TAC Meeting Notes, April 2026

The official summary minutes. Nine pages, 5,613 words, authored by Jessica Fenby (MDE) and generated 2026-05-14 — roughly three weeks after the meeting. This is the single richest source in the corpus and the closest thing available to a transcript.

ImportantThis document is not published

The PDF is not in the public repository. It records Session 4 in full, including the MTTC test-form overlap ceilings, the form rotation policy, and the per-group form counts — all test-security sensitive. Redacting those from the corpus while shipping a PDF that contains them verbatim would accomplish nothing.

It is held in the restricted layer at restricted/raw/2026-04-23/minutes/, gitignored and local only. The extracted specifics are in restricted/mttc-form-security-detail.md.

The cost is auditability, for this one source. Every other document in raw/ can be checked against the page that summarises it. This one cannot, by anyone without the file — which includes every reader of the published site. The corpus asks to be trusted on its most-cited source, and that is worth stating plainly rather than burying.

Everything else from the meeting stays public: the agenda and all five presentation artifacts. None contains MTTC material; there was no MTTC deck.

Structure

An attendee block per day, then five sessions. Each session opens with a two-to-four sentence third-person abstract and then drops into a bulleted record using a consistent speaker-role-prefixed convention:

  • TAC Question: / TAC Feedback: — from the committee
  • OEAA Question: / OEAA Feedback: / OEE Question: / OEE Feedback: / Pearson Question:
  • A bare bullet immediately following a TAC Question: is the presenter’s answer.

Those markers are reliable enough to parse, which is what makes the corpus buildable at all.

What it does well

It preserves candour. “None of this is great” survives verbatim, quotation marks and all, as does the committee’s blunt assessment that a handful of forms produced over several years, each sharing most of its items with the others, cannot be called meaningfully different. (That remark quotes a specific overlap figure and is held verbatim in the restricted layer.) It carries real methodological specificity — NTILE(50), the 1,000-per-bucket stability rationale, the Wainer et al. (2001) unidimensionality procedure. The five-session structure is clean and the topical sub-headings are sensible.

What it loses

Individual attribution. Every question and every recommendation is attributed to a body, never a person. Five TAC members with distinct areas of expertise were in the room; the record cannot tell you which of them raised the ESSA amendment problem or which pressed on defensible inferences. The one exception is the closing note that Damian Betebenner offered to collaborate with Ji Zeng.

Everything in the images. The substance of the CT-PO simulation study is in chart images — see CT-PO vs SBAC Simulation Results Deck. None of it reaches the minutes beyond a one-line conclusion.

Everything in the speaker notes and hidden slides. See TAC Growth Update Deck, April 2026. The repeal of educator evaluations and school grades from state law, and the post-COVID limit on WIDA projection horizons, appear nowhere in the minutes.

Catalogued defects

Recorded here rather than corrected silently, because the error rate of the current process is evidence for the argument in documents/ai-native-tac.qmd.

In the minutes Should be Kind
Session 5: 04/204/26 04/24/26 Date typo
Damion Betebenner Damian Betebenner Name
JI Zengs's Ji Zeng’s Name
OEEA Question OEAA Question Body
The APG timeframe The AGP timeframe Acronym
Meaningly Different Test Forms Meaningfully Different Test Forms Heading
Adequate Growth Targets (AGPs) Adequate Growth Percentiles Term
Correlation between all content areas is between 2.7 and 2.8 0.7 and 0.8 Figure — decimal lost
Cronbach's coefficient theta Cronbach’s coefficient α Statistic misnamed
unideminsional unidimensional Spelling
Smarter Balance s using Smarter Balanced is using Garbled
piece wising modeling piecewise modeling Term
Englisher language (source deck) English language Spelling in source

Two of these matter substantively. A correlation “between 2.7 and 2.8” is impossible and would mislead any reader who did not already know the answer. And Cronbach’s α and coefficient theta are different statistics — the deck says α, and α is what the Wainer procedure calls for.

A discrepancy the minutes probably get right

On the DRC exploration, the minutes record predictors SGP 2023 and SGP 2024 with outcome SGP 2025. The source deck’s slide 14 says predictors SGP 2024 and SGP 2025 with outcome SGP 2025 — which is self-contradictory, since an outcome cannot also be its own predictor. The minutes are coherent and are followed here.