Grant brief · current priority

Fund the independent validation study

Palimpsest is an MIT-licensed public evidence workbench for censorship measurements and reproducible AI evaluations. Its current validation study is pre-registered and published. Funding is needed for the two independent human coding passes that remain.

Fund independent validation Inspect the frozen study

Problem

A machine label is not human validation

Palimpsest's published lexical classifier labels AI responses, but no inter-rater agreement figure exists yet. Without independent human coding, the project correctly describes those labels as unvalidated.

Work already complete

The study is ready to code

145-row frozen sample
Published codebook and protocol
Committed answer-key digest
Agreement and verification scripts

Funded deliverable

1 · Code independentlyTwo coders apply the published codebook and attest that they did not inspect the classifier or answer key.

2 · Calculate openlyThe published script calculates agreement against the pre-registered thresholds.

3 · Publish the outcomeCoded sheets, agreement result and limitations are published, including an unfavourable result.

Falsifier

The grant does not buy success

The frozen protocol states that Krippendorff's alpha below 0.667 rejects the labelling scheme. That outcome remains publishable evidence, not a failed deliverable. Funders receive no control over questions, methods, findings or publication.

Budget and contact

$1,800 planned target

This forward honoraria budget covers two independent Mandarin-speaking coders. It is not an invoice, amount spent or funds received. Request a current breakdown through the project's public issue form; do not post confidential documents there. The aggregate ledger labels missing periods rather than reporting them as zero.

Evidence: palimpsest.info · Source and study: github.com/beepboop2025/palimpsest · No paywall · No supporter-only evidence · Watch the censor, never the censored.