What is legibility scoring?
Legibility scoring grades how cleanly AI agents, aggregators, and search algorithms can parse a block of disclosure text. It returns a 0-100 score, a verdict, and the specific spans most likely to be misparsed.
The six factors
Legibility is not one thing, so the score decomposes into six transparent factors. Hedging density counts vague qualifiers per 100 words. Quantification measures the share of sentences carrying a number. Sentiment stability counts polarity-flip connectives. Negation clarity counts double-negative and inversion phrases. Structural clarity reads average and over-long sentence length. Extractability counts the data points an agent can pull out.
Each factor is rated pass, warn, or fail with a plain detail, so the score is fully auditable rather than a black box.
Why a score plus spans
A number alone tells you there is a problem but not where. Legibility scoring pairs the score with the exact sentences driving misread risk, each with a remediation. That lets a writer fix a filing surgically.
The scoring is deterministic and offline-safe, so the same text always produces the same score. It is an editorial and machine-legibility analysis, never investment advice, a prediction, or a recommendation.
How LyraMind implements it
The score starts at 100 and subtracts more for each failing factor than each warning. Two or more fails, or a score under 55, returns HIGH_MISREAD_RISK; one fail or two warnings returns REVIEW_LANGUAGE; a clean read returns AI_LEGIBLE. The flagged spans are ranked by how many issues each triggers and how long it is.
An optional graded mode can ask a panel of live models to parse the text and measure their disagreement behind the same scan contract. The scan records its verdict and the flagged questions to the flywheel and the Trust Ledger unless a higher-level report records its own composite entry.
Related concepts
Legibility scoring is the sentence-level primitive under AI readiness and misread risk. Its factors drive the institutional simulation's per-persona comprehension and the Corporate AI Readiness Report.
- Six factors: hedging, quantification, sentiment stability, negation, structure, extractability
- Each factor rated pass, warn, or fail with a plain detail, so the score is auditable
- 0-100 score with verdict AI_LEGIBLE, REVIEW_LANGUAGE, or HIGH_MISREAD_RISK
- Pairs the score with the exact spans to fix, each with a remediation
- Deterministic and offline-safe; the same text always scores the same
- Editorial and machine-legibility analysis only, never advice or a prediction
Questions
What score is considered legible?
A clean read with no failing factors and few warnings returns AI_LEGIBLE. A score under 55 or two or more failing factors returns HIGH_MISREAD_RISK.
Is the score deterministic?
Yes. The default scan is deterministic and offline-safe, so identical text always yields the same score and verdict.
Does legibility scoring rewrite my text?
No. It flags the spans most likely to be misparsed and suggests a fix for each, leaving the rewrite to you.