Research
Borrowed science, named out loud.
The profile markers draw on validated constructs from published research — each named below in one plain line. Every claim the system makes carries receipts to real messages, and nothing on this page or in the product is a personality verdict.
-
Interpersonal circumplex.
Agency and communion — the two axes beneath the directness and warmth markers: how someone moves things forward, and how someone stays connected while doing it.
-
Need for cognition, elaboration likelihood.
Whether a contact wants the full reasoning or the one-line takeaway — and which route persuasion actually travels with them: the argument itself, or the cues around it.
-
Regulatory focus.
Promotion versus prevention: whether a message lands better framed as an opportunity or as a risk contained. A person's job can fake this signal, so it is read with that caution attached.
-
Language style matching.
Convergence toward a contact's own function words, formality, and message length — used as alignment guidance for drafting, never as a raw similarity score.
-
Conversational receptiveness.
Acknowledgment before answer. In any disagreement the recipe is fixed: credit their point, hedge your own, name the agreement, then differ.
-
Computational politeness.
How much softening a contact produces in a request, and how much they require before one reads as polite. Today's markers come from politeness theory; a lexicon-based scorer is in development.
-
Speech-act taxonomy.
What a message is doing — asking, committing, informing, urging — as the future backbone for drafting archetypes. In development; not in the shipped build.
-
Burrows Delta.
The authorship-attribution measure that will score whether a draft actually sounds like you, not just acceptably generic. In development; not in the shipped build.
Never scored
What this system refuses to measure.
The list is binding on the code, not a promise in prose — computing anything on it raises an error and stops, and a test fails if that ever changes. Restated here because it belongs next to the science:
- No discretion or forwarding-risk scores. "Will this person keep what you tell them to themselves?" is a question the system never computes an answer to.
- No protected-attribute inference. Nothing is derived on or about protected characteristics — not as a score, not as a label, not quietly as a feature inside another score.
- No clinical labels. The map describes how correspondence has behaved. It never diagnoses the person behind it.
The claims frame
Idiographic prediction, with nomothetic priors.
Every claim in a profile has one shape: how this contact has tended to respond to messages from you, stated with calibrated confidence. The evidence is idiographic — your own correspondence, receipts attached. Where that evidence is thin, population-level research stands in as a prior and is labeled as exactly that. Below the confidence floor, the system asks a plain question instead of asserting an answer.
What a claim is never: construct measurement of personality. The constructs above shape what the markers look for. They are not verdicts Cartographer issues about who someone is.
Validation
Notebooks and validation results publish here.
None exist yet. When validation has been run on real corpora, the notebooks and results publish here — including the ones that come back unflattering. Until then this section stays empty; filling it with an invented figure would break the site's own rules.