# Scientific benchmark task proposal

Blank local draft. Describe your scientific task, sources, checks and feedback. A few sentences per field are enough; mark unknowns. Existing workflows and an RSI study design are optional. An initial suggestion starts suitability and format discussion and has no reviewed-data status.

Use public descriptions and source links. Keep private records, credentials and hidden evaluation answers out of this proposal.

Full datasets may remain at their current host. Include a versioned data link or access instructions, file format and approximate size alongside the source paper.

## Research track (optional)

Choose the main scientific area: `molecular-discovery`, `materials-science`, `biology-biomedicine`, `physics-engineering`, `earth-environment`, `mathematics-computation`, or `other-interdisciplinary`. For Other, describe the specific field.

- Research track (`research_track`):
- Specific field or secondary area (optional):

This is catalog classification. The package manifest's separate `track` field still identifies the review route (`candidate_resource` or `rsi_task`).

## 1. Scientific problem

- Working title and field:
- Scientific objective, inputs, useful outputs and constraints:
- Your relationship to the work:

## 2. Reproducible starting workflow (optional)

- Existing scientific method, tool or recoverable code version:
- Public source, version, dependencies and input example:
- How it could be replayed; missing pieces:

## 3. Evidence of a limitation (optional)

- What was actually run or observed, with a public evidence link:
- Suspected cause or untested hypothesis (separate from observation):
- Known failures, uncertainty or saturation; say if no run exists:

## 4. Optional RSI study design

- What diagnostic, experimental-design or modification-selection mechanism could change:
- Permitted changes and components that must remain fixed:
- How the changed version could be installed and called to drive a later self-modification:
- What evidence could identify the version actually used:

This section is for contributors proposing an RSI experiment alongside their resource. Ordinary benchmark and data contributions can leave it blank.

## 5. Feedback and independent checks

- Available scientific feedback and its query or resource cost:
- Reference method, scientific validity checks and invalid outputs:
- Related scientific families and any available independent evaluation cases; state if none exists:
- Existing separation of development, selection and final assessment, if any; unresolved overlap:
- For an optional RSI study: common starting states and budgets for comparing old/new improvement procedures:

## 6. Sources, rights and participation

- Original paper title, authors, citation or DOI (if available):
- Original creators, public data/code links and exact versions:
- What this package adapts from the original study, and what you are adding:
- Measured / computed / generated origin; units and known transformations:
- Access, licenses and permissions still needed:
- Small contribution you could offer and help needed:

## Contributors (optional)

One shared dataset can have several contributors. Add one row per person; include everyone who prepared the contribution. The source paper keeps its complete author list.

| Name | Institution (optional) | Contribution | Institutional city (optional) |
| --- | --- | --- | --- |
| | | | |
| | | | |
| | | | |

- Team notes or contact person:
- Source for paper affiliations, if relevant:
- Map preferences or people who prefer not to appear:

Public profile details are confirmed before publication. Locations represent institutions or voluntarily shared cities, never private addresses. All contributors link to the same dataset listing.

## Permission when submitting publicly

- [ ] I have the right to share this original proposal text under CC BY 4.0. Linked resources retain their own licenses.

## Discuss the draft with the project

Send a review enquiry by email: [xxiao9@utk.edu](mailto:xxiao9@utk.edu), with **RSI4Science** in the subject. Include public source links, a representative example and any preparation questions; attach this draft if useful. Permanent contact page: https://rsi4science.org/contact/.

Agree a permitted transfer route before sharing restricted files. Sending an enquiry does not accept the resource or grant publication permission.

A proposal is not a reviewed task. Ordinary benchmark tasks do not require an agent, an RSI experiment or a positive improvement result. Record original sources and actual contributions. Confirm names and affiliations with contributors before public attribution.

## When the package is ready

Use `track: "candidate_resource"` for an ordinary benchmark task and omit `rsi`. Use `track: "rsi_task"` only when supplying the optional executable RSI study contract. The current checker requires scalar input/target records; agree a representation first for unsupported formats. Before formal review, prepare `manifest.json` and `records.json` using the submission-format guide, run the automatic checker, and resolve every blocking issue. A sample check is preparation only. Provide the complete declared review package and report; retain protected final-test data separately. The team then reviews the science, implementation and rights, and records explicit approval before public release. Original-paper citations and contributor credit accompany approved releases.

The links below provide the package guide, checks and contact route.

- Format and commands: https://rsi4science.org/downloads/submission-format.md
- Local browser check: https://rsi4science.org/check/
- Human review record: https://rsi4science.org/downloads/submission-review.md