Legal Notes — NCAA All Sports Rosters 2025-26 (v2.1.0)
This is an informational summary by the dataset author (Dharit Shah, dharits3@gmail.com, independent researcher). It is not legal advice, and no attorney-client relationship is created by reading it. It records the legal analysis behind the release design so that users, archives, and schools can see the reasoning. Where authority is unsettled, that is stated.
The dataset: individual-level rosters for all 28 NCAA sports, 2025-26
athletic year — 513,655 athletes, 1,087 schools, D1/D2/D3 — scraped from
official school athletics sites. As of v2.1.0 the public release is
named: first_name/last_name are distributed columns, alongside major,
previous school, height/weight, and per-sport season-stats sidecars. Every
distributed field is an institution-published fact from a public roster, bio,
or stats page; per-athlete demographic predictions and SES joins exist only
in a research tier that is never distributed. Companions:
DISCLOSURE_RISK.md,
ETHICS_REVIEW.md, OPT_OUT.md.
1. FERPA
Statute/regs: 20 U.S.C. § 1232g; 34 CFR Part 99.
Roster data — including names — is textbook directory information. The regulatory definition at 34 CFR § 99.3 expressly lists: name, photograph, major field of study, "participation in officially recognized activities and sports," "weight and height of members of athletic teams," dates of attendance, degrees/honors, and most recent previous educational institution. That list reads like a roster page schema because roster pages are the canonical use case — and it now also reads like this dataset's v2.1.0 schema: name, major, previous school, height, and weight are distributed fields precisely because they sit inside the enumerated directory-information categories the schools themselves invoked when publishing them. Directory information may be disclosed without consent under 34 CFR § 99.31(a)(11), subject to the § 99.37 conditions: the school must give public notice of what it designates as directory information and a window for students to opt out in writing. Schools that publish rosters have made exactly this designation.
FERPA obligations attach to funded institutions, not to third-party re-users. FERPA is a Spending Clause condition: it applies to educational agencies and institutions receiving Department of Education program funds (DOE, Protecting Student Privacy FAQ; overview: CRS IF13155). Its redisclosure limits (§ 99.33) bind parties that received education records from an institution under a consent exception — they do not reach an independent researcher who collected information the schools had already published to the open web under the directory-information exception. There is no FERPA private right of action even against schools (Gonzaga Univ. v. Doe, 536 U.S. 273 (2002), oyez.org/cases/2001/01-679); enforcement is DOE funding conditions on institutions.
Practical conclusion for this dataset: the scraped roster facts — names included — were disclosed by the schools themselves under FERPA's directory-information exception, and FERPA imposes no obligation on this project's collection or redistribution of them. v2.1.0 distributes names but still less than what schools publish: no photographs, no biographical text, no contact information, no birthdates.
The one residual edge — sharper in a named release: a student who filed a § 99.37 directory-information opt-out should never have appeared on a public roster page at all. If one did, that is the school's compliance failure, not a re-user's — but such a person could now appear by name in this data through no fault of their own. The dataset's opt-out process (OPT_OUT.md) covers this remainder: any athlete, a parent/guardian, or a school on an athlete's behalf can have the rows removed without stating a reason, with a 14-day target and a public removals ledger.
2. Right of publicity / NIL
The right of publicity targets commercial appropriation of identity — using a person's name/image/likeness to sell or endorse something — not the dissemination of factual information about them (overview: Free Speech Center, MTSU). A statistical dataset of institution-published facts performs no endorsement and attaches no one's identity to any product.
Controlling authority on named athlete facts — now directly on point. C.B.C. Distribution & Marketing, Inc. v. Major League Baseball Advanced Media, L.P., 505 F.3d 818 (8th Cir. 2007), cert. denied, 553 U.S. 1090 (2008) (opinion PDF via Yale ISP; Harvard L. Rev. note): the First Amendment protected a commercial fantasy-sports product's unlicensed use of players' actual names plus performance statistics, because the information was readily available in the public domain. In the de-identified releases that holding applied a fortiori; for v2.1.0 — athletes' actual names plus roster facts and season statistics, distributed free and non-commercially — it is the fact pattern itself, at a weaker level of commercial exploitation than the use the Eighth Circuit protected. For college athletes specifically: Daniels v. FanDuel, Inc., 109 N.E.3d 390 (Ind. 2018) (certified question), aff'd, 909 F.3d 876 (7th Cir. 2018) (Ind. opinion; 7th Cir.) — fantasy operators' use of college players' names, pictures, and statistics fell within the newsworthiness exception to Indiana's right-of-publicity statute, read broadly to cover "all types of factual, educational, and historical data."
State NIL statutes do not change this. The 30+ state NIL laws (and the House settlement framework) govern athletes' ability to be paid for endorsements and schools'/NCAA's ability to restrict that — they create no new cause of action against factual data compilations, and they inherit the right-of-publicity news/public-affairs exemptions (e.g., Cal. Civ. Code § 3344(d): "news, public affairs, or sports broadcast or account" is exempt; ABA overview of NIL landscape). Caveat honestly stated: right of publicity is 50-state common/statutory law with real variation; C.B.C. and Daniels are one circuit and one state court, and the First Amendment boundary is contested at the margins (transformative-use splits). A free, non-commercial factual compilation sits well inside the protected core those cases mark out; the margin matters for downstream commercial users, addressed next.
The license line, stated explicitly: the dataset is released under CC0 1.0 (LICENSE), and CC0 covers the database compilation only. It does not grant, waive, or license any name/image/likeness or right of publicity in any listed athlete. Those rights belong to the athletes and are not the maintainer's to convey. A downstream user who puts athlete names to commercial or endorsement use — merchandising, advertising, implied endorsement of a product — bears their own compliance burden under the publicity and NIL law of every relevant state. This note is carried in the release documentation as a no-endorsement clause; it puts downstream users on notice that the license conveys no persona rights.
3. Scraping
Collection was logged-out scraping of public web pages. The Ninth Circuit held the CFAA's "without authorization" concept inapplicable to publicly available, no-login data — hiQ Labs v. LinkedIn, 31 F.4th 1180 (9th Cir. 2022, on remand from the Supreme Court) (opinion) — reaffirmed in Meta v. Bright Data (N.D. Cal. 2024, applying hiQ to logged-out scraping; order). Caveats: hiQ is one circuit and arose on a preliminary-injunction posture (the case later settled with judgment against hiQ on contract grounds); website terms-of-service claims are civil breach-of-contract questions, not criminal access ones, and this project never accepted any school site's ToS by account creation. Scraping was rate-limited against public pages of the very institutions whose disclosures FERPA authorizes. The move to named distribution changes nothing in this section: the facts doctrine and the access analysis do not depend on which lawfully collected fields ship.
4. Copyright
Facts are not copyrightable. Feist Publications v. Rural Telephone, 499 U.S. 340 (1991) (opinion) — no "sweat of the brow" protection; a compilation gets at most thin protection for original selection and arrangement, which does not cover the underlying facts. Roster facts (name, position, height, weight, hometown, high school, major) are facts; schools' arrangement of them is not reproduced here. The US has no sui generis database right (unlike the EU); Congress has repeatedly declined to create one. Consequence: neither the schools' pages nor this dataset's facts carry enforceable copyright in the data itself. Unchanged by the named release — a name is as much a fact as a hometown.
5. Why CC0 + citation request
Because Feist leaves (at most) thin, uncertain rights in a factual compilation, an attribution license (CC BY / ODC-BY) would assert conditions on rights that likely do not exist — unenforceable and a known source of license-stacking friction. CC0 1.0 waives whatever thin rights exist and is the default at Dryad, Figshare, and Dataverse for exactly this reason (Dryad, "Why does Dryad use CC0?", CC0 FAQ). It also satisfies FAIR R1.1 (clear, machine-readable license) cleanly. Attribution is requested as a scholarly-norms citation request, not a license condition. Two limits of CC0, both stated in the release documentation rather than left implicit:
- CC0 waives the maintainer's rights in the compilation; it conveys no third-party rights — not the athletes' publicity/NIL rights (§2), and not any right to misuse names in ways other law forbids.
- CC0 makes use conditions unenforceable, so the release's use guidance (no individual-level sensitive inference, no re-identification-adjacent misuse) is a stated norm, not a term. See DISCLOSURE_RISK.md §3.
The research tier (BISG predictions, SES/income joins, tract identifiers, mobility joins) is not CC0; it is repo-internal, never distributed, and never deposited anywhere.
6. The opt-out mechanism as a good-faith control
No law surveyed above requires a removal process for republished
directory-information facts. One exists anyway
(OPT_OUT.md): row-level removal keyed on athlete_id, covering
the combined file, the stats sidecars, the per-sport splits, and the sample
files ("every distributed artifact" per OPT_OUT.md), propagating to the next version on
all three distribution platforms (Hugging Face, Kaggle, GitHub), 14-day
target, recorded in a public removals ledger
(RELEASE_NOTES.md, Removals section — empty at v2.1.0).
Legally it functions as a good-faith control: it operationalizes the § 99.37
residual (§1), gives named individuals — including parents/guardians of
listed minors — a working remedy no statute obliges, and evidences the
non-exploitative character of the release should any margin question in §2
ever be tested.
Rewritten 2026-08-14 for the v2.1.0 named release, carried forward from the 2026-07-07 notes; legal citations were verified against the sources linked above as of the original writing.