Docs

The rubric: eight factors and a verdict

The eight writing factors PostLens scores, how they combine into a 0–100 score, and the thresholds behind GO / WAIT / IMPROVE.

Every draft is read against eight factors. Each is scored 0–4 by the model and normalised to 0–100. The overall score is a weighted sum minus two penalties: one when the bait check fires, and one when the generic/formulaic check fires (a text-quality cue about templated writing, never a claim about who wrote it). All of that comes from one request carrying eleven questions: eight factor scores, one genre choice, and the two checks. The rubric version is stamped on every result so a saved row always knows which rubric produced it.

The eight factors

  • Opening hook — does the first line earn the second?
  • Clarity — is it easy to follow on one read?
  • Specificity — concrete detail over generality.
  • Reader value — is there something in it for the reader?
  • Evidence cues — are claims grounded rather than asserted?
  • Point of view — is there a real stance?
  • Audience fit — does it match who it is for?
  • Discussion value — does it invite a genuine reply?

Score to verdict

The verdict starts as a threshold on the whole-number score: IMPROVE below 50, WAIT from 50 to 74, GO at 75 and above. A weighted floor then applies: when the factors sitting at level 0 or 1 together carry at least a tenth of the total weight, the verdict is forced to IMPROVE whatever the sum says, so a draft cannot average its way past a serious weakness. Every factor weighted a tenth or more can trigger it alone; only the lightest factor, discussion value, cannot. There is no buffer zone around the thresholds; a score of 50 reads WAIT and a score of 49 reads IMPROVE. The per-factor panel shows which factors are soft so the verdict is actionable, not just a label.

Weights live in shared/core.mjs and nowhere else. The blog post signals, not scores explains the reasoning behind leading with the verdict.