Real workflow data.
Real expert evaluation.
For AI training.
Iceberg curates proprietary corpora from operating companies and runs the human-data operations on top — task design, rubrics, gold standards, scoring, and expert review. Built for vertical capability work and long-chain reasoning.
Operating a business with a corpus? For data owners →
What Iceberg Does
Iceberg brings together proprietary workflow corpora from operating companies and human-data operations led by domain experts with evaluation experience.
Not a self-serve portal. Not a vendor marketplace. Curated, hands-on engagements scoped to a specific domain and a specific goal.
Two arms of the work
Proprietary workflow corpora from operating companies
Real documents from real businesses — leases, claim files, dispatch logs, underwriting packets, investor reports. Native formats. Not scraped, not synthetic. Anonymized in the public catalog; firm names disclosed under NDA.
See the catalog →Task design, rubrics, gold standards, scoring, expert review
Outsourced human-data work staffed by domain practitioners with lab experience. Specialty: long-chain reasoning and other indeterministic work where rubric quality and reviewer judgment dominate.
How we work with AI training teams →Where the work is hard
Short tasks with clean answers are well-served by existing vendors. The hard work is the other kind — multi-step domain workflows where the “right answer” depends on methodology and the order of operations, and where a plausible-looking output can be quietly wrong. That’s the work we focus on.
Long-chain reasoning
Multi-step workflows where rubric quality dominates. We design and version rubrics like product.
Domain expertise
Reviewers are practitioners who have actually run the workflow — not generalists applying a checklist.
Indeterministic scoring
For tasks with no single right answer. Multiple reviewers, calibrated, with inter-rater agreement tracked.
Defensible methodology
Rubrics, gold standards, and scoring runs documented end-to-end. Auditable for procurement and research alike.
Worked Examples
We’ve published a set of commercial-real-estate workflows end-to-end under our methodology — task decomposition, rubric authorship, gold-standard construction, and model scoring across frontier models. Concrete artifacts, not case-study slides.
Early Lease Termination & Lender Consent
Office · 7 models · 18 scored fields
Lease Abstract → NER Calculation
Industrial · 7 models · Single + multi-step
Start a conversation.
Confidential. NDA on request. No slide deck required.