Questions
The questions buyers actually ask
Taken from first calls and vendor reviews. Where the honest answer is inconvenient for us, it is written that way.
Engagement
- How does a first engagement start?
- With a paid pilot, usually two weeks. You send a rubric and the acceptance bar you already use. We return a two-page feasibility memo, a staffing plan and a fixed quote. If we miss the bar, the rework is at our cost.
- Why is the pilot paid rather than free?
- A free pilot is staffed like a free pilot. Paying for it means we assign the same named pod, the same reviewers and the same QA pipeline you would get in production, so the result actually predicts the engagement.
- What is the smallest engagement you take?
- A pilot batch in the low thousands of items, or a single benchmark workstream. Below that the calibration overhead outweighs the value and we will say so rather than take the work.
- Can you work inside our tooling?
- That is the default. Most labs and platforms mandate their own annotation platform, task queues and data environment. Our people work inside your stack under your access controls. We are deliberately tooling-agnostic and will not ask you to adopt a platform.
- How quickly can a pod be staffed?
- We hold roughly 1.3 times the deployed demand on a calibrated bench, so a standard pod is staffed within days. Expert-tier domains such as clinical or legal review take longer because credential verification is part of onboarding.
Quality
- What happens if a batch misses the acceptance bar?
- It does not ship. Internal QA runs two points above your bar, so a predicted miss is reworked before delivery rather than after. If something escapes and you find it, the rework is at our cost within the window written into the SOW.
- How do you measure inter-rater agreement?
- Cohen’s kappa or Krippendorff’s alpha on a double-rated sample, reported per rubric dimension rather than as a single headline number. Production does not begin until the pod clears the agreement gate on two consecutive calibration rounds.
- What is in the QA evidence pack?
- The per-item review trail, agreement data by dimension, blind gold-task results and the defect taxonomy for the batch. It ships with the delivery, not on request.
- Will you audit work produced by another vendor?
- Yes, and it is often the best way to start. We publish the metrics either way, including findings that are commercially inconvenient for us.
Security and compliance
- How is confidential model data handled?
- Work is performed inside your environment by default, so the data never leaves your systems. Where we do host, it is encrypted at rest and in transit under least-privilege role-based access with 90-day access reviews and deletion certificates at project close.
- Who signs an NDA?
- Two layers. A company-level mutual NDA with you, plus an individual NDA binding every contributor to that specific engagement. Both are signed before any task access is granted and tracked in a compliance register with expiry dates.
- Are contributors verified?
- Government-ID KYC and address verification for everyone, plus reference checks for reviewer roles and above. Named accounts only, never shared logins, with same-day deprovisioning on offboarding.
- Do you hold SOC 2?
- Not yet. The security policy pack, incident-response plan and pre-drafted vendor questionnaire responses exist today, and SOC 2 Type I is on the roadmap with Type II following. We would rather state the position plainly than imply a certification we do not hold.
Commercial
- How do you price work?
- Four models: fixed-scope project delivery, a monthly retainer for a dedicated pod, per-item or retainer pricing for QA over someone else’s output, and project pricing for benchmark engineering. Rates depend on domain, expertise tier and volume, and are quoted per engagement.
- Who owns the work product?
- You do. All work product is assigned to the client by contract, and every contributor signs an IP-assignment agreement as part of onboarding.
- Will our name be used as a reference?
- Only with written consent. Until then, engagements are published as anonymised patterns with the client and volumes obscured, which is why the work section reads the way it does.
Something not covered here? Send the scope and we will answer it directly.
Ask us