Roughly 70% of companies that IPO trade below their offer within a couple of years. The scorecard's whole job is to separate the ~30% that rise from the ~70% that fade. The edge isn't the prospectus — everyone has that. It's that we can see the private history in MARS: every round, every backer, equity vs debt, the valuation path. That private lens is what a public-filing-only read can't see.
Two IPOs can file near-identical prospectuses and have opposite private trajectories — an up-round vs a down-round, tier-1 crossover backers vs no-name money, equity-funded growth vs a debt-financed buildout. The public offering hides all of it; MARS surfaces it. The scorecard turns that asymmetry into a grade.
The dimensions, scored on a real, marquee IPO — CoreWeave (CRWV), IPO 2025-03-31 at $40 (deal $1.7B). The point isn't the grade; it's that the scorecard surfaces a tension a headline read misses — elite backing wrapped around a debt-funded balance sheet:
| Dimension | Grade | What the data says |
|---|---|---|
| Backer pedigreeMARS · who + when | A | Elite. Pre-IPO Series C ($1.1B) led by Coatue, Altimeter, Fidelity, Magnetar; the IPO itself anchored by NVIDIA (strategic — whose chips they buy). Top-decile validation. |
| Capital structureMARS · equity vs debt | D | Debt-financed buildout. Of ~12 verified rounds, nearly all are Debt / Convertible Debt — ~$2.3B pre-IPO and a ~$19.7B torrent after (incl. a single $8.5B facility). Only one true equity round in the trajectory. Structurally fragile if the AI-capex cycle wobbles. |
| Offering mechanicssections · 424B4 | C | Proceeds go to debt, not growth. ~$1.0B of net proceeds to repay a term loan + $0.5B to a debt escrow. Dual-class 10:1 founder control. 180-day lockup (expired 2025-09-27). |
| Financials trajectorysections · prospectus | — | Extraction gap — the prospectus financials weren't pulled for this issuer
(no_financials_extracted). Falls back to the post-IPO 10-Q. A real coverage hole to close. |
| Post-IPO catalystslockup · insider · 13F | next | Lockup-expiry supply, insider behavior at unlock, institutional 13F uptake — the dimension that turns a static grade into a time-series. Roadmap. |
The read: CRWV scores top-decile on backing and bottom-decile on balance-sheet fragility — a high-conviction, high-leverage bet on the AI-capex cycle continuing. That tension is the honest output, and it's exactly the nuance the 30/70 split needs — invisible to anyone reading only the "hot AI IPO" narrative.
A prospectus tells you the deal size and the risk factors everyone else reads. MARS tells you how the company was actually financed and by whom — the equity/debt mix, the tier of the backers, the up-round/down-round path. Those are the features that separate survivors from faders, and they're structurally unavailable to a public-filing-only competitor. That's the moat, same as the graph: depth = domain × extraction × linkage.
Every grade is deterministic — backer tier, debt/equity ratio, use-of-proceeds split, insider-sell % are computed, not vibed. The model writes the story around the numbers; it never invents them. Same discipline that keeps the report builder honest.
mars_funding_history / investor_portfolio — every round, backer tier, equity-vs-debt. The two hardest-to-replicate dimensions.
sections__ipo_offering — use of proceeds, dual-class, lockup schedule, greenshoe. 98% of the IPO universe.
sections__ipo_financials — prospectus revenue/margin/burn (mind extraction gaps → post-IPO 10-Q fallback).
sections__ipo_risk_factors — classified risk language (going-concern, concentration, dilution).
Lockup-expiry behavior + institutional uptake — the catalyst dimension.
round_date instead and
resolve the entity at present. And use obscure IPOs for scoring so the model can't recall
famous outcomes._ipo_scorecard_panel + ipo_scorecard REGISTRY section + sections__ipo_offering prefetch — committed locally (Task #25), undeployedThen: close the financials extraction gap (10-Q fallback), add the post-IPO catalyst dimension (lockup/insider/13F), and ship the section into the served report engine.