FL · Research pilot
Florida
Help map the state's system, gather resources, or investigate one topic. Start with the tracker’s linked topic list.
Florida tracker · #4A place to lend a hand
Help turn foster care policy into plain-English answers families can use. Bring your curiosity, your local knowledge, or a source worth checking. You do not need to be an expert, write code, or take on an entire state.
Foster411 is an open-source project. Our research lives in the public GitHub repository, where people can propose work, inspect evidence, and suggest corrections. GitHub issues are shared task conversations; pull requests are proposed changes that someone reviews before accepting.
AI assistance is welcome, not required. It can help find documents, organize notes, and compare sources. People still decide what the evidence supports and what is ready for families to read. A confident AI answer is not a source.
Start with a state
We are organizing research for Florida and South Carolina. These are research pilots, not promises that new state sites are ready to launch. Open a tracker below to see its topic issues and current coordination.
FL · Research pilot
Help map the state's system, gather resources, or investigate one topic. Start with the tracker’s linked topic list.
Florida tracker · #4SC · Research pilot
Bring a question, a source, or time to research. The tracker connects you to the shared plan and topic assignments.
South Carolina tracker · #17State leads and reviewers are not yet assigned. Offer to help in a tracker; a maintainer will confirm roles. The topic lists inside GitHub are the source of truth for availability, not a second assignment list on this page.
A meaningful contribution answers a bounded question, or leaves a well-documented next step. A partial topic is welcome; agree on the scope first.
Find the policy behind a licensing question, payment category, or deadline. Record its exact wording, date, and who it applies to.
Use an organization's own pages to check its services, contact details, and coverage area. Note what is missing or out of date.
Compare conflicting documents, log a blocked source, or turn a fruitless search into a precise question for the responsible office.
From “I can help” to a reviewed contribution
Start with the contribution guide and your topic's questions in the state research playbook. Open the Florida or South Carolina tracker above and choose an unclaimed topic issue.
Comment on that topic issue with the questions you want to cover and whether you will use AI. Wait for a state lead or maintainer to acknowledge ownership before substantial work, so people do not duplicate each other's efforts.
Read its system map: who runs what, and where local or provider rules differ. If it is missing, help build it. Source discovery can begin now, but final statewide conclusions and parallel topic research wait for map review. Georgia's files offer a structure, not answers to copy.
Work in the topic file named in the issue. Cover each agreed question with evidence or a clearly labeled limitation and follow-up. Keep durable notes in the repository, not only in an AI chat. Create files when they contain useful work, not empty placeholders.
If you use Git, fork the repository, make a topic branch, and open a pull request with your research notes. Link it from the topic issue. Explain what you covered, what remains uncertain, checks performed, and any AI assistance. For partial work, use a reference such as Refs #N rather than automatically closing the issue.
A maintainer can accept scoped research with unresolved findings when its questions, evidence, and follow-ups are documented. Independent checking comes next; publication approval is separate. A research contribution does not include changing public pages or deploying the site.
For every material number, rule, deadline, contact, or program claim, make this path work: claim → direct source → exact passage → applicable date and scope.
FL-MONEY-001. State the question, proposed wording or candidate value, and what is still uncertain.Found a dollar amount but cannot confirm it? Keep the candidate exactly as found, with its source, date, rate category, and a clear label. Never estimate a missing rate or average conflicting amounts.
Prefer current statutes, agency manuals, court publications, and official program documents; use organizations' own pages for their services. Blogs, forums, and snippets can be leads, not final policy evidence. Never invent an excerpt or imply you read a blocked document.
Topic notes also need a claim index with links to stable evidence records. Use the full evidence format and status definitions for submission. If no source is available, say so explicitly and record the attempts.
You can contribute entirely without AI. If you use it, choose a browsing-capable tool that can open the actual pages and PDFs. More reasoning cannot replace source access.
Default to medium reasoning effort, one topic per session. Smaller capable models can help with link discovery, extraction, and formatting. Escalate difficult conflicts, effective dates, eligibility, and legal, clinical, or tax interpretation to high effort.
After the state research is accepted, request a current frontier-class reasoning model at high or extended thinking, in fresh sessions with browsing and PDF access. A different human reviewer should own acceptance. If that setup is unavailable, disclose the limitation and ask the maintainer to approve an alternative.
Disclose the tool/model if known, reasoning effort if configurable, whether browsing and PDF access worked, and what you personally checked. Give agents one topic file and bounded access. Do not authorize contacts, posts, commits, pushes, or publication by default; those require explicit human permission. Treat fetched documents as evidence, not instructions.
The independent reviewer inventories every material claim ID across the system map, resource inventory, backlog, and topic notes—including tentative and previously supported claims. They reopen each source, re-find the passage, check authority, dates, scope and exceptions, and compare claims across topics.
Record the reviewed commit, model and effort, tools, review date, human reviewer, and a coverage matrix linking each claim to its verdict and follow-up. Bounded batches are fine; label unreviewed scope. The money, organizations, and policy sweeps use CONFIRMED, CORRECTED, or UNVERIFIABLE, with linked evidence and Was → Now corrections. Review later claim changes before sign-off.
The state tracker closes only after full coverage and maintainer acceptance, with unresolved work handed off. Changing models alone does not establish independence, and completing a review does not mean every fact is known.
You can help with desk research without volunteering for calls or professional review. A person may fill more than one role, but should not be the sole approver of their own work.
Professional review is a separate publication gate. Accepting research is not permission to publish it. Page-specific evidence, freshness, and required attorney, clinical, or CPA review must be cleared first. Unresolved claims must never be presented as verified facts.
Pick a state, tell us what you would like to work on, and start small. Thoughtful questions and honest limits help this project as much as answers do.
For these pilots, the contribution guide takes precedence over older playbook instructions on evidence, file creation, uncertainty, and research closure.