N Nimdex

Methodology & editorial posture

Nimdex is a data-journalism investigation into Ireland's planning system. This page explains where the data comes from, how we sample it, and the rules we follow about what we do and don't show.

Where the data comes from

The Nimdex corpus is assembled from three public sources:

For the editorial sample of objections behind the quiz, we extract submission text from PDF objection letters using OCR, then run a two-tier tagging pipeline. Tier 1 is rule-based: it scores submissions for length, emotional register, and theme coverage. Tier 2 is an LLM grader (Claude Haiku) that assigns a press grade (A, B, C), an Irish-register score, a specificity score, and a topic.

What "successful" means in the quiz

The MVP cards use a working definition: a submission is treated as "successful" if it appears to have stopped the development at first instance (council refusal) or on appeal to ABP. The strictest definition — that the project did not actually get built — requires extra verification per case, and is not in the MVP.

Per-card outcomes in the launch sample are reviewed against the public file. Where the outcome is genuinely ambiguous (settled, withdrawn, partial grant), the card is excluded from the deck.

Editorial posture: no submitter names

We never show submitter names on the game screens. Quotes are real, verbatim, from public planning files — but the legal name on the file does not appear on any card or reveal. This is a defamation-hygiene rule, not a data-availability one. The names are in the underlying record, but they don't belong on a card you might share to social media.

We also avoid drone-fear quotes from the Manna prominent dataset on the lead deck. Wind, telecoms-mast, and data-centre objections lead instead. This is an editorial choice: the goal is to surface the texture of Irish objection prose, not to single out one applicant.

What the gravity stats are

Numbers in the gravity panels (e.g. "79% of judicial reviews never reach a written verdict") are calculated from our courts.ie scrape and the application-level outcome data. They come with the report. If a stat is rounded or sampled, the report says so.

What we don't do

Who built this

Nimdex is a Prime Directive AI project, built in collaboration with Inference Advisory. Editorial direction by Matt Cortland.

Contact

Spotted something wrong? Have a case you think belongs on the deck? Email hello@primedirective.ai.

← Back to the quiz