Self-audit · Episode 1 · July 1, 2026
We audit our own house first
On: The site's own 'legibility gap' thesis
Before this show grades anyone else, it grades the host. Here is our own 'legibility gap' thesis on the same rubric, with the receipts you can re-run.
Episode one points the Legibility Index at the site's own recurring argument, that complex systems fail whatever they cannot make legible. We steelman it, score it on three axes, and log two receipts you can check: one re-derived live in your browser, one guarded against the essay it came from. The point is to earn the right to score anyone else.
01Steelman first
The strongest version of our own claim: a modern state governs through registries, cutoffs, and scores, and a person the register cannot represent (a pre-digital deed holder, a child one point the wrong side of a threshold, a non-native writer flagged by a detector) is not merely unlucky but structurally locked out, because the gate is built to act only on what it can read. The failure is in the instrument, not the person.
02The clips
The state cannot assist what it cannot see; the recovery moves at the speed of the register, and the register is only partly written.
A roughly four-percent per-sentence false-positive rate does not stay small: across a whole paper it compounds into better-than-even odds of a wrongly flagged sentence.
Excerpts are quoted under commentary and linked to the source. This show never re-hosts audio it does not own.
03The score
Can any evidence move the claim, or is it built to survive every outcome?
Mostly falsifiable: the thesis makes concrete, checkable predictions (registry completeness, a cutoff that flips eligibility). It earns a point only where it reaches for 'systems' as a general mover without naming a test.
How far apart are the defensible claim retreated to and the strong claim sold?
A small gap between the defensible claim (specific gates demonstrably fail specific people) and the sold claim (legibility explains the outcome). Real, worth flagging, not egregious.
Does a clean number, 2x2, map, or framework stand in for messy reality?
The essays lean on clean single numbers (a registry percentage, a compounding percentage) to carry an argument about messy institutions. The numbers are sourced, but a tidy figure standing in for a tangled system is exactly the move we grade in others, so we grade it here.
The Index scores the claim and the rhetorical move, never the person. Every axis cites the exact evidence above.
04Roll the receipts
We told readers that a roughly four-percent per-sentence false-positive rate crosses even odds over a twenty-sentence paper. Re-derive it here: 1 - (1 - 0.04)^20.
Source: Computed: 1 - (1 - 0.04)^20We cite that AI detectors flagged 61.3% of TOEFL essays by non-native English writers as AI-generated, versus 3.2% of essays by US-born writers. This receipt fails the build if that figure ever drifts from the essay it is quoted in.
Source: Liang et al., 2023 (Patterns) ↗Computed receipts re-derive live in your browser and in CI; sourced receipts are cited and fail the build if the source drifts. We compute what they sourced.
05What would change our mind
If pre-digital, paper-deed owners outside the electronic register could in fact complete assisted filing at scale, the 'the gate cannot open' claim would fail and the false-legibility score would rise, because the clean 'registry completeness' number would have been standing in for a process that actually works around it.
Right of reply
Named subjects always get the last word here, verbatim. If this critique is about your work and you want to respond, write jake@jakelawrence.xyz and your reply runs at the top of this page.
Listen
More: all episodes · how this show checks itself · the research Claims Ledger