Being readable isn't the same as being repeated.
Most of the agent-readiness industry stops at the plumbing: can a crawler get in, is there something to parse. Necessary, and nowhere near enough, because a machine that can read you perfectly can still describe you like everyone else.
Legibility gets a machine to read you. Narrative decides what it says back, and it's part of how Beacon works, not an add-on.
Does your story survive contact with the models?
Narrative builds on legibility.
One product, two layers, scored separately and never blended. The base layer is mechanical and mostly a fix. The top layer is judgment, and it's the part that stays yours.
The mechanics of being parseable. Mostly a fix, and the foundation the narrative layer reads from.
Four steps, and every one leaves receipts.
This is the machinery behind three of the four numbers in your read, and behind the monthly Resonance loop once the record is set. Nothing is vibes; every score is re-runnable.
We read your site the way a model does, homepage, about, product, pricing, docs, your newest posts. Not a headless browser pretending to be a person: the actual text a machine can lift.
From those surfaces we extract the record: category claim, audience, core promise, differentiators, proof points. Extracted, never inferred, if your site doesn't say it, the field comes back blank, and the blank is the finding.
Then we turn around and ask the models, best tools in your category, what is this company, how does it compare. Every question runs across two models, three times each. One run is an anecdote. A batch is a rate.
Each answer is judged against what your site actually claims: named or not, accurate or not, your framing or the category's. Every run ships verbatim with the judge's call. Failed runs count against the score, it degrades, it never inflates.
Four dimensions. Four questions.
Each one maps to a layer of the story on the front page, and each is a question you could answer yourself, if you had a week and a hundred model runs to spare.
When something goes looking for you, is there anything there?
Is the record reachable and articulated everywhere a machine looks? Reachability comes from the live access probe; articulation comes from the crawl.
Wherever your story shows up, is it the same story?
Category claims that agree with each other. No contradictions between your pricing page and your docs. Nothing stale enough to read as a different company. Drift here becomes drift in the answers.
Swap your logo for a competitor's. Would anyone notice?
How much of your story could only be yours, ownable claims, language nobody else could use, specifics that can be checked. Add competitors to your read and the comparison gets sharper.
Do the models say it back, in your words, unprompted?
The output metric: how often the models name you, get the facts right, and use your framing. Three rates, per model, with transcripts. This is the number the monthly loop watches.
Stated is a story that exists only on your site. Believed is one the models repeat on their own. The work of Beacon is moving you right, and the monthly read is how you watch it happen.
Measured by machines. Authored with you.
The read tells you where the story fails. It can't tell you what the story should be, someone has to decide what's true and say it in language nobody else could use.
That's the Sightbox half of Beacon: we don't stop at making you machine-readable, we shape what the models perceive, so the answer an agent gives about you is the one you'd write yourself. Authored with you, not generated.
Access and Record are getting easier for everyone. This layer won't, because it was never technical. It's the part competitors can't copy.
Find out whether your story survives the retelling.