Paper Instruments

AI for strategic projects leads

Ultramarine Work turns your project notes, the client’s spec and the task platform’s exports into plain-English contributor guidelines, yes-or-no rubric criteria and the weekly report on throughput and acceptance.

Try this mode free

7 days free, then from $5 a month.

Project-report-week-46.xlsxExample
Task typeAcceptedAcceptanceReworkHandle timevs target
Finance Q&A41284%6%38 min+12
Multi-step reasoning65579%9%26 min+15
Python debugging29871%14%47 min-62
Rubric writing18688%4%52 min+6
All task types1,55180%9%36 min-29

NotePython debugging fell from 82% to 71% acceptance after Monday’s guideline change and is 62 tasks short.

Ask it for

  • “Turn my notes for the tax law project into guidelines a CPA with no AI background can follow, with a one-page quick reference.”

    Tax-law-guidelines-v2.docx
  • “Review the 40 rubrics from Tuesday’s batch and mark every criterion that isn’t yes or no, needs the reference answer, or penalizes a missing point.”

    Rubric-review-batch-14.xlsx
  • “Build this week’s report from the task export, with throughput, average handle time, acceptance and rework by task type against the target.”

    Project-report-week-46.xlsx
  • “Python debugging acceptance fell from 82% to 71% after Monday’s guideline change. Find which rules and reviewers drove it before my 3pm client call.”

    Acceptance-drop-week-46.docx
  • “Plan staffing to deliver 6,000 accepted tasks by December 12 at our current handle time and acceptance rate, and show how many more contributors we need.”

    Staffing-plan-Dec-12.xlsx
  • “Update the project budget with October’s contributor hours and pay rates, and forecast spend to the end of the contract against the client’s PO.”

    Project-budget-Oct.xlsx

Custom Strategic Projects Leads (Frontier Data) Skills

  • Contributor task guidelines

    Rewrites your rules into guidelines and quick references an outside expert can follow, one instruction per rule, with worked examples and dated change notices.

  • Response grading rubrics

    Writes and reviews rubric criteria a grader can answer yes or no from the response alone, one point per line, with penalties only for real errors.

  • Operating metrics and reviews

    Builds the weekly report on throughput, handle time, acceptance and rework by task type and contributor tier, with the reason each number moved.

  • Ad hoc analysis and findings

    Answers one question from the exports, such as why acceptance fell after a guideline change, and says what the data can and cannot show.

  • Workload capacity and staffing model

    Works out the contributor hours and headcount a delivery date needs at current handle time and acceptance, and shows the gap week by week.

  • Project budget and cost forecast

    Tracks contributor hours, pay rates and review costs against the project budget, and forecasts spend to the end of the contract.

What’s inside the mode

  1. Custom skills

    6 battle-tested playbooks for Strategic Projects Leads (Frontier Data)

  2. Tools

    Read and edit files, run code, search the web

  3. Connectors

    Connections to your data sources and apps

  4. Office skills

    Real Excel, Word, PowerPoint and PDF files

  5. Style & Design

    Professional, restrained design

Connected to the apps you already use

  • Airtable
  • Asana
  • Jira
  • BigQuery
  • Gmail
  • Outlook
  • Google Calendar
  • Google Drive
  • Google Docs
  • Google Sheets
  • Google Slides
  • SharePoint & OneDrive
  • Excel
  • Dropbox
  • Box
  • Slack
  • Microsoft Teams
  • Zoom
  • Granola
  • Notion
  • Additional custom connectors

How it works

  1. 1

    Put the project folder together

    Bring the client’s spec, your notes on the task, the current guidelines and change log, task and review exports from the platform, contributor rosters with pay rates, and the delivery schedule.

  2. 2

    Ask for the piece you need

    Ask for a guideline rewrite, a rubric review, the weekly report, a staffing plan or a look at why a number moved. Every figure names the export it came from, and anything it could not reconcile is listed for you.

  3. 3

    Keep the content yours

    The task’s rules and the quality bar come from you and the client. You check every file it touched, decide what contributors see, and send the update yourself.

Who it’s for

Strategic projects leads, program leads and operations managers at frontier data companies, who run the expert contributor projects that produce training and evaluation data for AI labs.

Private until you share it

  • Private by default

    Your folders, workstreams and chats are yours alone until you share a folder.

  • Only what you give it

    A workstream works from its own folder and the connectors you turn on. A cloud workstream cannot reach files on your computer.

  • SOC 2 Type II

    Paper Instruments is SOC 2 Type II, and companies can request a zero data retention agreement.

  • You review everything

    Every file it touched is marked, and you decide what to use and what goes out.

Questions

Will it write our task guidelines for us?

It writes from the content you give it. You supply the task’s rules, the edge cases and the quality bar, and it puts them in plain English that an expert outside AI can follow, with one instruction per rule, examples of good and bad work, and a change notice when a rule moves. Where your notes leave a gap, it lists the question for you instead of inventing a rule.

What makes a rubric criterion good enough to ship?

A grader can answer it yes or no from the response alone. Each line checks one thing, names the specific fact or step this task needs, and spells out what counts. Penalties cover real errors, such as a wrong figure or unsafe advice, and a point the response left out loses its credit without a penalty. It checks a contributor’s rubric against those rules and marks each line that fails one.

How does it work out a staffing plan for a delivery date?

It works from your own numbers. It takes the tasks left, the delivery date, current average handle time, acceptance rate and the hours each contributor tier works, then computes the hours needed each week and the gap to today’s roster. Each assumption sits in its own cell, so you can lower the acceptance rate after a guideline change and see the date move.

Can it read our task platform or the client’s portal directly?

It works on the exports you put in the folder, such as task logs, review scores and contributor rosters. It can also read Slack, Google Sheets, Gmail and the other connectors you turn on. It does not log into the task platform or message contributors, and you send every update yourself.

Where does contributor and client data go?

Ultramarine Work changes files only in the project’s folder. When a request needs it, it can read files elsewhere on your computer, never passwords, keys or app data, and you can turn that off in settings. The model runs on Paper Instruments’ servers, so the content it works on is sent there to be processed. Paper Instruments is SOC 2 Type II, and our security and compliance details are on our compliance page.

Our client agreements restrict how project data is handled. Can we use it?

That decision belongs to your company, so take it to your security or legal team first. Content the model works on is sent to Paper Instruments’ servers to be processed, so they should read the compliance page against your client agreements and NDAs, which take precedence.

See every mode

Try the Strategic Projects Leads (Frontier Data) mode in Ultramarine Work

Powered by Ultramarine 1 Flash, more than 10 times cheaper than frontier LLMs.

Try this mode free

7 days free, then from $5 a month.

See how Ultramarine Work works