Susquehanna (SIG) Interview ProcessTier 2

The Mettl screen exactly as it was sat in September 2026 (on-screen title 'The Susquehanna Problem Solving Assessment', one section, palette of 17 tabs with an 'Attempted: n/17' counter, 'Revisit Later' bookmarks, autosave). All 17 questions recovered word for word with their figures, in the real order and the real answer widgets: numerator / denominator blanks for fractions in simplest form, a dollar-prefixed blank (dice game), a %-suffixed blank (bakery), hh:mm blanks (canoe), a two-digit AB entry (skyscraper grid) and one five-option radio question (knights and knaves). Every family is a number-rotation of the classic bank drilled in the 22-question set below, but three prompts changed shape this sitting: the dice game pays the LOWER roll and $0 on doubles, the bakery runs out of croissants (the 80% item), and the news question asks for P(fake). Fourteen of the 17 have a bank twin (linked); the balance mobile, the exercise table and the plain 4x4 skyscraper grid live only in the timed test.

The signature SIG screen is 17 timed questions, first ~9 easy then ~8 hard - probability, expected value, Bayes, constraint logic, gambler's ruin, and a lattice DP. The bank is fixed and only the numbers shuffle between sittings, so drilling it for speed is the whole strategy. Sitting reports (Aug 2026, and the Sep 2026 sitting captured word for word above) match this replica family-for-family - no genuinely unseen problems - though a couple of items run heavier live: the canoe question makes you solve for the paddling speed p, and the balance mobile hides the simple halving branch among many distractor branches (~2 min to find it). This drill runs all 22 recurring families in one 60-minute sitting; for the exact 17-question sitting use the verbatim Sep 2026 replica above.

QuantVault's own candidate files describe the 17-question, 60-minute sitting above, but other prep sites, and a candidate quoting SIG's invitation email, describe a 20-minute Mettl Quantitative Evaluation, cited as 16 typed questions with no going back (14 short items and 2 longer ones; +1/-1 per one source). Both formats appear to be in use, so this replica covers the short one.

A full 2027-cycle sitting of the 17-question screen leaked in mid-2026 (16 of 17 problems recovered). Thirteen are number-rotations of the classic bank drilled in the timed screen above; three question families are genuinely new to the rotation - a two-dice payoff-matrix EV with a doubles penalty, a stars-and-bars flavor count, and an expected-consecutive-pairs linearity question. This timed set runs exactly those three, with the sitting's numbers and figures.

A proctored coding OA (camera/mic/screen-record) that rewards clean modeling and edge-case handling over exotic algorithms - the pair-sum-with-updates problem is the one real optimization gate. All four recurring problems are below.

A 4-problem CodeSignal coding test for the Strategy Developer track: matrix construction and manipulation, a cyclic transit-log time computation, and a count-pairs aggregation. All four are below.

The Philadelphia equity-research OA - finance-flavored, three sections: easy stats / expected value, option & equity VALUATION (multiples, EPS vs P/E, reading statements), and general market knowledge (M&A, board structure, bonds). Take the timed screen below; it expects a real finance background, not the quant math OA.