The 2026 draft, measured.

What the IACC is and why its language matters

The Interagency Autism Coordinating Committee is a federal advisory committee within the U.S. Department of Health and Human Services, established by the Children's Health Act of 2000 and reconstituted under the Combating Autism Act of 2006 (P.L. 109-416). Its statutory mandate, codified at 42 U.S.C. § 280i-2, is to coordinate federal autism activities, monitor research and services, make recommendations to the HHS Secretary, and develop and update a strategic plan for autism research, services, and policy. The committee has been reauthorized five times, most recently by the Autism CARES Act of 2024 (P.L. 118-180), which extends its authority through September 30, 2029, and broadens its scope toward services, supports, and policy alongside research.

The IACC operates under the Federal Advisory Committee Act (FACA), which requires balanced membership, public meetings, and publicly available records. Its membership by statute includes federal officials — the directors of NIH, NIMH, CDC, and other HHS agencies — alongside public members appointed by the HHS Secretary: researchers, clinicians, self-advocates, and family members. The designated federal official is Susan A. Daniels, Ph.D., Director of the Office of National Autism Coordination at NIMH.

How earlier plans were produced. Between 2009 and 2023 the committee produced strategic plans through a multi-stage process: a public Request for Information, analysis of the response, scientific workshops or working-group meetings, publication of draft sections for review, full-committee deliberation with public comment, a committee vote, and NIH publication. The 2021–2023 plan, for example, drew on 537 public comments, a working-group meeting in July 2022, draft review across three full-committee meetings, and a final vote before publication in September 2023. Each plan built incrementally on its predecessors, reusing established research questions, retaining community-negotiated vocabulary, and adding new priorities without discarding the existing framework.

January 2026: a complete reconstitution. On January 28, 2026, HHS Secretary Robert F. Kennedy Jr. appointed 21 new public members to the IACC, replacing every outgoing public member at once rather than staggering replacements as in prior cycles. Dr. Sylvia Fogel was named chair. The current public roster includes clinicians, parent advocates, nonprofit executives, a nonspeaking college student, and a high-school student — but no research university faculty in autism neuroscience, no representatives of large autism service organisations, and no members of the Autistic Self Advocacy Network or similar self-advocate organisations that sat on prior committees.

The resulting working draft, released for public comment on July 31, 2026, was assembled without the multi-stage RFI-and-workshop cycle that produced every earlier plan. Its own preface states: "Our drafting process did not have the benefit of a complete institutional record or formal handoff from the prior planning cycle." The linguistic analysis that follows measures what changed when the committee, the process, and the plan were all replaced at once.

Where the draft's style sits versus every earlier plan

Each bar represents one style measurement. The first two views express it as a z-score — against the eight earlier strategic plans, or against all 45 earlier documents; red bars indicate |z| > 2, values outside the range of anything the committee has previously published. The third view presents the pairwise comparison with SP-2023 as a ratio. The draft is the most obligation-laden document in the record ("should" at 7.8× the SP-2023 rate) and employs the longest, most polysyllabic vocabulary the committee has produced. On grade-level readability it falls alongside the 2017 and 2023 plans rather than beyond them, because its sentences are shorter even as its words are longer.

The readability finding rests on the main prose. Removing every list-typeset-as-prose block in the draft changes its overall Flesch–Kincaid grade by 0.1. The obvious objection — that Appendix C reaches FK 30.6 because a clinical-protocol list was set as continuous prose (373 words, 5 sentences, 58 semicolon-delimited items, longest sentence 121 words) — was checked: 26 such blocks appear in the draft (1,790 words, 2.1% of the text) against 3 in SP-2023, and their removal does not materially alter the result.

large (|z| > 2, or ≥2× vs SP-2023)notable (|z| > 1, or ≥1.4×)within range · bars are log-scaled and clipped · hover a row for a plain-English explanation
Z magnitudes against only eight prior plans are inflated by their small spread — read the direction, and use the SP-2023 view for the size of the change.

Seventeen years of language, one discontinuity

The style fingerprint above provides a static comparison; these charts render the same measures as time series across all eight earlier strategic plans (2009–2023). The gold star marks the 2026 draft, and the dashed segment indicates the magnitude of the jump from SP-2023. The default view groups complementary or contrasting measures side by side; the dropdown selects any individual series.

The draft names autism less than half as often as any plan before it

The stylistic shift documented above would be of limited consequence if the draft continued to address the same subject in comparable proportions. It does not. Which label a document prefers — "people with autism," "autistic people," "on the spectrum" — varies with the field; person-first usage has been declining across autism writing for a decade, reflecting deliberate community preference. Counting all naming conventions together removes that variable: every explicit reference to autism or to autistic people, tallied in one non-overlapping pass regardless of wording. That total occupied a band of 199–314 per 10,000 words for seventeen years (mean 268, sd 33). The draft registers 108 — 4.8 standard deviations below the series mean, 2.6× below SP-2023, and 1.8× below the lowest earlier plan (SP-2013, 199 per 10k). Both components decline: references to autistic people fall from 113 to 28 per 10k, and autism-as-a-topic mentions from 164 to 81. The apparent rise in identity-first "share" is an artefact of the collapsing denominator.

The space vacated by autism references is occupied by federal machinery. Every plan from 2009 to 2023 named autism between 8× and 36× more often than it named a federal agency (agency acronyms 8–32 per 10k). The draft contains 167 agency acronyms per 10k and names agencies 1.5× more often than it names autism at all.

One newly introduced term accounts for a substantial share of the draft's remaining person-level references: "profound autism" appears approximately 50 times — a rate of 5.5 per 10,000 words — in a record where the term had never previously appeared. The draft proposes its adoption as a "standardized functional designation … for research and policy purposes" and couples it to a prevalence claim: that the share of children meeting the designation "has risen alongside the broader trend."

references to autism or autistic people, per 10k wordsfederal agency acronyms, per 10k words · same cleaned text, running headers and reference lists removed · full counts in autism_reference_by_plan.csv, regenerated by the notebook

Not only less visible: less agential. A dependency-parse proxy for the semantic proto-role framework of Hoefer & Martin (SCiL 2026) classifies each entity mention as agent (grammatical subject of an active-voice verb) or patient (direct object or passive subject), following Dowty’s (1991) proto-agent properties. Three measures result. Visibility is the existing reference count: agencies appear at 166.9 per 10k in the draft versus 108.2 for autistic people — the only plan in the series where agencies are named more often. Relative agency asks: of all actor-role mentions in the document, what share belongs to each class? In SP-2023, autistic people held 57.8% of actor mentions and agencies held 42.2%. The draft reverses this: agencies hold 80.2%, autistic people 19.8%. Proto-role agency share asks: when each class does appear in a clause, how often is it the actor rather than the acted-upon? In the draft, agencies occupy the agent role 82.2% of the time — the highest in the series — while autistic people score 57.4%, mid-range but on a far smaller base. The combined effect: the draft’s autistic people are both less visible and less likely to be the ones acting; federal agencies are both more visible and more agential than in any earlier plan.

Widening the non-agency side does not restore the balance. The seventh pair is not agencies versus nonprofits alone. It is federal agencies versus one combined non-agency side: autistic people and the less-profit-motivated organisations the plans themselves name — nonprofits, advocacy groups, community partners and members, families, caregivers, parents, and public stakeholders. Same visibility and actor-role measures as the agency-versus-person pair above. The term list was harvested from the plan texts (full inventory in community_partner_terms.csv); federal “supporting partners,” community-as-setting, and industry partnerships are excluded. Combined visibility sat at a 309 per 10k mean across the 2009–2019 plans and 335 in SP-2023. The draft is at 120 — below agency visibility (167), the only plan in which agencies out-name that entire coalition. Of actor-role mentions, the combined side held 67.8% in SP-2023 and a 50.4% earlier-plan mean; the draft gives it 24.9% and agencies 75.1%. Existing agency-versus-person figures (166.9 vs 108.2; 80.2% vs 19.8%) are unchanged.

Fifty-five of eighty trends reversed — in a corpus where reversing half is normal

The subject-reference decline documented above is one finding. This section poses the broader question: across every content word with a strong monotonic trend in the 2007–2023 record (|Spearman ρ| ≥ 0.5), how many does the draft continue and how many does it reverse? The control is essential: it applies the draft's own scoring rule to every earlier document against its own prior record, with trends refitted on the documents preceding each one and judged against its own trailing eight-year window. On that like-for-like basis the 28 earlier documents reverse 25–76% of their own trend sets (median 48%). A reversal share near half is the corpus norm, and the measure imposes no minimum magnitude of change. The draft's 69% ranks second of 29 documents and is the highest among the strategic plans — SP-2013 25%, SP-2019 42%, SP-2023 48%, SP-2017 61% — though the stricter pairwise test below yields a more qualified conclusion.

The pairwise check, requiring no window or baseline choice. Comparing each plan only with the plan immediately preceding it, on one fixed vocabulary of 386 terms: the draft changes 62.2% by a factor of two or more — the largest step in the series, against 50.5% for 2019-vs-2017 and 13–35% across the annual-update era — with 45.1% of terms at least halved and the median term at 0.60× its SP-2023 rate. This constitutes the rupture finding in its strongest form, as it requires no trend fit and no baseline. On the strict pairwise reversal rule, however — plan-series trend, judged against the predecessor's own level, with a 2× magnitude floor — the draft reverses 32.4% while SP-2023 reverses 42.8%: SP-2023 was an expansionary rewrite whose increases reversed falling trends, whereas the draft predominantly pushes already-falling terms further down. The draft is the most divergent plan-to-plan step in the record, but not the most trend-reversing one. Full table in consecutive_plan_change.csv.

red = the 2026 draft · each document scored by the draft's rule against its own past · per-term detail in reversal_control_by_term.csv and trend_reversal_walkforward.csv

The substantial reversals view applies a floor: the term was used at least 5 times per 10,000 words in the 2018+ documents and the draft at least halves it. Thirteen terms clear that threshold, including health, disparities, adulthood, racial, suicide, sex, and justice. The remaining forty-two are individually unreliable (61 of the 80 terms are reversed by 8 or more documents in the control); the thirteen substantial reversals are the appropriate figures to cite.

What the draft says more and what it stopped saying

The trend reversals above identify which direction terms moved; keyness quantifies how statistically distinctive each word is in the draft relative to the earlier plans. The measure is weighted log-odds with an informative Dirichlet prior (Monroe et al. 2008). Positive z indicates the draft over-uses a term; negative z indicates under-use. All 3,579 measured words and 4,939 bigrams are searchable and sortable below. Rare newly introduced terms (leucovorin, NAPTI, dashboard) carry modest z-scores by design of the estimator; they appear under coined & retired in the following section.

Coined here. Retired here.

Keyness measures relative emphasis; vocabulary turnover measures absolute novelty. Coined: words used 5 or more times in the draft that never appeared in any earlier IACC publication. Retired: words used in at least 3 earlier strategic plans that the draft never uses. Dashed-border chips denote names of people or organisations. Among the coined terms, several carry substantial policy weight: leucovorin (24 occurrences in a record where the term had never appeared), NAPTI (a proposed National Autism Precision Therapeutics Initiative), biomarker-defined, and implementation-ready. Among the retired: evidence-based, scientists, advocates, vaccine, and collaborative.

Almost nothing carried forward

The vocabulary turnover above identifies which terms entered and exited the lexicon; this section measures a distinct property — the extent to which the draft's actual text reappears verbatim from the earlier record. The metric is the share of 8-word sequences already present in earlier IACC publications. The draft inherits 0.03% from the full record and 0.02% from SP-2023. The 24% average for earlier plans is carried by the annual-update era — SP-2010 inherited 53%, SP-2013 40% — when each plan was a revision of the last. The appropriate comparison is the recent one: SP-2023 took 0.94% of its text from SP-2019, and SP-2017 took 1.1% from SP-2013; modern plans were already near-clean-sheet documents. The draft falls roughly 50× below that recent norm, not 800× below the historical one. The committee's process — a full reconstitution with no institutional handoff (see background) — accounts for part of this gap. The draft's own preface acknowledges as much:

"Our drafting process did not have the benefit of a complete institutional record or formal handoff from the prior planning cycle."

inherited from the immediately preceding planinherited from any earlier publication

Thirty-seven sections, all of it new text

The preceding measures treat the draft as a single document. This table disaggregates it into its 37 outline sections, each profiled identically: readability (FK grade), prescriptiveness ("should" per 1,000 words), share of text found nowhere in the earlier record, and nearest earlier document by content. Click a column header to sort.

What was removed, what was added back, and what remains lost

The linguistic changes documented above did not occur in a vacuum. They coincide with the largest disruption to U.S. federal science funding in the modern era. Three questions situate the IACC draft within that broader context: how much science funding was removed, how much was directed specifically to autism research, and what the difference represents.

1. Science funding removed. Between February and June 2025, the NIH obligated approximately $8 billion less in grant awards than it had in the same period of FY 2024 — 62% of the prior year’s level — and terminated over 1,800 grants, in what the Government Accountability Office determined violated the Impoundment Control Act of 1974 (GAO decision B-337203, 5 August 2025). A JAMA analysis of the first wave (February–April 2025) documented 694 terminated NIH grants with a cumulative award value of $1.81 billion, of which $544 million had not yet been expended. At the National Science Foundation, over 1,300 grants totalling approximately $700 million in unspent funds were terminated or frozen in 2025 (Science News, citing Grant Witness data). In October 2026, the New York Times reported that NSF had carried over an additional $1.4 billion — approximately one-fifth of its primary research budget — that Congress had appropriated for research, redirecting it toward administration priorities including AI and quantum computing. By Nature’s January 2026 accounting, more than 7,800 NIH and NSF grants had been terminated or frozen in total (5,844 NIH; 1,996 NSF), with roughly 2,600 not reinstated, representing $1.4 billion in unspent funding still cancelled. The Brennan Center for Justice places the combined NIH and NSF total at over $3 billion frozen or terminated in 2025, with approximately $1.4 billion remaining cancelled or frozen as of early 2026. These figures measure money actually withheld or terminated from congressionally appropriated funds, not merely proposed budget reductions. The administration’s proposed FY 2026 budget sought far larger cuts — 41% at NIH, 57% at NSF, 24% at NASA — but Congress largely rejected those proposals in the enacted FY 2026 appropriations.

2. Autism funding added back. The Autism CARES Act of 2024 (P.L. 118-180) authorised $28.1 million per year for CDC autism programmes and $56.344 million per year for HRSA autism research and training programmes for FY 2025–2029. The FY 2026 appropriations bill funded both lines at their authorised levels and restored $8 million in Department of Defense autism research (Autism Speaks summary). NIH received a total programme level of $47.493 billion for FY 2026 (CRS report R43341), a $458 million (1.0%) increase over FY 2025; the enacted bill directed NIH to “maintain and enhance investments in [autism] research” and to implement Autism CARES Act requirements, but did not specify a dollar amount for the autism-research increase within the NIH total. Autism Speaks characterised the combined federal autism allocation as “nearly $400 million.”

3. The net. The two figures are not directly comparable: the science-funding removals above represent money withheld or terminated from existing appropriations across NIH and NSF during 2025, while the autism-specific figures are enacted annual authorisations for FY 2025–2029. With that qualification, the scale is clear. The GAO-documented NIH obligation shortfall alone — $8 billion in five months — exceeds twenty years of combined CDC and HRSA autism authorisations at the Autism CARES Act’s new rates. The approximately $3 billion in NIH and NSF grants frozen or terminated in 2025 is roughly eight times the “nearly $400 million” in annual federal autism funding. Congress rejected the administration’s proposed structural cuts and provided a modest NIH increase for FY 2026, but the grants already terminated or frozen during 2025 represent research capacity — laboratories, personnel, multi-year protocols — that a subsequent appropriation does not automatically restore.

Selected reporting. The following articles substantiate the figures above; each is linked to its original source.