Governing principles for evidence-based, human-accountable authentication at Poshmark. Defines the relationship between AI decision support and human judgment.
The Authentication Team exists to protect marketplace trust through evidence-based, consistent, and defensible item review. The function serves two obligations simultaneously: protecting buyers from counterfeit risk and ensuring sellers receive fair treatment grounded in assessable evidence.
Artificial intelligence can improve the speed and consistency of review, but accountability for every outcome remains with the Authenticating Agent.
The team operates under a single governing principle: evidence before assumption. Every verdict must be grounded in visible, assessable evidence from the submission. A gap in visibility is an evidence gap — not a signal of inauthenticity.
Incomplete photo coverage, unclear markers, or insufficient submission quality means the case requires better evidence, additional review, or escalation. It does not, on its own, support a reject decision.
Where evidence is inconclusive, the default path is further evidence gathering, additional review, or escalation rather than rejection. Items reviewed under materially similar conditions should reach consistent outcomes regardless of Authenticator, workload, or queue pressure.
VERIFIED is the authentication platform. RAVEN (Authentication Assistant) is the AI decision engine operating within it. The distinction is intentional and material.
RAVEN provides advisory decision support and does not supersede trained Authenticator judgment. A RAVEN verdict is a structured recommendation produced from available photographic and contextual evidence. It is one input into the review process — not the final determination.
RAVEN supports standardization, pattern recognition, and review consistency. It does not independently authorize an outcome, and no RAVEN recommendation is operationally final without human acknowledgment.
RAVEN's role is bounded: it produces a recommendation based on available evidence and surfaces the reasoning behind it. Accountability for the final decision remains with the Authenticating Agent.
The Authenticating Agent is responsible for assessing the available evidence, confirming its completeness and relevance, applying policy, and owning the action taken on the case. This accountability is not transferable to an automated system.
A RAVEN recommendation may be overridden when available evidence supports a different conclusion. Overrides are a legitimate and expected feature of a human-led review system. They are also governed.
Override authority is restricted to internally signed-off Authenticators. Every override must be documented and auditable. The review record must reflect that an override occurred, the basis on which it was made, and the identity of the accountable Authenticating Agent.
This requirement exists to protect consistency, enable quality review, and ensure that override patterns generate learning signals for standards, training, and tooling improvement over time.
Where an Authenticator cannot confidently support an outcome, the case should be escalated rather than resolved through speculation. Escalation is the appropriate safeguard for material uncertainty — not a sign of failure.
Cases warrant escalation when evidence is incomplete or conflicting, when the submission presents unusual risk, when the item falls outside the Authenticator's confidence level, or when neither an Authentic nor a Counterfeit outcome can be supported with adequate certainty.
Escalation routes upward to an Authenticator with greater authority or subject matter depth. Where internal review remains materially uncertain, approved third-party authentication may be engaged at Senior or Manager discretion.
The system is human-led. RAVEN supports the review process. Agents conduct the review. Internally signed-off Authenticators hold override authority. Senior Authenticators resolve material uncertainty.
Every decision, revision, override, and escalation must be traceable in the record. The Authentication Team's objective is not throughput alone — it is accurate, fair, and defensible decisions that can be understood, reviewed, and improved over time.
The Authentication Team operates under the following doctrine: