# The 2026 IACC Strategic Plan working draft, measured against the IACC publication record

Source document: `IACC Strategic Plan Working Draft July 17.pdf` - working draft, marked subject to errors and revision. Figures are quoted against SP-2023, the plan it replaces, with the mean of the 8 earlier plans second.

Corpus: 46 documents in the analysis set (8 earlier strategic plans, 37 other publications, 1 draft); 15 inventoried documents excluded as too short or easy-read/at-a-glance editions.

## Size and shape
- Draft: 81,788 words on 271 pages of prose (65 reference pages removed, trailing reference list from p.272); earlier plans averaged 32,326 words (max 60,494, SP-2023).
- Sentences: 19.2 words per sentence vs 28.4 in SP-2023 (21.2 across earlier plans) - the draft's sentences are the shorter ones.
- Word length: 6.62 letters and 2.21 syllables per word vs 5.90 and 1.95 in SP-2023; 49.4% of words run 7+ letters vs 40.6%, and 35.4% are polysyllabic vs 26.4% - the highest in the record on both.
- Readability, all six formulas, with the draft's rank among the 46 analysed documents (1 = hardest): flesch reading ease 0.6 (rank 1); flesch kincaid grade 17.9 (rank 3); gunning fog 21.8 (rank 2); smog 18.0 (rank 3); coleman liau 21.6 (rank 1); ari 19.3 (rank 3). The draft is the most extreme document on Flesch Reading Ease and Coleman-Liau, which weight word length most heavily, but ranks behind SP-2023 on the four grade-level formulas because its sentences are shorter. The defensible claim is the word-length one, not that it is the least readable document by every measure; Flesch Reading Ease below about 10 is an extrapolation past the calibrated floor of that scale.
- Lexical diversity (MATTR-500): 0.551 vs 0.486 in SP-2023 (0.490 across earlier plans).

## Voice
- Deontic index (share of must/shall/should among modals): 0.66 vs 0.19 in SP-2023 (0.21 across earlier plans).
- 'should': 10.5 per 1,000 words vs 1.4 in SP-2023 (7.8x) and 1.1 across earlier plans (9.8x). The study's most extreme measurement, and the one least sensitive to the choice of comparator. 'will' 0.2 vs 3.0.
- Passive constructions: 18.4 per 100 sentences vs 25.1. Hedges 4.8/1k vs 7.7; boosters 2.6/1k vs 2.2.
- First-person plural 0.09/1k vs 1.19; acronyms 28.9/1k vs 27.8.

## Features on which the draft is unlike any earlier strategic plan (|z| > 2)
- should_per_1k: draft 10.53, earlier plans 1.07 +/- 0.30 (z = +31.1)
- deontic_index: draft 0.66, earlier plans 0.21 +/- 0.05 (z = +8.9)
- polysyllable_pct: draft 35.43, earlier plans 24.02 +/- 1.48 (z = +7.7)
- syllables_per_word: draft 2.21, earlier plans 1.86 +/- 0.05 (z = +6.8)
- letters_per_word: draft 6.62, earlier plans 5.64 +/- 0.17 (z = +5.9)
- mattr: draft 0.55, earlier plans 0.49 +/- 0.01 (z = +5.8)
- nominalisation_pct: draft 11.77, earlier plans 7.55 +/- 0.77 (z = +5.5)
- long_words_pct: draft 49.41, earlier plans 37.85 +/- 2.15 (z = +5.4)
- coleman_liau: draft 21.59, earlier plans 15.91 +/- 1.13 (z = +5.0)
- flesch_reading_ease: draft 0.57, earlier plans 27.92 +/- 8.21 (z = -3.3)
- will_per_1k: draft 0.23, earlier plans 2.76 +/- 0.84 (z = -3.0)

## Naming autism (per 10,000 words; draft vs SP-2023 vs mean of earlier plans)
- person-first ('people with autism'): 2.6 vs 19.2 vs 54.8
- identity-first ('autistic people'): 22.0 vs 58.7 vs 9.2
- 'on the (autism) spectrum': 0.0 vs 32.9 vs 9.9
- 'autistic' (any use): 24.8 vs 72.7 vs 12.6
- ASD (cs): 1.2 vs 21.2 vs 156.9
- 'autism spectrum disorder(s)': 0.1 vs 1.0 vs 5.2
- 'disorder(s)': 4.8 vs 9.3 vs 21.3
- 'condition(s)': 26.9 vs 24.5 vs 15.6
- 'profound autism': 5.5 vs 0.0 vs 0.0
- nonspeaking / minimally verbal: 3.5 vs 1.2 vs 2.2
- neurodivers*: 0.0 vs 0.7 vs 0.2
- identity-first share of naming phrases in the draft: 90% (SP-2023: 75%) - but this is a share of a shrinking denominator: identity-first phrasing is 22.0 per 10k in the draft against 58.7 in SP-2023. Person-first usage has also been falling across autism writing for a decade by community preference, so no single naming term separates this draft's choices from its era's. See the next section for the measure that does.

## How often the draft names its own subject
- Every explicit reference to autism or autistic people, counted once each: 108 per 10,000 words, against 278 in SP-2023 and 268 +/- 33 across the 8 earlier plans, whose range is 199-314 (z = -4.8).
- Both components fall: references to autistic people 28 vs 113 per 10k, and autism-as-a-topic mentions 81 vs 164. The finding does not depend on which naming convention is in fashion, nor on the choice of baseline.
- Federal agency acronyms 167 per 10k vs 12: the draft names agencies 1.54 times per autism reference, where every plan from 2009 to 2023 sat between 0.03 and 0.12.

## Largest lexicon shifts vs earlier plans (per 10,000 words)
- Naming autism | ASD (cs): 1.2 vs 156.9
- Governance and money | agency acronyms HHS/NIH/CDC/FDA/CMS/... (cs): 166.9 vs 18.6
- Voice and modality | 'should' (cs): 105.3 vs 10.7
- Governance and money | federal: 71.2 vs 10.0
- Naming autism | person-first ('people with autism'): 2.6 vs 54.8
- What is to be done | implementation: 39.0 vs 4.1
- What is to be done | service(s): 28.2 vs 62.1
- Governance and money | IACC (cs): 4.5 vs 38.0
- What is to be done | support(s): 61.7 vs 33.2
- What is to be done | intervention(s): 20.9 vs 46.9
- Voice and modality | 'will' (cs): 2.3 vs 27.6
- Life course | safety / wandering / elopement: 29.8 vs 7.6
- Causes and biology | immun*: 27.8 vs 5.9
- Causes and biology | regression: 21.6 vs 1.4
- What is to be done | biomarker(s): 24.8 vs 5.6

## Keyness (weighted log-odds vs strategic plans 2009-2023)
- Most over-used words: federal, agencies, sleep, implementation, cms, domain, evidence, clinical, pathways, evaluation, nih, safety, hrsa, priority, regression, action, measures, subgroups, motor, medication, metabolic, defined, guidance, standards, functional.
- Most under-used words: asd, research, services, studies, iacc, children, autism, interventions, question, community, people, years, needs, early, factors, spectrum, brain, objectives, adults, risk, disorder, needed, development, recommended, efforts.
- Most over-used bigrams: life course, lead agencies, federal autism, national autism, medication effects, clinically meaningful, functional outcomes, standards care, data elements, sleep disruption, strategic objectives, caregivers burden, immune activation, trials design, federal implementation.
- 176 words (5+ uses) never before seen in an IACC publication, e.g. dashboard, workplans, compact, comparative-effectiveness, napti, biomarker-defined, implementation-ready, praxis, tier, speech-motor, prescribing, neural-circuit, evidentiary, mental-health, motor-access, leucovorin, ndr, deterioration, post-meal, civil-rights.
- 54 words used in 3+ earlier plans that the draft never uses (names excluded), e.g. sharing, enhance, scientists, evidence-based, repository, mutations, infants, various, aspects, prevented, ndar, happening, children's, regions, suggested, stem, collaborative, showed, advocates, explore.

## Similarity
- Nearest earlier document by content: SP-2023 (cosine 0.29); nearest strategic plan: SP-2023 (cosine 0.29). For comparison, consecutive earlier plans had cosine 0.41 on average.
- Nearest earlier document by style (Burrows' Delta): SP-2011 (Delta 1.13).

## Text reuse
- Effectively none of the draft is inherited text: 0.03% of its 8-word shingles occur anywhere in the earlier record (68 shared shingles in total, the most - 14 - with SP-2023; the longest shared passage is 16 words).
- The draft inherits 0.02% from SP-2023. The like-for-like comparison is the recent one, not the 24% all-plans average: that average is carried by the annual-update era, when each plan was a revision of the last (SP-2010 53%, SP-2011 31%, SP-2012 3%). The modern plans were already near-clean-sheet documents: SP-2017 took 1.12% from its predecessor and 4.4% from the record; SP-2019 took 5.89% from its predecessor and 19.4% from the record; SP-2023 took 0.94% from its predecessor and 14.7% from the record. Against that recent norm the draft is roughly 54x lower, not the ~812x the historical average implies.

## Trends across 2007-2023 and the draft
- Of the 80 terms with a strong monotonic trend (|rho| >= 0.5), the draft reverses 55 (68.8% of its trend set). Control: every earlier document scored the same way - trends refitted on the documents before it, judged against its own trailing 8-year window - reverses 25.0-75.9% (median 48.2%), so a reversal share near half is the corpus norm and the measure has no magnitude floor. The draft is nonetheless at the top of that distribution: rank 2 of 29 documents, and the highest of the strategic plans (SP-2013 25.0%, SP-2017 60.6%, SP-2019 42.1%, SP-2023 48.2%, draft 68.8%).
- Plan against plan, on one fixed vocabulary of 386 terms: the draft changes 62.2% of it by 2x or more against SP-2023 (45.1% at least halved), the largest step in the series - the seven earlier steps run 13.0-50.5% - and the median term ends at 0.6x its SP-2023 rate, so the change is near-uniform contraction rather than substitution. On the stricter pairwise reversal rule (trend fitted on the plan series only, 2x magnitude floor) the draft reverses 32.4% against SP-2023's 42.8%: it is the most divergent plan-to-plan step in the record, but not the most trend-reversing one.
- 13 reversals clear a magnitude floor (used 5+ per 10k in the 2018-on documents and at least halved by the draft): health (111.1 -> 47.0), disparities (19.8 -> 1.3), living (14.5 -> 5.4), adulthood (11.9 -> 2.2), experience (11.5 -> 4.0), receive (11.4 -> 4.0), physical (10.8 -> 2.7), racial (9.5 -> 0.7), improving (8.2 -> 2.4), meet (7.9 -> 2.7), suicide (6.6 -> 2.2), sex (5.4 -> 1.6), justice (5.1 -> 0.0). Report these rather than the full 55.
- Rising terms the draft continues: care, regulation, accountability, outcomes, access, timely, accommodations, participation, conditions, dysregulation, employers, primary, transportation, consistent, barriers.
- Rising terms the draft reverses: college, veterans, accessibility, suicide, prefer, equity, arts, ensuring, demographic, leadership, inpatient, justice, health, living, sex.
- Falling terms the draft continues to drop: developing, prevented, clues, caused, concerned, happening.
- Falling terms the draft revives: syndrome, treatment, abnormalities, turn, happen, hold.

## Inside the draft
- 37 outline sections of 300+ words; novel text share ranges 100-100% (median 100%).
- Most prescriptive sections ('should' per 1,000 words): Priority Therapeutic Domain 7: Autonomic Dysfunction (18.5); Priority Therapeutic Domain 9: Epilepsy and Network Excitability (17.9); Priority Therapeutic Domain 1: Neurotransmission, Neural Circuit Funct (17.2); Priority Therapeutic Domain 8: Sleep and Circadian Regulation (16.3); Life Course Domain VI: Caregiver Stabilization, Respite, Navigation, a (16.2).
- Densest prose (Flesch-Kincaid grade): Appendix C (30.6); Contributing Factors and Causality (22.2); Priority Therapeutic Domain 6: Mitochondrial, Redox, Metabolic Functio (20.6); Implications for Federal Action (20.6); Life Course Domain III: Improving Communication and Access for Nonspea (20.3).
- Nearest earlier documents across sections: RTC-2017 (11), SP-2017 (11), SP-2023 (7), SP-2012 (3), PUBA-2012 (3).

## Files
- 01_corpus_timeline.png
- 04_identity_first_share.png
- 04_lexicon_strategic_plans.png
- 04_subject_reference.png
- 05_draft_style_zscores.png
- 06_keyness_draft_vs_prior_plans.png
- 07_content_map_mds.png
- 07_similarity_strategic_plans.png
- 07_style_dendrogram.png
- 08_inherited_text_by_plan.png
- 09_consecutive_plan_change.png
- 09_lexicon_trends.png
- 09_reversal_control.png
- 09_strategic_plan_drift.png
- 09_style_trends.png
- 10_draft_section_novelty.png
- autism_reference_by_plan.csv
- consecutive_plan_change.csv
- content_similarity.csv
- dashboard_review_checks.csv
- documents.csv
- draft_sections.csv
- draft_zscores.csv
- dropped_vocabulary.csv
- keyness_bigrams_vs_prior_plans.csv
- keyness_words_vs_SP-2023.csv
- keyness_words_vs_prior_plans.csv
- lexicon_per10k.csv
- list_as_prose_blocks.csv
- new_vocabulary.csv
- pairwise_reversal_detail.csv
- reuse_carryover_by_plan.csv
- reuse_draft_vs_each_doc.csv
- reversal_control_by_term.csv
- strategic_plan_drift.csv
- style_delta.csv
- style_features.csv
- substantial_reversals.csv
- term_trends.csv
- trend_reversal_control.csv
- trend_reversal_walkforward.csv

## Method notes
- Sentence splitting, syllable counting and passive detection are heuristic and applied identically to every document; reference lists, running headers/footers and tables of contents were removed before measuring; easy-read/at-a-glance editions and documents under 2000 words were excluded from comparisons.
- Comparator: figures are quoted against SP-2023 first because that is the plan this draft replaces. The 8-plan mean overstates several changes - 'ASD' fell from 157 per 10k across the plan series to 21 in SP-2023 before the draft took it to 1.2 - and on identity-first naming the two comparators disagree in sign.
- Z-scores rest on 8 stylistically consistent plans, so their denominators are small and their magnitudes inflated: read direction from them and size from the pairwise comparison.
- What this cannot show: intent. A genre change, a jargon-dense register and near-zero inherited text are equally consistent with a committee working without a handoff under deadline, which the draft's preface states. The document is also a working draft released for public comment, measured against finished publications.