The Lyceum Labs research programme
Rigorous analysis of the public data the SEND system leaves on the table.
England's education departments publish remarkable statistics: every local authority's assessment requests, refusals, mediations and tribunal appeals; absence and exclusion rates split by SEND status; plan timeliness; home-education counts. Most of it is released, tabulated — and never seriously analysed.
We analyse it the way academic research is supposed to work: hypotheses preregistered before estimation, every number script-generated from checksummed snapshots, null results reported next to significant ones. Our aim is to make the system's own data legible to the families inside it — and harder for the system to ignore.
Working papers
The SEND Friction Index: Measuring the Administrative Burden on Parents
Six official measures of the administrative fight — missed deadlines, refusals, challenges, complaints, overcrowded special schools — combined into one score for every local authority. Friction is persistent (ρ = 0.72 year-on-year), barely related to deprivation, and where it is highest, ~11% more families per SD exit to home education the next year. The headline validation test failed, and we say so.
Refused, Not Appealed: The SEND Appeals That Never Happen
Tribunals find for families in ~99% of decided SEND appeals, yet 88.6% of the 71,836 refusals to assess in 2023–24 were never contested there. We show the missing appeals are patterned by deprivation — and that the gradient is specific to the costly channel, not to families' willingness to challenge.
Local authority explorer
The panel behind the papers, opened up: refusals, dispute rates and context for every English local authority.
Ethics & integrity statement
Public data only
We analyse published official statistics at local-authority level. No personal data, no scraping behind logins, no children identifiable anywhere in the pipeline.
Preregistration before estimation
Hypotheses, exact model specifications, exclusions and predicted effect sizes are locked in a public document before confirmatory analysis begins. Deviations get an appendix, not a quiet edit.
Reproducible by anyone
Raw snapshots are pinned by SHA-256 checksum. The full pipeline — download, panel construction, models, figures, these web pages' data — rebuilds with one command.
AI-assisted, human-directed
Analyses are executed by an LLM research agent under human direction, and we say so. The discipline that makes this trustworthy is the same one that protects against human bias: preregistration, pinned data, open code.
An explicit point of view
We build tools for SEND families and we are not neutral about children getting the support the law entitles them to. The methods are designed so that our motivation cannot bend our findings.
Corrections welcome
If you find an error in our data handling or analysis, we want to know, and we will publish the correction.
Current data build · panel e6835de90531… · manifest 80236d0f112a… · generated 2026-06-11