Editorial review · 260605-002
How ORA’s piece on The Stanford Law Study Tells Us More About Who Pays Than Who Wins scored.
Read the article →Solid reporting. Some issues but credible overall. The reader is well-served.
Accuracy
Core study figures (75.33-75.92% win rate, 3.53% vs 12.06% harm flagging, 16 professors, 40 questions, ~3,000 matchups) are attributed to a Stanford Law press release dated June 3, 2026, which is post-cutoff but sourced (-0). The $60,000 tuition and 40% adjunct figures are load-bearing but weakly cited, with the ABA stat unlinked and tuition citation pointing to a Hacker News thread (-10). NotebookLM is reasonably characterised and methodological caveats are flagged honestly.
Balance
The piece is openly opinionated but represents the methodological defence of the study, the legitimate concerns about hallucination, and the limits of the contract-law domain fairly (-0). It centres a labour-impact frame that other coverage missed and names it as such rather than smuggling it in (-0). One deduction for not quoting any administrator, dean, or legal-AI vendor in their own voice on a contested policy topic (-8).
Concerns (4)
- minoraccuracy
“now exceed $60,000 per year”
Citation points to a Hacker News thread, not a tuition source.
Evidence: Footnote 2 is a HN front page link, inadequate for a specific dollar figure.
- minoraccuracy
“More than 40% of law school instruction is delivered by adjunct or contingent faculty”
ABA 2024 data cited inline but no link or document reference provided.
Evidence: The stat block names ABA but supplies no traceable citation in footnotes.
- minoraccuracy
“Stanford Law study findings and methodology”
Post-cutoff, source attributed to Stanford Law press release.
Evidence: Published June 3, 2026, two days before review date; cannot independently verify.
- minorbalance
“(source set)”
No administrator, dean, or AI vendor quoted in their own voice.
Evidence: Piece characterises vendor behaviour and institutional motives without direct sourcing from those parties.
Reproducibility
How this review works: read the methodology. Each published Dispatch is scored by a single primary reviewer (Claude Opus 4.7) against the public rubric. A second model (Gemini 2.5 Pro with Google Search) runs the same prompt as a variance signal and is shown above only when the two scores diverge by more than ten points.