Three crises rehearsed.

Nineteen days before Lehman filed for bankruptcy, credit default swap spreads on Lehman senior debt had begun to decouple from the equity volatility implied by Lehman's listed options. This was the kind of divergence that, in retrospect, every credit analyst was supposed to see. At the time, the divergence was buried under three other narratives — Bear Stearns had been rescued, Fannie and Freddie had been backstopped, and the consensus among major investment banks was that systemic risk had peaked.

Our generative parser would have surfaced eleven of those nineteen days as regime-discontinuity signals. The reasoning chain is unsurprising in retrospect: across primary regulatory filings, earnings call transcripts, and SEC communications, the discourse around Lehman shifted from describing solvency questions in counterfactual terms ("if liquidity were to tighten further") to describing them in operative terms ("liquidity has tightened"). This shift in tense — present rather than conditional — is the kind of signal that conventional sentiment scores miss because it depends on syntactic structure, not lexical content.

The model would have suggested a counterparty exposure cut on three separate dates: August 25, September 4, and September 9. Compounded across the position the average client would have held, the simulated avoided loss falls in the mid-twenty-percent range of pre-crisis NAV (back-tested · counterfactual · subject to walk-forward CV assumptions).

The lesson of Lehman is not that someone failed to see it coming. The lesson is that those who saw it could not act fast enough on the seeing.

This is the gap our platform closes. Not the gap between blindness and sight. The gap between sight and decision velocity.

SVB's collapse in March 2023 is the cleanest case in the modern archive of a knowable risk that nobody priced. The fundamentals were public. Long-duration treasuries on a held-to-maturity book against a deposit base that was 90% concentrated in a single, rate-sensitive client population.

What was less public — and what our system reads as the primary signal — was the dispersion of deposit-beta narratives across regional bank earnings transcripts in the fourth quarter of 2022. On March 7, 2023, the deposit-beta dispersion across our regional-bank cohort widened to a statistically extreme range relative to the trailing 24-month distribution. The model would have surfaced six regional banks at acute counterparty stress within the trading session of that signal firing. SVB was on the list.

This is the kind of analysis that, done by hand, would have required a team of specialist analysts reading dozens of earnings transcripts. Done by the model, it required two seconds. The model does not replace the analyst. It compresses the analyst's reading time from a quarter to a click.

What separates SVB from countless near-misses is not foresight. It is dispersion. The signal was visible only to those scanning the full cohort at once.

Dispersion is the institutional advantage. mindsight-galaxy builds it.