SAPINOVER OS v2.0 :: INITIALIZING...
000%
We read the current Form ATS-N of all 33 operating NMS Stock ATSs. In 53% of off-exchange dollar volume, the venue operator classifies counterparties by their measured trading behavior. In 29%, it can reclassify them without notice. And no venue anywhere lets a subscriber choose its own category.
Every NMS Stock ATS is required to describe, in a public filing, whether it sorts its participants into groups and whether anyone gets to choose who they trade with. The disclosures sit in Part III, Items 13 and 14 of Form ATS-N, they are amended whenever the mechanics change, and almost nobody reads them side by side.
We did. We pulled the current restated form of every operating NMS Stock ATS from EDGAR, resolved per venue rather than per filer because one broker-dealer can run several pools, and asked one question of each: who decides who you meet? That yields 33 rulebooks covering 99.77% of the $29.19 trillion in ATS notional reported to FINRA over our 32-week window.
The answers sort cleanly into four working levels. At the bottom, nothing: every order meets every order. One step up, opt-out tools: the venue defines no categories, but a subscriber can exclude specific counterparties. Above that, operator assignment: the venue sorts participants into categories by who they are, a broker, an institution, a liquidity provider, and interaction rules follow. And at the top, operator grading: the venue classifies flow by how it behaves, measured, usually monthly, from the venue's own post-trade data.
Here is the census, weighted by where the dollars actually trade.
Figure 1 · Who decides who you meet
Share of ATS notional by classification level of the venue's current Form ATS-N, Items 13 and 14. FINRA Rule 4552 weekly data, 32 complete weeks, 2025-12-15 to 2026-07-20.
53.5% of US off-exchange dollar volume now trades in venues whose operator classifies counterparties by measured behavior. Not by who the subscriber is. By what its orders did: where the midpoint moved after its fills, how often it added liquidity against taking it, its message rates and order-to-trade ratios. Fourteen venues run a regime of this kind, and they include four of the five largest pools in the market.
Add the venues where the operator assigns categories by identity rather than behavior and the total reaches 67.3%. Two thirds of the off-exchange market is a place where the operator, not the subscriber, decides what kind of counterparty you are.
And inside the behavioral group sits a narrower and more striking number: 6 venues, carrying 29.3% of all ATS notional, disclose that they may reclassify a participant or restructure its segment without telling them. Sigma X2's form states that if a subscriber's behavior materially changes, “the Taker Category will be updated without notice.” LeveL's states the operator may “designate new channels or consolidate existing channels without communication to the subscriber.” Virtu MatchIt's liquidity profile designations, in the form's own words, “cannot be contested.”
Nearly a third of the off-exchange market can change what kind of counterparty you are without telling you it happened.
The grading machinery is remarkably uniform across venues that built it independently. The core input, nearly everywhere, is the mark-out: where the midpoint stood some interval after your execution, compared with where you traded. Flow that is consistently followed by adverse midpoint moves gets sorted away from flow that is not. UBS grades on a rolling three months of reversion metrics. Bank of America's Instinct X scores message rates, add-take ratios and reversion statistics. Citi's Citi-ONE ranks liquidity-removing flow into four Taker Levels “based on a Mark-out Analysis performed by CGMI.” BNP's Cortex recategorizes some orders dynamically, mid-session.
Two design choices repeat. First, the metric is computed by the operator from the operator's own data, and the subscriber typically sees its grade only on request, if at all. Second, the grade attaches to flow, not to the firm: several forms are explicit that different channels or business units of the same subscriber can carry different classifications. The market has converged on treating an order's measured aftermath as its identity, which is precisely the practice we described from the market maker's side in our five-constituency map. What is new in the filings is that the venues themselves now run it.
Figure 2 · All 33 venues, placed
Each circle is one operating NMS Stock ATS, sized and positioned by share of ATS notional, in the lane of its classification level. Dashed rings mark the 6 venues whose forms reserve reclassification without notice.
Our typology had five levels. The census filled four of them. The one that came back empty is the one a market designer might have predicted would dominate: the venue defines categories and the subscriber elects among them. A menu. Not one of the 33 operating forms describes that structure as its base regime. Where categories exist, the operator assigns them; what subscribers elect is at most a filter layered on top of an assignment already made.
That empty middle is the finding under the finding. The off-exchange market did not split the difference between openness and control. Venue by venue, amendment by amendment, it moved classification authority to the operator, because that is where the data lives. An operator sees every fill on its book and can grade all comers with one consistent yardstick. A subscriber electing categories from a menu sees only its own experience. The information asymmetry decided the architecture.
The exceptions are as instructive as the rule. The single largest ATS in America, Intelligent Cross at 14.6% of notional, discloses no operator-run classification of counterparties at all: its Items 13 and 14 describe subscriber block-and-allow lists and hosted-pool exclusion tools, nothing more. Its differentiation lives elsewhere in the form, in how the match itself is timed. Two newer venues make refusal a feature: AlphaX US states it “will never make any counterparty selection decisions on behalf of a Participant,” and PureStream confines selection to subscriber-elected order instructions.
And then there is the night. All three overnight-dedicated venues, BOATS, Bruce and Moon, sit at the bottom level: no segmentation, no counterparty selection of any kind. Moon's form says it in five words: “Subscribers cannot select their counterparties.” Bruce's closes the item with “No other counter-party selection or restrictions are available.” BOATS answers the form's own yes-or-no questions with No and No. The only all-to-all corner left in the US off-exchange market is the overnight session, the same corner that is growing faster than everything around it and the same corner where, as we wrote in The Room and the Pipe, no negotiation mechanism exists at any hour.
For an exchange executive the census reads as a map of a product an exchange cannot ship. A national securities exchange's rules must not permit unfair discrimination among members; its book is the open field by statute, and that is its virtue. The regime described above, operator-run counterparty grading with interaction rules attached, exists in the ATS wrapper because the ATS wrapper is where it can exist.
That reframes the acquisitions we covered in The Room and the Pipe. Cboe owns BIDS, whose form discloses an operator-determined Scorecard level that subscribers can filter on. Nasdaq is acquiring LeveL, whose channel regime is among the most explicit behavioral frameworks in the census. An exchange group that owns a graded pool alongside an ungraded book is not merely diversifying volume. It is acquiring the one market-structure capability its own registration forbids, in the one wrapper where regulators have accepted it, disclosed on a public form.
The census also says where the white space is. Two thirds of off-exchange notional is operator-classified, and the classification engines are converging on the same mark-out mathematics. The differentiated positions remaining are the extremes: the genuinely neutral venue that makes refusal its product, and the fully graded venue that makes the grade transparent enough to trust. The crowded middle is variations on the same monthly mark-out arithmetic.
For a principal market maker the census formalizes something the desks have known by feel: the profiling now runs both ways. The same firms that tune their own interaction by measured counterparty signature are themselves being tiered, by name, in public filings, on the venues' own mark-out arithmetic. A taker category at one venue, a liquidity profile at another, a source category at a third, each computed from the venue's book, each governing what your quotes are allowed to meet.
Three practical consequences follow from the filings alone. Your effective venue access is now a portfolio of grades that can move monthly, and at 6 venues, without notice. Your behavior at one venue is invisible to another, so the same strategy can carry different grades in different pools, and managing the dispersion is itself a strategy. And because the grade attaches to flow rather than firm, internal segregation of strategies by connection or channel is no longer only an operational choice; it is grade management, and several forms explicitly accommodate it.
For an asset manager the asymmetry is stark. The venues holding two thirds of off-exchange volume grade your flow continuously from their own data. Your public disclosure in the other direction covers almost none of what you send: the public routing reports cover held orders only, and the amended execution-quality reports arriving this September still exclude not-held orders by rule. The institutional parent order, worked through a broker's algorithms, is graded everywhere and disclosed nowhere.
One instrument closes part of the gap, and it is the least used document in Regulation NMS: Rule 606(b)(3). On request, your broker must disclose, for your not-held orders, venue-by-venue routing and execution detail for the prior six months. It is customer-specific, it is not public, and the clock is seven business days. Paired with this census, it supports concrete questions: which of my flow reached venues that grade counterparties without notice, what channel was my flow assigned to, and did my broker elect any counterparty filters on my behalf. The venues wrote down how they sort you. The request letter is how you find out what they decided.
The levels are judgment calls at the margin. The line between identity-based assignment and behavioral grading is clean in the paradigm cases and blurry in two or three venues that gate eligibility with behavioral tests. Moving every borderline call down a level changes the behavioral share by low single digits and the operator-classified total not at all.
Disclosure is not practice. A form describes what a venue may do and must disclose, not how aggressively it does it. A venue with sweeping disclosed discretion may use it rarely; a venue with modest disclosure may apply it energetically within its bounds. The census measures the architecture, not the temperature.
Forms lag. Amendments post on a delay and we read each venue's latest restated form, which for a few quieter venues dates from 2024 or 2025. A regime adopted since a venue's last restatement would be missed, though material changes require prompt amendment, so the lag is bounded by the rule itself.
And the notional weights are a snapshot. Shares move. The census weight of the behavioral group has a floor well above half under any recent window we can construct, but the precise 53.5% belongs to this window alone.
The census. The current Form ATS-N of every operating NMS Stock ATS, from EDGAR: the latest full restated primary document per 013- file number, resolved per ATS rather than per filer, because one broker-dealer can operate several pools. 33 venues, covering 99.77% of ATS notional over 32 complete FINRA Rule 4552 weeks (2025-12-15 to 2026-07-20). Of the 36 MPIDs in the FINRA NMS-equity data, Luminex ceased operations on March 27, 2026, and OTC Link's two OTC-market systems file no separate ATS-N. Moon's ATS-N is filed under OTC Link LLC's CIK as file 013-00200; the form's MPID field, not its name field, identifies it.
Classification. Part III Items 13 and 14 of each form were extracted and classified into the four-level typology described in section 01, each classification carrying a verbatim supporting quotation. Every quotation printed in this piece was verified character-for-character against the extracted filing text before publication. Blue Ocean files its narrative as PDF exhibits amended in changed pages only, so its classification rests on the structured answers in its own XML filing: Item 13(a) segmentation, No; Item 14(a) counterparty designation, No.
A formatting trap. Form ATS-N styles Item 14 as “Counter-Party Selection” with a hyphen. Unhyphenated text searches undercount these disclosures, and all matching here was hyphen-insensitive.
On the exchange comparison. The statement that an exchange cannot operate a regime of this kind rests on the Exchange Act's requirement that exchange rules not permit unfair discrimination and on the fair-access provisions of Regulation ATS, which apply to an ATS only above volume thresholds. It is a statement about rulebook architecture, not a legal opinion about any particular design.
