← Trust Center

How we score you

The honest version. Six signals, each with known strengths and weaknesses, combined because no single signal is enough.

Last updated: 2026-04-14 · Version 1.1.0

Our four principles

  • No source is the truth

    Every signal has known weaknesses. Fingerprint research is emerging. Self-report is biased. Resumes favor measurable work. Peer panels are small. We combine multiple imperfect signals because their errors do not correlate — when three independent sources agree, the signal is real.

  • Disagreement is data

    When your fingerprint pattern points one way and peer feedback points another, we surface the conflict rather than averaging it away. Our peer-divergence engine classifies every dimension as confirmed, challenged, blind spot, or over-estimation. Disagreement is often the most useful insight.

  • Confidence is variance-based

    Clarity scores are computed from source agreement, not source count. Two sources that strongly agree give higher clarity than five sources that contradict each other. You can see which sources drove a recommendation and which disagreed.

  • Weights are documented, not hidden

    The weights and formulas used in scoring are shown on this page. Fingerprint is capped at 15% and scaled by per-user classifier confidence — a noisy scan contributes less than that cap. Weights normalize across active sources — if you skip fingerprint entirely, its weight redistributes proportionally to the other active sources.

  • Every recommendation is explainable

    On every trajectory page, a "Why this score" panel shows exactly which perspectives contributed to the final number, the top signals matching the trajectory, any signals holding the score back, and the single next input that would most sharpen your clarity. Explanations are deterministic — the same inputs always produce the same explanation, no AI prose, nothing to guess.

The six signals and their weights

Weights reflect signal quality — how hard each source is to game and how reliable its underlying data is. They are not claims about scientific certainty.

Fingerprint (DMIT)

15%

Cognitive pattern profile derived from on-device fingerprint pattern classification (Gardner's 8 intelligences). The 15% cap is multiplied by the per-user classifier confidence — a noisy scan contributes less, the freed weight redistributes across your other active sources.

Signal quality:
Bounded by signal quality. Crisp scans land at the full 15%; uncertain scans contribute proportionally less.
Known limit:
Dermatoglyphic mapping to cognitive traits is emerging research, not settled science. We capped the weight at 15% (down from 25% on 2026-04-14) so this signal cannot dominate your recommendations regardless of how the underlying field evolves.

Resume / demonstrated skills

20%

Skills extracted from your resume text with fuzzy matching across a curated synonym map.

Signal quality:
High — what you have actually done.
Known limit:
Biased toward measurable, documentable activities.

Psychometric

15%

Big Five personality and RIASEC interest codes from an adaptive anchor-probe questionnaire.

Signal quality:
Medium — validated instruments applied to self-report.
Known limit:
Self-report is reliable for personality traits but subject to social desirability and mood effects.

Experience reflection

15%

Enjoyment-weighted activity reflections — what energized you, what you would repeat.

Signal quality:
Medium — recall bias but valuable for surfacing fit over time.
Known limit:
We discount skipped reflections and weight energizing activities higher.

Context

10%

Life stage, education, work experience, favorite subjects.

Signal quality:
Low on its own, high for adjusting other sources.
Known limit:
Used to life-stage-tailor roadmaps and to soften fingerprint/psychometric calls that conflict with context.

Peer feedback

10%

External ratings from friends, teachers, and mentors, relationship-weighted (teachers weigh more on academic dimensions).

Signal quality:
Medium — triangulates self-report with outside view.
Known limit:
Small samples (typical n = 3–5) mean variance is high. We surface disagreement explicitly rather than averaging it away.

Weights normalize across active sources. If you skip a source, its weight is redistributed proportionally to the others — no empty sources inflate or deflate your score.

The honest read on DMIT

DMIT (Dermatoglyphic Multiple Intelligence Test) has believers and skeptics. We think the honest answer lives in between.

  • Dermatoglyphic Multiple Intelligence Test (DMIT) correlates fingerprint pattern types (whorl, loop, arch, and subtypes) with cognitive and personality traits. The method is widely used in India and parts of Asia.
  • Research on dermatoglyphic-brain correlation is active but not consensus. Peer-reviewed studies show moderate associations, not deterministic predictions. Where DMIT marketing claims "95% accuracy", we do not.
  • Our classifier uses classical computer vision (orientation field estimation + Poincaré index for singular point detection) to identify 8 NIST standard subtypes. The subtypes themselves are forensic-standard, used in criminal identification for over a century.
  • The mapping from pattern to intelligence follows the established DMIT methodology, calibrated with four clinical DMIT reports from accredited providers. Where our calibration disagrees with specific providers, we note it in internal documentation.
  • We treat fingerprint as one signal among six, never the sole driver. The weight is capped at 15% and further scaled by per-user classifier confidence — a user who opts out of the fingerprint scan loses up to 15% of a possible signal, not 100% of the assessment value.

When perspectives disagree

Our peer-divergence engine compares what you see in yourself against what peers see, across eight intelligence dimensions. Every dimension lands in one of four buckets:

Confirmed

You and peers rate this strength similarly. High confidence signal.

Hidden strength

Peers see more than you see in yourself. Worth investigating.

Over-estimation

You rate yourself higher than peers do. Not a judgement — could be confidence, or could be a blind spot worth testing.

Challenged

Sources disagree meaningfully. We surface the disagreement rather than averaging it.

How clarity scores work

A clarity score of "high" does not mean you have completed many perspectives. It means the perspectives you have completed agree with each other. Clarity comes from convergence.

Two strongly-agreeing sources beat five contradictory ones. If your clarity is low, the fix is rarely "add more sources" — it is usually to look at which sources disagree and why.

What we are not

  • Not a diagnostic tool. Our outputs are reflection prompts and growth trajectories, not medical or psychological diagnoses.
  • Not a replacement for a counselor or mentor. We work best as the diagnostic that feeds a human conversation.
  • Not a predictor of future success. Career fit is one variable among many — effort, opportunity, network, timing, and luck all matter.
  • Not settled science. We ship honest uncertainty instead of fake certainty.

Read more