What Is a Statement of Work (SOW) for Data Collection?

A custom dataset can only be checked against what was written down before capture. An order that names hours but not the task, the conditions or the acceptance test leaves every disagreement to be settled after footage exists. The statement of work is where a buyer fixes what a batch must contain, so a fast, on-demand collection can start from a signed spec instead of a conversation.

On this site the written skill spec carries the statement-of-work content: named skill, environments and conditions, acceptance criteria and target hours, agreed before a single contributor is briefed. Commercial terms sit in an order form or master services agreement signed by both parties. Prices, payment terms, minimum term, SLA penalties and warranty terms are not stated.

A statement of work for data collection is the part of a contract that says exactly what is being bought. A master agreement or order form sets the general commercial terms. The statement of work defines the specific task and deliverable under it. For a custom dataset, that means the task to capture, the geometry that must stay in frame, the conditions to cover, the volume and the format it arrives in. Without it, an order describes hours rather than data.

The site’s procurement guidance describes the same split. A master agreement or order form covers general commercial terms, a statement of work or spec document covers the task and deliverable, a data processing agreement covers personal data in the footage, and a license covers what the buyer may do with the dataset. Each answers a different question, and treating one as settling another is a common source of confusion once delivery starts. This page is general information, not legal advice.

The order form is the commercial document. The site’s terms state that Firsthand collects custom egocentric video datasets to a specification agreed with the buyer and delivers them under a separate written data license, and that commercial terms are set out in an order form or master services agreement signed by both parties. Nothing on the website is an offer capable of acceptance. Where an order form grants a perpetual, buyer-owned commercial license, those license terms control over anything on a web page.

The skill spec is the working document for the capture itself. The site calls it the written contract for a capture program: the named skill, the environments and conditions to cover, the acceptance criteria and the target hours. Its procurement guide lists “statement of work / skill spec” as a single row, so on this site the skill spec is where the statement-of-work content lives. Whether a buyer’s legal team wants a separate SOW wrapped around it is a contract question the site does not answer.

The site’s skill-spec guide gives five parts: the named skill, the geometry that must be visible, a condition budget in hours, the reject rule, and one acceptance test a reviewer can apply. Its scoping guide frames the same ground as five decisions made before capture: the exact task, the conditions and population to cover, the validated-hour volume, the labels and delivery format, and the consent and license terms.

Read together they give a checklist of what is captured, under which conditions, how much, in which labels and format, under which license, and how a batch is judged. Format matters because the SOW is where a buyer names the structure their training stack expects; the site’s procurement guide cites RLDS as one reference delivery structure an order can specify.

Billing is stated per validated hour, not per raw hour recorded, and every item must clear the signed accept criteria, anonymization and consent checks. Items that fail go to a visible reject log rather than a downgraded delivery. The SOW is the place those criteria are written down.

On-demand collection points an already-warm pipeline at a new brief, so the SOW is what turns a described gap into something that pipeline can execute. The site states that brief to spec takes the same week, and that the published clock starts at the signed spec: a median of 48 hours to the first validated batch where coverage is already deep, and three to four weeks where contributors must be recruited first in a thinner environment.

That start point is why a complete SOW is the main lever on speed. The site notes that a missing element, whether task, geometry, condition budget, reject rule or acceptance test, causes rework and disagreement mid-collection. There is no minimum program: a focused brief on one condition, one region or one missing scene is a normal starting point, and top-ups run against the original spec and accept criteria so new data stays consistent with what the buyer already holds.

For an unfamiliar spec, the site describes a Spec Pilot: a small, fast first batch captured against the written skill spec to prove the spec and accept criteria before volume. It is optional, billed per validated hour like a full collection, and delivered with the same reject log and coverage reporting. No figure is stated for pilot size or timing, so write both into the SOW rather than assuming them.

Samples are a different thing. The terms state that sample datasets are provided for evaluation only, and that redistribution, resale or use in a production or commercially deployed model requires a separate commercial license. A pilot proves your spec; a sample shows existing footage. Keep them apart in the document and in the plan.

A procurement team will expect terms the site does not publish. They are listed as not stated and should be raised in the spec conversation, not inferred: payment terms and prices; acceptance criteria beyond the stated QA gates (the signed accept criteria, reject rule and acceptance test); IP and warranty terms beyond the stated perpetual, buyer-owned commercial license; minimum term; SLA penalties and guaranteed delivery dates; turnaround and cadence for second and later batches; and change-request terms.

On licensing, the site states that exclusivity, where a program needs it, is a separate term negotiated in the order form, which is the governing document. It states no standard sharing or redistribution term, so raise that before the spec is signed. This is not legal advice: have counsel review the order form and any SOW, and treat the executed documents, not a summary on a web page, as controlling. When the scope is ready, request a custom dataset.

What is a statement of work for data collection?

The document that defines the specific task and deliverable under a master agreement or order form: the skill, geometry, condition cells, volume and format. It is general information here, not legal advice.

Is a skill spec the same as a statement of work?

The site’s procurement guide lists them as one row, so the skill spec carries the statement-of-work content. Whether a buyer wants a separate SOW document around it is a contract question the site does not state.

How fast can a custom collection start once the SOW is signed?

Brief to spec takes the same week. From a signed spec, the first validated batch arrives in a median of 48 hours where coverage is already deep, and three to four weeks where contributors must be recruited first.

Do I need a pilot written into the SOW?

No. A Spec Pilot is optional: a small, fast first batch that proves the spec before volume. No minimum program is required, and no pilot size or timing figure is stated.

Does the site state prices or payment terms?

No. Prices, payment terms, minimum term, SLA penalties and warranty terms are not stated, so raise them in the spec conversation and have counsel review the executed order form.

Who owns the data under a custom collection?

The site states a perpetual, buyer-owned commercial license, with exclusivity as a separate term negotiated in the order form, which is the governing document. This is not legal advice.