment, restructuring, generic impairment — the kind that shows up across many companies. Because the score is normalized corpus-wide, that common language is heavily discounted. The signal rewards unusual distress language (SVB's specific liquidity-mismatch disclosures, NKLA's EV-program language); it discounts garden-variety impairment. On a smaller, narrower corpus the same filing scored above the ceiling — the result is corpus-relative by design. We count Party City as a miss rather than pretend otherwise.
Rite Aid's 2023 filing contained escalating language around opioid litigation exposure — the scale of potential liability, the liquidity implications, and uncertainty around the company's ability to continue operations. The specific combination of phrases around legal reserves, going concern uncertainty, and available liquidity had shifted dramatically from prior filings and was unusual across the corpus.
Why 167 days is interesting
Rite Aid is the shortest lead time in our labeled set. The filing language was already acute — specific enough that the 167-day window to bankruptcy was the right read on urgency. Compare to BBBY (~24 months): the score alone doesn't tell you the timeline. The severity of the language and the rate of escalation across filings are better indicators of how much runway remains.
In each case, the elevated score came from language that was:
You can see the specific anomalous sentences for any of these companies — and the current scores for ~4,900 tracked companies — on FilingDrift. The company pages show the actual text from the filing that drove the score.
Questions about a specific company or methodology? Email hello@filingdrift.com