J C
You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$200,000 left to reach the funding goal.
A document review should let someone trace an answer back to the evidence. We want to make that easier to test and teach. Sentiq Robotics LLC proposes a public benchmark and teaching kit built from synthetic documents. Each case would contain a claim, its source material, a reference judgment, and a record of disagreements. The cases would cover problems such as missing evidence, conflicting passages, and conclusions that sound plausible but are unsupported.
We are seeking $100,000 for a twelve-month initial project, with a $200,000 goal for a larger case set and more independent review. Funding would pay the founder and project workers to produce and maintain the materials. Educators and small research teams would be able to download the case set, teaching guide, and reproducibility package. Sentiq is a for-profit LLC seeking support for this defined public-benefit project; acceptance and grant terms remain for Manifund to determine.
At the $100,000 level, our planned outputs are thirty synthetic cases, a documented review rubric, source-linked reference judgments, a reproducible evaluation package, and a teaching guide. Each released case would identify the relevant source passages, the limits of the evidence, and any unresolved reviewer disagreement.
Months 1 through 3: appoint the technical and independent review contributors, agree the rubric, and build a first batch of ten cases. Months 4 through 8: complete the thirty-case set, review the reference answers, and test whether another person can reproduce the recorded judgments from the supplied sources. Months 9 through 12: revise weak cases, finish the teaching guide, and publish the materials and a report of what worked and what did not. These are proposed milestones beginning after funding is available.
At $200,000, the planned set grows to seventy-five cases. The larger budget adds engineering time, independent review, accessibility work, and an external reproduction check. Between those amounts, we would publish a revised work plan matched to the funding received. We would seek feedback from educators and researchers, but no participant or adopter is committed today.
We would report completed and excluded cases, reproducibility problems, and reviewer disagreements. We would not treat agreement with a reference key as proof of real-world accuracy or improved learning. The new synthetic cases and teaching materials are intended for free public reuse; any software included in the release would first be checked for suitable rights. Customer records, restricted technology, and proprietary material outside that release scope are excluded.
The minimum request is $100,000 for twelve months. The expanded goal is $200,000 for the same period. All figures below are proposed gross project costs.
At $100,000:
Founder project lead: $48,000, based on 20 hours per week for 48 weeks at $50 per hour.
Reproducibility engineering: $24,000, based on 400 hours at $60 per hour.
Independent technical review: $12,000, based on 120 hours at $100 per hour.
Case authoring, teaching materials, and accessibility: $8,000, based on 160 hours at $50 per hour.
Project computing, hosting, and preservation: $3,000.
Employer costs, project accounting, and administration: $5,000.
Total: $100,000.
At $200,000:
Founder project lead: $60,000, based on 25 hours per week for 48 weeks at $50 per hour.
Reproducibility engineering: $60,000, based on 1,000 hours at $60 per hour.
Independent technical review: $30,000, based on 300 hours at $100 per hour.
Case authoring, teaching materials, and accessibility: $24,000, based on 480 hours at $50 per hour.
Project computing, hosting, and preservation: $10,000.
Employer costs, project accounting, and administration: $16,000.
Total: $200,000.
Founder pay is an explicit cost of dedicating time to this work and can support living expenses during the project. It averages $4,000 per month at the minimum or $5,000 per month at the expanded goal, before personal taxes. Payment would be for documented project work after funding is available, subject to the grant agreement and an appropriate employment arrangement. These estimates are not existing salaries, accepted supplier quotes, or promises of tax treatment.
The founder role covers project planning, contributor coordination, case planning, review scheduling, public documentation, and reporting. Technical work and independent judgment have their own budgets. Final rates, employer costs, any grant fees, and assignments would be reconciled before work is committed. Support from another funder or usable service credits would be disclosed and the same costs would not be charged twice. Accounting and administration exclude coordination and documentation already budgeted under the founder and contributor roles. Employer costs are provisional allowances to be reconciled with the actual staffing arrangements.
Jamell Cleary, founder and President of Sentiq Robotics LLC, would lead project coordination, records, budget, and public reporting. Engineering, case preparation, accessibility, and independent review would be paid assignments with defined deliverables. No outside specialist is represented as retained, and the independent reviewer would not simply approve their own case-writing work.
Sentiq has an internal local software demonstration that links claims to source passages, records review revisions, and exports evidence bundles. That is a starting point for this work. It does not establish the quality of the proposed benchmark, a completed user study, or a demonstrated learning benefit. The grant would fund new public materials and outside scrutiny of them.
The cases could be too simple to teach the difficult parts of evidence review, or reviewers could disagree enough that a single answer key would mislead people. Technical work may also take longer than planned, and we may struggle to recruit suitable independent reviewers or interested users.
We would test the first ten cases before expanding, keep disagreements visible, and remove cases whose reference judgments cannot be defended. If costs or staffing change, we would disclose the change and obtain any approval required under the grant agreement before reducing the promised case count or changing the funded scope. A useful but smaller release is one possible outcome; another is that the work fails to produce a useful teaching resource. We would publish that result, the spending report, and any reusable material. Unspent funds and any change of purpose would be handled under the grant agreement.
Our support so far has been noncash: credits through Anthropic's Claude program and benefits through NVIDIA Inception. We are also participating in an accelerator. These benefits are not cash raised for this project, and we have not assigned an unverified dollar value to them here. No cash funding for this project is recorded. Any additional cash or overlapping support would be disclosed before an award.
There are no bids on this project.