← Back to the evidence ledger

Personal assessment audit

Current status, October 12, 2026: the site holds 25 assessments, 398 evidence records, and 121 linked sources. Seventeen records keep a documented disagreement between sources instead of picking one. The October 3 review below covered the first 19 assessments; the dated addenda record everything added since.

Addendum, October 10, 2026: a twentieth assessment, the SSI-4, was added with its own source-checked records. See ADDITIONS-FLUENCY.md. The October 3 findings below are unchanged.

Reviewed October 3, 2026, America/Chicago, by OpenAI's Codex, working as a research agent at Ethan's direction. Codex opened and read the sources for all 19 assessments then on the site; earlier sub-agent research was treated as leads only. This record supersedes the numerical conclusions in the October 2 addition memos and older research review where they differ.

The audit checked edition, age, quoted administration task, norms, reliability methods, and diagnostic-accuracy context against current publisher pages, technical graphics, available manual introductions, original studies, and named technical reviews. PDF extraction was supplemented by visual inspection of the consequential tables and graphics. The sources distinguish information reported by a publisher from findings in a particular research sample. A source check does not independently validate an assessment or mean that its entire examiner manual was read.

At that review the site held 19 assessments and 311 evidence records. Each number has a record with its source, method, score/task, population, available sample size, cutoff, location, date, and limitations. A missing denominator remains missing; the norm sample is not substituted for the reliability or diagnostic sample. The Research page exposes additional records beyond the profile's summary fields.

What changed

Assessment-by-assessment findings

The table and instrument notes below summarize the checked evidence. The site's metric records contain the detailed source locations and sample context.

Assessment Checked age Quoted time Norm/sample context
PPVT-5 2:6–90+ Source 10–15 min Source N=2,720; ages 2:6–90+ Source
CELF-5 5:0–21:11 Source 30–45 min for Core Language Score; full battery varies Source N=2,380; ages 5–21; 47 U.S. states Source
PLS-5 Birth–7:11 Source 45–60 min current overview; older brochure 30–60 min Source N=1,400; norm data collected 2010; March 2008 Census targets Source
GFTA-3 2:0–21:11 Source 5–15 min Sounds-in-Words; other sections vary Source N=1,500; ages 2–21; 39 U.S. states Source
KLPA-3 2:0–21:11 Source 10–30 min publisher analysis estimate Source N=1,500; ages 2–21; 39 states Source
CAAP-2 2:6–11:11 Source 15–20 min Articulation Inventory Source N=1,486 U.S. children; 2013 Census targets Source
TNL-2 4:0–15:11 · publisher age conflict Source 15–20 min Source SALT reference database: 778 children, ages 4:0–14:11 Source
OASES 7+ · S 7–12, T 13–17, A 18+ Source 15–20 min · plus about 5 min scoring Source Adult development n173; separate S/T/A evidence required Source
WAB-R 18–89 years Source 30–45 min Part 1 · supplemental +45–60 · bedside 15 Source Original WAB: 150 aphasia/59 controls; later 215/63; WAB-R adapted tasks n20 Source
CLQT+ 18:0–89:11 Source 15–30 min Source Criterion-based · original studies n13/92/154/119; aphasia path n76 Source
CADL-3 18–95+ years Source About 30 min Source n115 adults with neurogenic communication disorders · 2014–2016 Source
CASL-2 3–21 years · test-specific age coverage Source Each test 5–10 min; GLAI ages 3–6: 30–45; ages 7–21: 45–60 Source U.S. age-stratified norms · n2,394 Source
OWLS-II Oral Form A 3:0–21:11 · written and Form B 5:0–21:11 Source 10–30 min per scale · not the full battery Source Co-normed U.S. scales · age-stratified n2,123 Source
REEL-4 Birth through 36 months Source About 20 min Source U.S. infants and toddlers · n1,019; demographics aligned with 2019 child population Source
Bayley-III 1–42 months Source 30–90 min battery · language-only time unverified Source Review n1,700; publisher 9.8% clinical cases; collection dates conflict Source
Arizona-4 1:6–21:11 Source 5–20 min per articulation test Source N = 3,192 · U.S. ages 1:6–21:11 Source
KSPT 2:0–5:11 Source 5–15 min Source Original KSPT: n447 typical speech; n263 disordered speech Source
PLS-5-Spanish Birth–7:11 Source 20–45 min Spanish · 20–65 min dual-language Source n1,150 · monolingual/bilingual Spanish speakers · U.S. and Puerto Rico Source
CELF-4-Spanish 5:0–21:11 Source Not reported in checked current publisher sources Source n1,019 U.S./Puerto Rico Spanish-speaking students · reported by 2018 research Source

Interpretation and remaining gaps

PPVT-5. The publisher's .97 summary and the review's mean split-half .97 are separate summaries. The review reports .94–.98 split-half, mean retest .84 (n213), and mean alternate-form .86 (n273); publisher summaries give retest .88 and alternate-form .84. The infographic's .98/.98/.95 values are negative predictive power at 10% prevalence, not specificity. A matched sensitivity/specificity pair remains unavailable from those sources. Root also rechecked its three-correct basal, six-incorrect ceiling, and one permitted repetition against publisher materials.

CELF-5. Publisher norm total is 2,380. The graphic supplies scorer estimates .99 Word Structure, .95 Formulated Sentences, .91 Word Definitions, and .96 Structured Writing. Accuracy varies with cutoff: .97/.97 at SS80, 1.00/.91 at SS85. It is not one universal accuracy rate. Recalling Sentences reversal, four-zero stopping rule, and no-repetition instruction were rechecked in the existing local manual text, printed pp. 26 and 74; no manual or test items are redistributed.

PLS-5 English. Current publisher time is 45–60 minutes; the older brochure says 30–60. The review's normative split-half values are AC .91, EC .93, Total Language .95; these differ in aggregation from an age-specific .80–.97 range. Retest .86–.95 uses n195 and 3–28 days. Examiner agreement .96–.99 uses n54; 91.9–100% scoring agreement concerns 200 protocols. Accuracy .83/.80 at SS85 retains the publisher/review age-label disagreement. The basal/ceiling also agrees with Hsiao et al.'s original study procedures; Pearson's FAQ confirms use of the lowest basal and ceiling. The repetition exceptions reported in Sami's study remain study procedures rather than an assertion of a fully checked U.S. manual rule.

GFTA-3. Technical graphic alpha ranges .94–.97 across sex/task groups; the comparison flyer gives an overall .94–.96 range. Both remain visible. The flyer gives 13.2% bilingual participants with English primary, not bilingual norms for any language. Accuracy at SS85 is .91/.81 for Sounds-in-Words. The .70–1.00 scorer range in the comparison flyer is in the GFTA-2 column and is not assigned to GFTA-3.

KLPA-3. The norm total 1,500 and 39 states are visible in the technical graphic. Retest .94 in the flyer differs from .92/.93 and unusual task labels in the graphic. Scorer ranges .98–1.00 versus .96–1.00 also differ. The SS85 accuracy pair agrees at .93/.83; lower-cutoff rows disagree. These are recorded as publisher-source conflicts.

CAAP-2. Norm n1,486, age 2:6–11:11, and the 15–20-minute inventory time are supported. Training reports scorer agreement 1.00 for the Articulation Inventory and >.99 for the Phonological Process Checklist. Retest significance p<.01 supplies no coefficient. Concurrent correlations .789–.948 with older GFTA/KLPA editions are not sensitivity/specificity.

TNL-2. Current publisher text differs on the lower age bound (4 versus 5); the finder retains the disclosed conflict. SALT's 778-child reference sample drawn from standardization is not established as the entire norm sample. A 2022 original study reports manual developmental subtest alpha .81 comprehension/.87 production and cohort scoring agreement 90.65–93.84%. Those percentages are not a retest coefficient or diagnostic accuracy. Public total-score retest and matched diagnostic-accuracy estimates remain gaps.

OASES. S/T/A forms cover 7–12, 13–17, and 18+ and have 60/80/100 items. The 2006 adult-development section alphas .92–.97 use n173; adult retest r .90–.97 uses n14 over 10–14 days. Neither is pooled evidence for all current child and adult forms. Percentage agreement and correlations are kept distinct.

WAB-R. Added AQ retest Pearson r .973, LQ .940 (n10), CQ .915 (n8), two weeks, chronic stroke, from Bond's 2019 thesis Table 4. This small research sample is distinct from manual norms. Rao's AQ mode concordance .942 concerns 19 PPA participants contributing 23 pairs, not 23 unique people or same-mode retest. A 2026 acute-stroke study reports 77% sensitivity/64% specificity at AQ93.7 in n171; it does not establish a universal cutoff. Original-WAB norm counts are identified as historical evidence.

CLQT+. Criterion-based interpretation, two administration paths, and age/time are supported. Development-study sample counts are separate cohorts. Public publisher descriptions do not supply exact current internal-consistency, retest, scorer, or matched diagnostic-accuracy coefficients; those remaining gaps are real within the inspected sources. Historical CLQT figures are not transferred automatically.

CADL-3. Publisher alpha .94 and retest .94 are reported. Norm n115 is not assigned to retest because the public retest summary does not give its sample, interval, or statistic. The public summary does not supply a matched diagnostic-accuracy pair.

CASL-2. Manual introduction reports internal consistency .85–.99 tests/.95–.99 indexes; two-week retest .73–.94/.88–.96; scorer ICC .86–.97. Detailed technical-table denominators are unavailable. The excerpt was personally read through indexed text; its direct CDN PDF returned 404 on this audit date, now disclosed in the source access field. Remote-equivalence results are a separate mode-agreement record. The GLAI SS85 .74/.84 pair concerns the mixed clinical validation sample; the .92/.92 pair comes from a more restricted moderate/severe sample and is not substituted. GLAI takes 30–45 minutes at ages 3–6 and 45–60 at 7–21; individual tests take 5–10 minutes.

OWLS-II. Oral Form A covers 3–21:11, written scales 5–21:11; Form B starts at 5. Time is per scale. The UNSW review supplies clinical internal reliability ≥.95/≥.93 and scorer ICCs .93–.96 by scale/form. Its n117 range .73–.94 appears in a paragraph whose method labeling is ambiguous; the site explicitly avoids presenting it as an established retest coefficient. Alternate-form .67–.96 (median .78; n319) is separate. Matched sensitivity/specificity remains unavailable.

REEL-4. Norm n1,019 and birth–36 months are supported. Publisher average reliability >.90 has no named method or score-specific coefficient, so it is labeled a general summary. Exact retest and diagnostic-accuracy values remain unavailable. Accuracy values for a different screening test in a study using REEL are not transferred to REEL-4.

Bayley-III. Review gives norm n1,700 and Language Composite split-half .93. Retest .67–.94 spans Cognitive/Language/Motor across n197 and is not a language-only range. Publisher/review norm-collection dates disagree. The current publisher identifies the edition as discontinued; the 30–90-minute time covers the battery, not an isolated language task. Correlation with WPPSI-IV is convergent validity. Language-specific scorer and matched diagnostic-accuracy gaps remain.

Arizona-4. Norm n3,192, age 1:6–21:11, and 5–20 minutes per articulation task are supported. Prahl and Schuele's 2022 original article reports manual internal-consistency .90–.97 and retest .96. Retest sample/interval are not supplied there. The 50-child speech-sound-disorder clinical sample is distinct from the norm total and is not treated as a complete diagnostic-accuracy denominator.

KSPT. McCauley and Strand's original 2008 review reports norm groups 447 typically developing and 263 with disordered speech. Two subtests have reported retest ≥.90; exact coefficients, n, and interval are not reproduced. That does not establish accuracy for childhood apraxia of speech. The 2026 combined manual format is not evidence of new norms for the original 1995 test.

PLS-5 Spanish. Spanish norm n1,150 is independent of English n1,400. Review gives normative split-half AC .80–.94, EC .80–.95, Total .87–.97; retest n193 with Total ≥.90 in three age groups and AC/EC below .90. Scorer .99/.99 involved rescoring 200 protocols. The reported .85/.88 accuracy pair lacks an explicit PLS cutoff and complete validation denominator in the review. A ≤77 threshold selected a reference group using an earlier unidentified measure and is not a PLS diagnostic cutoff.

CELF-4 Spanish. U.S. norm n1,019. Review normative alpha .65–.97 and split-half .62–.98 differ from clinical-group estimates. Retest corrected r .52–.93 uses n132, 7–30 days (mean 14). Fourteen trained raters are not 14 children. The 2018 cohort's validation n293 is distinct from its n656 parent cohort; at SS85 specificity is 65%, sensitivity 93% in Results versus 94% in Abstract. The paper's quoted manual .96/.87 pair is separate. Exploratory SS78 .86/.80 is study-specific. Current checked publisher text does not provide a completion time; the English-edition time is not substituted.

Additional numerical check

Codex rechecked Petersen et al. (2020), including Table 3: 31 bilingual Spanish/English children, 21 typical/10 language-disordered, ages 5:9–9:7. Combined dynamic-assessment predictors yield 90% sensitivity/90.5% specificity with linear discriminant analysis and 100%/95.2% with logistic regression. These are small-sample research classification findings, not psychometrics for a commercial test. Original study.

Source access and scope

Publisher pages were checked for the latest available information about the specified editions; a recently updated sales page does not imply recently collected norms. Technical reviews identify which manual statistics they reproduce. Numerical gaps say what the inspected source does not supply, rather than asserting that the number cannot exist anywhere. The CASL indexed excerpt's unavailable direct URL is explicitly marked; the checked GFTA/KLPA comparison PDFs and UNSW OWLS review returned PDF content successfully.

No test items, answer keys, norm conversion tables, examiner manuals, patient data, or private course files are redistributed. Cached research downloads remain outside the preview's served routes. Faculty validation remains pending. This is a local, source-linked planning prototype; no public publication occurred.

Addendum · 10 October 2026 (hole audit)

A second pass on every figure still showing "—": 13 new sources and 12 new records across TNL-2, Bayley-III, CLQT+, REEL-4, PPVT-5, OWLS-II, KSPT, CADL-3, SSI-4, OASES, GFTA-3 and CELF-4 Spanish. Carried-forward figures (first-edition TNL accuracy; original-CLQT pilot reliability) are stored with status inherited and wear a visible flag on the spec sheet; the Bayley-III pair is a predictive study in a very preterm cohort and is flagged as such. Details, sources and the list of holes that remain genuinely unreported are in HOLE-AUDIT-2026-10-10.md. All values await LSU faculty validation.

Addendum · 10 October 2026 (Week 6 additions)

Five speech-sound instruments from the COMD 4756 Week 6 deck were added with the same record standard: BBTOP-2 (PRO-ED page: n770, Word Inventory α .97, retest .87/.88, sensitivity .81 / specificity .82 at cutoff 90), GDAP (retailer-hosted publisher text: n880, α .71–.95, retest .98–.99, sensitivity .77 / specificity .80 at SS 90; manual unread), DEMSS (Strand et al. 2013: 89–91 % agreement; criterion-referenced, no norms; the 65 %/97 % accuracy pair restated in paywalled secondary sources is described in the record but not displayed), ICS (McLeod et al. 2012, 2015: α .93–.94, n803 norms, sensitivity .82 / specificity .58 against parental concern, flagged) and CTOPP-2 (PRO-ED and PAR: n1,900, composite α ≥ .85; retest and accuracy not public). A "Phonological awareness" clinical-area chip was added for the CTOPP-2. Figures from retailer-hosted publisher text are labelled as such pending manual confirmation. Details in CANDIDATES-SSD-2026-10-10.md.

Addendum · 12 October 2026 (interface review)

Claude (Anthropic) reviewed the site ahead of its first faculty demo, working at Ethan's direction. It walked every flow on desktop and phone screen sizes, ran an automated accessibility scan (axe-core, WCAG 2 A/AA), and fixed what an expert reader would trip on: a CASL-2 specificity note that described sensitivity, "≥" symbols that had separated from their coefficients (KSPT, PLS-5 Spanish), sample sizes written as "n117" instead of "n = 117", a chart label that overlapped its scale caption, and audit links that opened raw text or a missing page. Claude also designed and built the tiger-to-SLP intro animation on the Atlas. This review changed how figures are displayed, not the figures: no source was re-verified, and every number still traces to the records above.