What AI never decides

Never rejects a candidate

There is no auto-reject path in the product. No Stori output removes a candidate from consideration without a human choosing to do so. Mechanism: rejection actions exist only in employer-facing review surfaces, taken by a person.

Never makes the hire

Stori produces evidence and organization, not employment outcomes. Mechanism: the product has no decision output — there is nothing to auto-accept with.

Never ranks on personality

The Big Five self-report is returned to the candidate as self-insight and shown to employers as context. It does not contribute to ranking, prioritization, filtering, or exclusion. Mechanism: trait results are excluded from the evidence-ordering inputs by design.

Never scores speech delivery

Fluency, accent, pace, hesitation, and speech patterns carry no signal, positive or negative. Evaluation reads the substance of the transcript — can the candidate convey their thoughts so their intention is understood. Mechanism: the rule is written into the interviewer and every evaluation prompt, and candidates who need an adjustment (extra time, restarts, an alternative format) can request one before the interview.

Never uses protected characteristics or their proxies

Race, sex, age, disability, and other protected characteristics are not evaluation criteria; ZIP code is not used as a proxy. Location is used only as a job-relevant signal — proximity to a role's stated work location. Mechanism: scoring inputs are enumerated and audited against a proxy list.

Never infers demographics

Stori does not estimate a candidate's race, gender, or other protected characteristics from names, photos, voices, or geography — for scoring or for anything else. Any future bias monitoring will use voluntarily self-identified data, held separately from all scoring inputs. Mechanism: no inference pipeline exists, deliberately.

Never learns from hired / not-hired outcomes

No model in the system is trained on which candidates employers accepted or rejected, so historical selection bias cannot be absorbed as ground truth. Mechanism: there is no outcome-labelled training dataset in the architecture.

Never produces an unreviewable output

Every score, ordering, and interpretation traces to source evidence a human can inspect, and can be overridden — with overrides captured. Mechanism: receipts are generated with the output, not reconstructed after the fact.

Never fills a gap with a guess

Where evidence is thin, the output is “insufficient signal,” not an invented score. Mechanism: abstention is a first-class output value across the scoring pipeline.