QL Private Foresight grades the direction of the next mark move. This board grades the event: a sealed monthly probability that each private company IPOs, gets acquired, raises an up round, reprices down, or stays quiet over the next 12 months — graded automatically against confirmed outcomes when the horizon elapses. Baselines are stamped at seal time; sealed claims are never edited; the multi-class Brier score and per-event calibration tables are published win or lose.
Each call is a full probability vector with its driver breakdown at issuance — filing status, valuation scale, stage, round age, last step-up, and the secondary-market gap where a QL Mark exists. Deterministic, disclosed, no black box. Sort by the event you care about.
| Company | Sector | Next 12 months | Sorted by | Top call | Drivers |
|---|---|---|---|---|---|
Loading the sealed board… | |||||
What do this month's sealed odds imply, category by category, before a single claim is graded? Pick an event; every chart below is drawn from the board's own published aggregate — a projection of its claims, never a record. The first real grades land at the horizon.
Every sealed call's claimed probability for this event. Drag the threshold to keep only the board's most convinced calls and see what the sealed odds imply for them.
Sweep the threshold from 0 to 80%: the implied hit rate among the calls you keep rises, the share of the board you keep falls. Your slider is the marker.
The sum of the sealed probabilities per category — what the board expects the next 12 months to hold across all — names.
Mean claimed probability for the selected event across the board's largest sectors.
Four ways to interrogate this month's sealed board before a single claim is graded. Everything here is a projection of the board's own claims — sampled, re-composed, or re-thresholded — never a record. Sampling the claims cannot test the claims; the graded record below is the only test, and it is published win or lose.
Every call's outcome sampled from its own sealed vector, thousands of times. The dashed band on each bucket is the 90% range the board implies — each bucket sits on its own scale, so the question is only ever inside the band, or out. Roll one year and see where it lands.
Pick a sealed call and re-run xo-v1 with different drivers — a filing, a staler round, a secondary discount. The dashed pill is what the board sealed; the orange one is your what-if. Same base rates, same deltas, same clamps.
What the Brier score would read a year from now if the board is exactly calibrated — and how wide the luck band is on a board this size. The model must land below both grey markers to have beaten its baselines.
The calibration table fills only when boards grade. This is what each claimed band would realize under perfect calibration — with the 90% band a finite bucket carries. A realized rate inside the whisker proves nothing; outside it is a real miss.
Five dissections of this month's sealed board: what moved the odds, how the claims ladder by stage, scale and round age, how concentrated the expected events are, how decisive the vectors are, and what changed since last month. Every chart is computed from the board's own published claims — aggregate only, never a record. The graded record is the only test.
Start every call at the base rates, switch the driver families on one at a time in catalog order, and watch the board's mean probability walk to the sealed mean. Sequential by construction — softmax is not additive — so the order is the catalog's, and the last step lands exactly on the sealed number.
The sealed mean vector, bucket by bucket. Each bar is one bucket's mean claim across the five events; the number is how many sealed calls sit there. This is what the model believes about a Series C at $1B with a 3-year-old round — read it, then argue with it in the driver lab.
Sort the board by one event's claimed probability and cumulate: what share of that event's expected count do the top calls carry? A curve hugging the diagonal means the odds are spread evenly; a curve bowing up means a few names carry most of the expected events — and the record will be decided by those few.
A vector that spreads across all five events is undecided (entropy near 1); one that piles onto a single event is decisive (near 0). A decisive board earns or loses more per call — the record will show which.
The sealed board against the one before it: names entering and leaving the 500, how the mean vector moved, how each returning company's chance of any event shifted, and which top calls changed. A repriced round, a filing, or a stale mark crossing a threshold is what moves a call — the movers list names the ones you can see at your plan.
Sealed claims whose outcome is already observable before their board's 12-month horizon. Nothing here is in the record yet — a board only grades, in full, when its horizon elapses. This strip exists so you can watch claims meet reality in real time.
When the board says 30% chance of a down round, it should happen about 30% of the time. One table per claimed event — they fill as boards grade. The record must beat both published baselines: uniform (0.800) and the model's own sealed base rates.
| Claimed band | Calls | Realized |
|---|---|---|
No graded boards yet. | ||
Exit Odds are model probabilities from disclosed data, not predictions of any transaction, offer, or listing; private-market data is sparse and every call discloses its drivers. Nothing here is investment advice, an offer, or a solicitation. Past record ≠ future results.
Monthly sealed probabilities, per priced private company, for five 12-month outcomes: IPO, acquisition, up round, down round, or quiet. Boards are graded automatically against observable events and multi-class Brier-scored in public against base-rate and uniform baselines.
IPOs and confirmed acquisitions from the identity ledger, and disclosed repricings beyond a ±5% band. A company counts as quiet only once the 365-day horizon elapses with no observable event.
The sealed vectors feed QL Deal Screen’s exit pillar and its outcome simulation. They are model probabilities with a published record — not investment advice.