Project summary

I propose a small, freely readable teaching kit for students and small protein-research teams learning to keep experimental decisions traceable. It would connect candidate sequences, raw assay observations and the next batch of experiments through two clearly labeled synthetic rounds. The files and walkthrough would be usable without buying a DeltaFold subscription.

The purpose is practical scientific education: help readers recognize missing, failed and censored observations, preserve their meaning, and distinguish measured values from predictions and reference annotations. It is not a laboratory experiment or a claim that a model discovers better proteins. This proposal was prepared with AI assistance.

What are this project's goals? How will you achieve them?

The proposed output is one compact public package: synthetic CSV examples, a data dictionary, expected outputs, documented reproducibility checks and an accessible walkthrough. Each round would use one parent sequence and one consistent assay definition. The example would explicitly retain units, conditions, replicate types, source identifiers and the selection made before the next outcomes are revealed.

Proposed sequence after any separately reviewed grant acceptance: define the teaching task and acceptance checks in week 1; prepare the two-round examples and recovery checks in weeks 2-3; finish the walkthrough and publish the results in week 4. These are planning estimates for part-time work, not a commitment to full-time participation or a fixed start date. No hiring, laboratory work, private datasets or new compute service is required.

Only new synthetic examples and cleared explanatory material are proposed for public release. Licensing of any deliverable would be settled before an agreement; this application does not license DeltaFold's proprietary software or anyone else's material.

How will you know it worked?

Before implementation, specify the input files, expected results and checks. Acceptance would require two complete scripted rounds with zero lost source identifiers or raw outcome statuses; an exact round trip of the published example through its documented export/recovery procedure; and an explicit warning or rejection when an example mixes incompatible assay units or parent identities. The kit would include commands, versions and expected outputs so a reader can repeat the checks.

Publish the actual results, including any failure and unresolved limitation. These are software and teaching-material checks, not observed laboratory time savings, user-study results, predictive-accuracy improvements or biological validation. Independent usefulness and uptake would remain unmeasured until separately evaluated.

How will this funding be used?

The proposed request is $2,000, all for 40 estimated hours of founder work at a proposed $50/hour: 12 hours ($600) for synthetic examples and the data dictionary, 16 hours ($800) for reproducible checks and recovery examples, and 12 hours ($600) for the accessible walkthrough and final documentation. Both the minimum and maximum request are $2,000 because this is a single bounded package.

The rate is a budgeting estimate, not my current wage or an existing expense. For context, the U.S. Bureau of Labor Statistics May 2025 table lists a $50.14 median hourly wage for software quality assurance analysts and testers: https://www.bls.gov/news.release/ocwage.t01.htm . No matching funds, equipment reimbursement, GPU budget or another person's contribution is assumed. The requested funds would support future public educational work, not reimburse historical business spending.

Who is on your team? What's your track record on similar projects?

I am Adam Kaluza, DeltaFold's founder and the only person currently working on it, and a UC Berkeley student. I operate DeltaFold as a sole proprietorship; this is an individual proposal, not a university-sponsored application. I built the research workbench at https://deltafold.org.

Its deployed Experiment Campaigns workflow connects one-parent/one-assay variants, pre-outcome selections, sample handoffs and follow-up measurements. Published software verification includes two synthetic rounds and account-recovery checks: https://deltafold.org/updates#experiment-campaigns . That existing workflow is the starting point. The additional public teaching kit described here is proposed work, not an already completed deliverable. No other team member or partner laboratory is involved in this proposal.

DeltaFold already includes synthetic practice inputs and guided Help. This request funds a separate downloadable teaching package and independently repeatable checks, not rebuilding or reimbursing those completed features. Only new incremental work is included in the proposed budget.

What are the most likely causes and outcomes if this project fails?

The main risks are that the examples are too specialized to help learners, that recovery instructions are hard to reproduce, or that the scope takes longer than estimated. A small fixed task and explicit expected outputs limit scope; documenting failed checks prevents an incomplete demonstration from being presented as reliable. If the educational case proves weak, report that result and preserve the usable materials instead of claiming a successful scientific outcome. Any budget or scope change after funding would follow the separately agreed grant terms.

How much money have you raised in the last 12 months, and from where?

DeltaFold has raised $0 in external funding and has been personally funded by me. Historical founder spending is separate from this request and is not being offered as matching funds. Other funding and compute-credit applications are pending; none is represented here as an award. Public teaching work also appears in a broader grant proposal under preparation. If another source funds overlapping work, the same hours and deliverables would not be charged twice; any overlap would be resolved before accepting funds.