CLOSE V2 · PROSPECTIVE SHADOW

Find tomorrow's movers
before tomorrow.

A nightly ranking system for U.S. equities, frozen at the close and judged only after the next session arrives.

Explore tonight's pulse ↘ See how it works
Full core universeCS + ADRC
D-1 closedecision clock
Frozen modelno nightly refit
NIGHTLY SIGNAL CONSOLE Awaiting first run
PREVIEW
01
—First prospective pick will appear here
—
Model frozen PIT features SEC reserve only
D-1 CLOSE✦FULL MARKET RANKING✦POINT-IN-TIME✦TOP10 / TOP20✦IMMUTABLE RECEIPT✦60 SESSION CONFIRMATION✦DISCOVERY ≠ EXECUTION PNL
01 / PULSE

Tonight, compressed.

One market close in. One ranked universe out. No outcome data is allowed into the prediction receipt.

LATEST RANKING

Next-session watchlist

Awaiting the first prospective Close V2 run.

The shell is live. Real picks will replace this state without changing the interface.

MODEL STATE

Close V2

FROZEN
V2A3
FamilyLambdaRank
SignalsTechnical + PIT history
SEC in scoreNo
Input gateFull universe
MODEL SHA9b8b7c…abcbcf
PROSPECTIVE CLOCK

Evidence, not vibes.

SHADOW
0/ 60

Formal review unlocks only after 60 untouched settled sessions.

OOF R@1011.65%
OOF R@30059.92%
EXECUTION REALITY

Can the move still be caught?

E0
D-1
close
D
open
D
high
Open proxy sessions0
Top10 open→high—

Spread, slippage, halts and fill failures remain outside E0. No executable PnL claim.

02 / METHOD

A narrow contract, by design.

The system is deliberately boring where rigor matters: fixed clock, fixed information set, fixed model artifact.

01
◐

Close

Wait for the U.S. regular session to finish. Nothing from tomorrow enters the feature set.

02
⌁

Rebuild

Reconstruct the point-in-time CS + ADRC universe and promoted technical history features.

03
↗

Rank

Score the entire eligible core universe with the frozen Close V2 model. No Top300 pre-gate.

04
◇

Freeze

Write Top10, Top20 and full ranking with cryptographic receipts before the next outcome exists.

03 / VALIDATION

Research evidence has layers.

Development evidence, prospective evidence and execution evidence are shown separately so one cannot quietly impersonate another.

A3 WALK-FORWARD OOF121 TEST SESSIONS
11.65%

Actual Top10 Recall@10

How often a true next-session Top10 mover already appeared inside the model's Top10 across the committed walk-forward test folds.

Development evidenceNot prospective
CANDIDATE COVERAGEA3
59.92%

Actual Top10 Recall@300

The reason Top300 was useful in research — and the reason it is not allowed to become a permanent production input gate.

Full-universe scoredTop300 diagnostic only
PROSPECTIVELIVE CLOCK
0/60

Untouched settled sessions

This counter, not retrospective demos, determines when Close V2 earns its first formal prospective review.

CollectingNo early stop
04 / EXECUTION

A great ranking can still be a bad trade.

So the product keeps discovery and execution on different rails. Close V2 asks what moved. The execution layer asks whether a real entrant could still capture it.

E0 · Regular-open proxy E1 · Timed / premarket entry E2 · Spread, slippage, halts, fills
E0Price realityDescriptive
E1Entry timingFuture validation
E2MicrostructureRequired for PnL
!

No executable-return claim. The current product remains a discovery system until the execution stack earns that language.

05 / RESEARCH VAULT

What survived. What didn't.

Research history stays visible, but rejected ideas do not leak back into the production score.

RESERVE
A5A

SEC identity + timestamp ledger

Audited PIT infrastructure stored for a future, more specific hypothesis.

Dormant research asset
REJECTED
A5B

Structured SEC event features

Did not improve the frozen full-universe ranking gates.

Not in Close V2 score
FUTURE
V2.x

Only after evidence

Any new signal must freeze its hypothesis and beat the current baseline out-of-sample.

No silent upgrades
MOVE IT UP

One prediction.
Then reality gets a vote.