# SpeechLP aggregate data dictionary

Source: Children’s Articulation Practice in SpeechLP Across Two Months, manuscript dated November 3, 2026. Definitions below come from its Methods and Supplementary Methods; the full manuscript controls interpretation.

## Observation and outcome

- **Profile:** family account plus normalized child-profile label, not an independently verified unique child. The primary cohort has 101 profiles from 101 accounts.
- **Source window:** January 22, 2026 inclusive through October 8, 2026 exclusive, America/Toronto time (last included date October 7).
- **Month 1 / Month 2:** elapsed [0,30) and [30,60) days after the first observed eligible card. Each day is 24 hours. These are not calendar months or a pre-practice baseline.
- **Accepted-card accuracy (%):** 100 × accepted scored cards / eligible scored cards. A card can be accepted after repeated utterances and auditory modeling. Individual utterances and replay counts were not measured. This is not first-attempt, spontaneous-naming or SLP-rated accuracy.
- **Matched profile-and-sound pair:** identical sound label observed in the same profile in both periods. There are 660 correlated pairs, not 660 independent children.
- **Primary mean:** sounds weighted equally within each profile, then profiles weighted equally. Pair-weighted and sound-specific summaries use their stated alternative denominators.
- **Change (pp):** Month 2 minus Month 1 in percentage points. Exact positive/negative/zero differences define Higher/Lower/Same; no clinically meaningful threshold is imposed.
- **95% CI:** nominal percentile interval from 10,000 whole-profile bootstrap resamples, seed 20261014. Exploratory and uncorrected for multiple comparisons. Supplement S2 describes the carried-over 10+ session interval.
- **Sessions:** recorded session identifiers within profile, split after gaps exceeding 30 minutes; assigned by session start. Cards are assigned by card timestamp.
- **Practice days:** distinct device-local dates with eligible cards. Cohort totals sum profile-specific dates.
- **Recorded age:** from the first eligible card, not independently verified.

## Files and fields

| File | Structure and interpretation |
| --- | --- |
| `aggregate-tables.csv` | Long format. `Table` is original table number 1–15; `Caption` is its full label; `Row` identifies the row; `Column` names a column; `Value` is the displayed value, including units/interval strings and suppression labels. |
| `aggregate-tables.json` | `tables[]` contains `number`, `caption`, ordered `columns` and ordered `rows`. `primary_profiles` is the cohort count. Other top-level fields identify title, analysis/manuscript/version and rights. |
| `sound-results.csv` | One of 24 sound labels per row. `Matched profiles` is the sound-specific paired denominator; `Compared cards` counts cards used; monthly means are percentages; `Mean change (pp)` and `CI low/high (pp)` are percentage points; `Higher`, `Lower`, `Same` count pairs by exact direction; `Suppressed` signals withheld result cells. |
| `study-results.json` | `primary_accuracy` holds primary and secondary accuracy summaries; `primary_sound_rows` holds sound results; `practice`, `age_bands`, `first_month_frequency`, and `day_one_sensitivity` hold their labeled aggregate analyses. `month_definitions` and `sound_suppression` explain windows and withholding. Author, conflict, version and rights fields give context. Retained precision can exceed rounded manuscript values. |

Main Tables 1–7 retain their numbering. Original export Tables 8–15 correspond to Supplementary Tables S1–S8. The supplement crosswalk explains figure numbering.

## Missing and suppressed values

`Suppressed`, `Withheld`, a true `Suppressed` flag, and empty/null withheld result fields must not be interpreted as zero. Sound results based on fewer than five matched profiles are suppressed; additional cells are withheld to prevent reconstruction. Recorded-age cells also use complementary suppression. Preserve this withholding and do not infer hidden cells from totals.

## Interpretation and rights

Primary result: 69.4% to 74.9%, +5.5 pp (95% CI 0.4 to 10.4). Recorded sessions and practice days declined. Scoring changed over time, with no app-version adjustment and an unquantified contribution to the difference. Automated acceptance has not been validated against SLP ratings. These preliminary descriptive findings do not establish clinical articulation improvement, treatment efficacy or a causal dose response.

CC BY 4.0 applies only to `aggregate-tables.csv`, `aggregate-tables.json`, `sound-results.csv`, and `study-results.json`. Manuscript, figures, software and source-record rights are reserved. No participant-level data or source records are supplied.