Skip to Fulmerica links
Worth Building · Chapter 165 October 2026A draft for public use

How to build shared tools in the open

Most tools in this catalog will be built by small teams, civic technologists, and nonprofits with two or three engineers, encoding rules that change by place and month. Between 2023 and 2025 the volunteer brigade network lost its sponsor, Benefits Data Trust closed, and the federal digital team 18F was eliminated. This chapter applies the catalog's ideas to how shared tools are built, reviewed, governed, and handed on, as AI writes a growing share of the code. It maps ten territories with current facts, things to build, who is already building them, scores, and where a reader could start.

01The idea, in shared civic software

Shared tools that people can read, review, and carry on

A more capable system that leaves people more isolated and more dependent is a step backward, whatever its benchmark scores.

The things that keep a shared tool alive already have names. The maintainer who answers the issue on a Sunday. The reviewer who reads the change before it ships. The decision record that says why a rule was coded one way. The county volunteer who knows which pantry closed. Each is easy to skip with two people and a grant ending in June. Stack Overflow's 2025 survey found 84 percent of developers using or planning to use AI tools, and only 29 percent trusting their accuracy. The 18F team's nearly 11 years and more than 450 projects ended on 1 March 2025. The work ahead calls for rules written once, in the open. Each change should show who reviewed it and why, and any team should be able to pick up the tool when the first one stops.

SceneThe default pathWhat we could build instead
A state changes a ruleEach screener hard-codes the change in a private repository. Some are wrong for months. Nobody can see which version a household was judged by.One open, versioned rule file per state, with the citation and a named reviewer. Every helper app pulls the same file, and the change is a public record.
A pull request a model wroteSix hundred lines merge because the tests pass. Six months later nobody on the two-person team can explain them.Small changes, a named human reviewer, a note on what the model drafted, and a reviewer from a partner team for anything touching money or eligibility.
A design decisionSettled on a call and lost in chat. The next team argues it again or repeats the mistake.A one-page decision record in the repository, opened as a proposal, with the options, the choice, who decided, and a date to revisit it.
The funder leavesThe nonprofit closes, the servers go dark, and partner agencies lose the tool overnight.Code, configuration, and documentation are already public. A handoff plan written at launch names who can run it, and a partner forks it the same week.
A new county chapterVolunteers build their own resource list from scratch. It dies when the lead moves away.The chapter configures the shared core for its county, keeps records in a common format, and has a vote on the core's roadmap.
A security report on a FridayA lone maintainer has no policy, no help, and a new European law arriving with nobody to explain it.A security policy and a bill of materials sit in the repository. A steward organization answers, and the fix reaches every program on the core.
02Seven commitments, seven examples

What each commitment means when the subject is shared code

CommitmentIn building in the open, it meansA concrete instance
1 · Judge by what happens to peopleJudge a shared tool by households served correctly and by teams able to run it after the first team left. Stars and downloads are inputs.A rules engine reports how many screeners use each state's rule file and how often their answers match a reviewed test case.
2 · Keep people in chargeCode, configuration, and decisions are public. A named person reviews every change to rules that decide money or rights. Any partner can fork and leave with its data.A county running a shared directory exports every record in the open standard and moves to another host within a week.
3 · Build capability rather than dependenceEach project leaves its partners able to read, change, and run the tool. Documentation and decision records are part of the product.A legal aid paralegal updates a state's rule file and runs the tests without calling a vendor.
4 · Make slow harm visibleSurface the slow decay. A library with one maintainer. Review times stretching. Rule files months behind the statute. Directory records nobody has checked in a year.A dependency map showing that a dozen civic tools rest on one library with a single maintainer and a weak security score.
5 · Design for groupsThe team, the federation of chapters, and the partner agencies are the users. Governance is part of the design.A federation charter that gives each county chapter a vote on the core's roadmap and a right to fork with its records.
6 · Keep the effort that builds judgmentLet AI draft boilerplate, tests, and first translations of rules into code. Keep the human review, the partner's comment on a proposed decision, and the caseworker's account of what a rule means in practice.A rule drafted by a model and reviewed line by line by a policy analyst, with the draft, the edits, and the citation kept together.
7 · Name choices as choicesClosed code bought with public money, a single funder with no handoff plan, and unreviewed AI changes are choices. Name them, and point to the teams that chose otherwise.The SHARE IT Act and the EU's Interoperable Europe Act already tell public bodies to share code or prefer open solutions. A county can write the same rule into its next contract.

The three questions, applied

ThingMore capable or more dependent?Closer to each other or to the system?See, steer, and leave?
PolicyEngine's open rules engine and APICapable; any screener can run and inspect the same rules, and one screener built on it reports more than 65,000 families helpedCloser; several tools share one rule setOpen code that anyone can read, correct, or fork
A closed eligibility screener licensed to one agencyDependent; every rule change waits on the vendorCloser to the vendorRules hidden; leaving means starting over
An AI coding assistant with no second reviewerFaster at first; 66 percent of developers report answers that are almost rightCloser to the tool; the review conversation disappearsHard to see what changed and why
Open Referral's directory standardCapable; a small directory can exchange records with a 211 lineCloser; many directories share one formatOpen specification; records move to any system
OpenStreetMapCapable; 2.25 million people have contributedCloser; local mappers keep their own places currentPublic edit history, and the map can be copied
A volunteer network held by one sponsorCapable; brigade volunteers built local toolsCloser within each chapterAbout 60 chapters lost fiscal sponsorship in 2023 when the sponsor changed course
03Where to push

Where to push when building shared tools

Place to pushWhat it meansIn building in the open
ParadigmAI as a meansA civic tool is a public asset held in common. Its purpose is people served well by many teams over many years, including after any one funder leaves.
Goals and metricsCapability, trust, connectionTeams able to run a tool they did not write. Rule files current with the statute. Days from a security report to a fix. Chapters still active after five years.
Information flowsMake drift visible; the most practical leverDecision records in the repository. Security scores for every dependency. Which agencies listed and published their code. Which directory records are stale, by county.
RulesLegibility, portability, interoperability, exitOpen licenses in public contracts. Shared formats for directories and for eligibility rules. A bill of materials with every release. A handoff plan as a condition of each grant.
Self-organizationCooperatives, civic groupsVolunteer chapters with a charter and a vote. Maintenance cooperatives among teams that share libraries. Review exchanges among small teams. Foundations that serve as stewards under European law.
Why building in the open, specifically

This is the domain where the catalog's ideas apply to the people who build everything else. Linux and Wikipedia, the precedents the catalog cites, both run on a shared core, public decisions, and contributors who can see and steer the work. The Linux Foundation reported spending $8.4 million on the kernel in 2025, and a community vote that year filled 2 seats on the Wikimedia Foundation's board. OpenStreetMap reached 2.25 million contributors by mid-2025 the same way. Civic technology learned the opposite lesson the hard way. The brigade network depended on one sponsor until 2023, Benefits Data Trust on thinning grants until 2024, and 18F on one agency until 2025. Small teams cannot each afford security, review, rules research, and a succession plan. Together they can, when the core is shared, the decisions are written down, and every member can leave with its work.

04Ten territories

The map of building in the open

Each territory is scored on the catalog's five-part rubric: a shared layer nobody owns, a buyer or deadline, outcome data, the three questions, and fit. Each card names which of the nine reusable parts it builds on. Here the parts are also what these practices protect, because place configuration, standards cohorts, and outcome commons all fail when nobody can read, review, or inherit the code.

How to read a card: its score adds five judgments, each from 0 to 2 (shared layer nobody owns, buyer or deadline, outcome data, the three questions, fit), for a total out of 10.

B1Score 8 of 10Reuses: place config (rules as code), standards cohort, outcome commons

Open rules that every helper can run

The screeners, navigators, and legal aid tools that each encode the same eligibility rules, the states that change them, and the households who get a wrong answer when one copy goes stale.

Why now

PolicyEngine, a nonprofit, publishes a free, open rules engine for US and UK taxes and benefits, built on the OpenFisca framework. Its application programming interface (API) powers screeners including MyFriendBen and Amplifi. MyFriendBen's own account reports more than 65,000 families helped and more than $40 million in benefits since its 2023 launch. It opened in North Carolina as its second state and won a $2.4 million Gates Foundation grant to expand, with a goal of $100 million in benefits. Georgetown's Beeck Center published four experiments on 24 March 2025 that used AI to translate benefits policy into code.

Plays

  • A rule file per state and program with a version, an effective date, a citation to the statute or manual, and a named reviewer, which any screener can pull.
  • A shared set of test households, identifiers removed, that every engine must pass before a rule change goes live, with disagreements between engines published.
  • AI-drafted rule translations that a policy analyst reviews line by line, with the draft, the edits, and the citation kept in one decision record.

Already there

PolicyEngine, OpenFisca, MyFriendBen, Amplifi, and the Beeck Center's Digital Government Hub. Shared test households and a public review record for each rule change are missing.

Score

Shared layer 2 · Buyer or deadline 1 · Outcome data 1 · Three questions 2 · Fit 2

B2Score 9 of 10Reuses: standards cohort, place config, outcome commons

One format for every resource directory

The 211 lines, United Ways, clinics, and referral apps that each keep a list of the same pantries and shelters, and the people who call a number that no longer works.

Why now

Open Referral's Human Service Data Specification (HSDS) began in 2014 with Code for America as a co-sponsor. The Alliance of Information and Referral Systems, now Inform USA, endorsed it in 2018 as the industry standard for exchanging resource data. Version 3.0 moved the core format from comma-separated values (CSV) files to JavaScript Object Notation (JSON) and made the standard API-first, and version 3.1 has followed. Open Referral says Findhelp, United Way 211, and Unite Us have adopted it, and New York City's Office of Opportunity adopted it in 2018. The 211 taxonomy that labels services holds more than 9,000 defined terms.

Plays

  • A free validator and conformance badge for HSDS 3.x feeds, run monthly for any directory, with results public.
  • A standards cohort that funds five small community directories to publish HSDS feeds, then wins a 211 or a county as buyer of a shared regional feed.
  • A correction loop. When a caller finds a closed pantry, the fix reaches every directory that holds the record, and the count of stale records by county is public.

Already there

Open Referral, Inform USA, United Way 211 lines, Findhelp, Unite Us, and county human services departments such as Allegheny County's. A shared, public freshness record is missing.

Score

Shared layer 2 · Buyer or deadline 1 · Outcome data 2 · Three questions 2 · Fit 2

B3Score 8 of 10Reuses: standards cohort; new: decision log

Decisions written down where people can see them

Small civic teams, their partner agencies, and the next maintainer, who inherits code with no record of why it works the way it does.

Why now

A study at the 2026 International Conference on Software Architecture found that architecture decision records (ADRs) are still rare in open source but more common each year. About half of repositories that keep them hold 1 to 5 records. Of about 5,800 records studied, about 3,674, or 63 percent, were opened already marked accepted, which suggests the discussion happened elsewhere. Missouri's transportation department published ADR 0001 to adopt the practice for its engineering policy guide. The Office of Management and Budget (OMB) memo M-25-21 of 3 April 2025 required agencies to post AI strategies publicly within 180 days.

Plays

  • A decision record template for civic tools that names who is affected, which partner was consulted, and when the choice will be revisited, written so a caseworker can read it.
  • A proposal-first workflow, where the record opens as proposed and partners comment before acceptance, following open projects such as pygeoapi's request-for-comments process.
  • A public index across civic projects, so a team choosing an eligibility engine or a login provider can read what ten others decided and why.

Already there

ADR practice in open source, pygeoapi's request for comments (RFC) process, Missouri's transportation department, and agency AI strategies. A shared civic index does not exist.

Score

Shared layer 2 · Buyer or deadline 1 · Outcome data 1 · Three questions 2 · Fit 2

B4Score 7 of 10Reuses: standards cohort; new: review exchange

Reviewing code a machine wrote

Two- and three-person teams at nonprofits and agencies who write much of their code with AI assistants and have nobody else to review it.

Why now

Stack Overflow's 2025 survey of 49,009 developers in 166 countries, fielded from 29 May to 23 June 2025, found 84 percent using or planning to use AI tools, up from 76. Trust in their accuracy fell to 29 percent from 40, and 66 percent cited answers that are almost right. GitHub reported on 28 October 2025 that nearly 80 percent of new developers used Copilot in their first week, and merged pull requests rose 23 percent to 43.2 million a month. The 2025 DORA report, from Google's DevOps Research and Assessment team, linked AI adoption to higher throughput and still to higher instability. Microsoft's chief executive said in April 2025 that 20 to 30 percent of code in its repositories was written by AI.

Plays

  • A review exchange where small civic teams swap reviewers for changes touching eligibility, money, or personal data, with a shared checklist and a record of who reviewed.
  • A change note convention that says which parts a model drafted, which tests a person wrote, and what the reviewer checked.
  • A size limit for AI-drafted changes to rules code, so a person can actually read each one.

Already there

DORA's research, GitHub's review tools, and the review norms of large open projects. A cross-team review exchange for civic code does not exist.

Score

Shared layer 1 · Buyer or deadline 1 · Outcome data 1 · Three questions 2 · Fit 2

B5Score 8 of 10Reuses: outcome commons, member-owned data, standards cohort

Paying the people who keep shared code alive

The maintainers of the libraries civic tools depend on, often one or two people, and the public funders who have begun to treat their work as infrastructure.

Why now

Germany's Sovereign Tech Fund reported investing €33.6 million in 96 technologies by late 2025, with a 2025 budget of about €19 million against €17 million in 2024. Its 2025 fellowship pilot paid 6 maintainers for a year across 18 critical technologies, and the 2026 round, open to community managers and technical writers, closed on 6 April 2026. The National Science Foundation's Pathways to Enable Open-Source Ecosystems (POSE) offered up to $300,000 for Phase I and $1.5 million for Phase II in 2025. A grants summary reports that its successor, PESOSE, offers up to $40 million for 40 to 60 awards, due 1 September 2026. The Open Source Security Foundation's Scorecard, launched in November 2020, scores projects from 0 to 10 on automated checks.

Plays

  • A dependency map for civic tools that shows which libraries the catalog's builders share, their Scorecard results, and how many people maintain each.
  • A pooled maintenance fund among civic teams that rely on the same libraries, with spending decided by members and published.
  • Joint fellowship and grant applications for the maintainers of civic libraries, such as rules engines and directory validators.

Already there

The Sovereign Tech Agency, the National Science Foundation, the Open Source Security Foundation, the Linux Foundation, and GitHub Sponsors, where PolicyEngine has a page. Civic dependencies have no map.

Score

Shared layer 2 · Buyer or deadline 2 · Outcome data 1 · Three questions 1 · Fit 2

B6Score 8 of 10Reuses: deadline engine, evidence file, standards cohort

The steward duty and the bill of materials

Small open-source projects and the nonprofits that steward them, which European law now names, and the agencies that buy software with a list of its parts.

Why now

Reporting duties under the European Union (EU) Cyber Resilience Act began on 11 September 2026, and the law applies in full from 11 December 2027. It creates a lighter role for open-source stewards, who need a documented security policy, vulnerability handling, and cooperation with authorities, and who face no fines. The Cybersecurity and Infrastructure Security Agency (CISA) released draft minimum elements for a software bill of materials (SBOM) on 22 August 2025. The draft adds component hashes, licenses, tool names, and generation context, and comments closed on 3 October 2025. CISA's Secure by Design pledge grew from 68 companies at launch in May 2024 to more than 250 by the end of 2024 and 367 listed signers in 2025, according to a security firm's count.

Plays

  • A steward kit for small projects with a security policy, a disclosure inbox, an SBOM generated in every build, and a plain checklist for the 2027 obligations.
  • A shared steward organization that several civic projects join, so one responder handles reports for all of them.
  • A bill of materials attached to every release of shared civic tools, so a county can see what it is installing and what needs an update.

Already there

The Open Source Security Foundation's work on the Cyber Resilience Act, CISA's guidance and Scorecard listing, and foundations that act as stewards. A steward service sized for civic projects is missing.

Score

Shared layer 2 · Buyer or deadline 2 · Outcome data 1 · Three questions 1 · Fit 2

B7Score 9 of 10Reuses: standards cohort, records pipeline, place config

Public code, and procurement that buys open

Agencies that pay for software, the vendors and nonprofits who write it, and the residents whose rules it encodes.

Why now

The SHARE IT Act, Public Law 118-187, was enacted on 23 December 2024. It requires federal agencies to share custom-developed source code across government and publish lists of it, building on the 2016 Federal Source Code Policy. OMB memo M-25-21 of 3 April 2025 extended sharing to custom AI code and models, and M-25-22 the same day set rules for buying AI. The EU's Interoperable Europe Act, Regulation 2024/903, tells public bodies to prefer solutions without restrictive licenses, such as open source, when they are otherwise equivalent. The European Commission published an open-source roadmap in July 2025, and its open-source strategy proposes procurement reform and support for maintainers. The Digital Public Goods Alliance counted more than 220 verified digital public goods and 50 members in its report of 11 February 2026.

Plays

  • Model contract clauses for state and county software. Code delivered under an open license, data exportable in named formats, decision records and tests included, and a handoff plan.
  • A tracker of federal code inventories under the SHARE IT Act, showing what each agency listed and what it published.
  • A buyer's guide that matches common civic needs to verified digital public goods, and the place configuration each needs.

Already there

The Digital Public Goods Alliance, the EU's Open Source Observatory and its procurement handbook, federal code-sharing rules, and European campaigners for public money, public code. Model clauses sized for counties are missing.

Score

Shared layer 2 · Buyer or deadline 2 · Outcome data 1 · Three questions 2 · Fit 2

B8Score 9 of 10Reuses: place config, standards cohort, member-owned data, outcome commons

Many small programs on one core

Volunteer chapters, county programs, and small nonprofits that each serve one place, and the shared core that lets them run the same tool with local settings.

Why now

Code for America ended fiscal sponsorship of roughly 60 volunteer brigades in February 2023, and their contracts ended that June. The change followed a $100 million gift in 2022 for its Safety Net Innovation Lab. Former brigades regrouped on their own. OpenStreetMap has 10 million registered users and 2.25 million contributors as of mid-2025, and about 50,000 people mapped in an average month of 2025, according to the Heidelberg Institute for Geoinformation Technology (HeiGIT). Wikipedia's community vote filled 2 Wikimedia Foundation board seats in 2025.

Plays

  • A core and configuration design. One maintained codebase, one configuration file per place, and a chapter starter kit, so a new county launches in weeks.
  • A federation charter that says what chapters own, what the core owns, how chapters vote on the roadmap, and how any chapter forks and leaves with its records.
  • A shared outcome record across chapters, so the federation can show funders what many small programs did together.

Already there

OpenStreetMap's local chapters such as OpenStreetMap US, Wikimedia's affiliates, former brigades, and Open Referral's network. A charter template for civic software federations is missing.

Score

Shared layer 2 · Buyer or deadline 1 · Outcome data 2 · Three questions 2 · Fit 2

B9Score 8 of 10Reuses: evidence file, place config, member-owned data

When the funder leaves

Partner agencies and residents who rely on a civic tool run by one nonprofit or one federal team, and the people who would have to pick it up.

Why now

Benefits Data Trust closed in August 2024 and laid off 273 people, after serving more than 120,000 people in 2023 with partners including Pennsylvania's human services department and the City of Philadelphia. The General Services Administration (GSA) eliminated 18F on 1 March 2025, dismissing about 90 employees after nearly 11 years and more than 450 projects. The U.S. Digital Service was renamed the U.S. DOGE (Department of Government Efficiency) Service on 20 January 2025. GSA launched USAi on 14 August 2025, offering agencies chat, code generation, and summarization at no cost. California retired GetCalFresh, used by more than 7 million Californians over a decade, in phases by September 2025 as its BenefitsCal took over. A new agreement keeps Code for America running digital outreach after 30 September 2025.

Plays

  • A handoff plan template written at launch. Who can run the tool, where code and data live, how partners are told, and what costs continue.
  • Code, configuration, and documentation public from the first day, so a partner can fork within a week.
  • A register of civic tools at risk, with each one's funding horizon and a named successor.

Already there

The 18F alumni site, Code for America's transition work with California, and the Digital Public Goods Alliance's standard. Nobody keeps the register.

Score

Shared layer 2 · Buyer or deadline 1 · Outcome data 1 · Three questions 2 · Fit 2

B10Score 7 of 10Reuses: outcome commons, standards cohort; new: decision log

The nonprofit's AI decisions, made together

Nonprofit staff who already use AI tools, the executives who set policy, and the clients whose records pass through those tools.

Why now

A survey by NTEN and Bridgespan of 917 respondents, fielded from late April to mid-June 2026, found 98 percent of nonprofits using AI in some way and 61 percent with official use. That official use splits into 33 percent in pilots, 24 percent team-wide, and 4 percent fully integrated. Among executives, 63 percent reported no AI-specific funding of any kind, and the survey found 37 percent with no training and 58 percent with no roadmap. Executives were nearly twice as likely as staff to strongly agree that AI held untapped opportunities, 44 percent against 23. The San Francisco Fed published a study of early AI adoption in community development in March 2026.

Plays

  • A library of AI use policies written as decision records with staff and published, so the next nonprofit adapts one instead of starting blank.
  • A staff-led review circle that tests a tool on real casework before adoption and records what it got wrong.
  • An outcome commons of AI tool trials across nonprofits, consented and anonymized, by task and result.

Already there

NTEN, Bridgespan, and the San Francisco Fed's community development team. The shared policy library and trial record are missing.

Score

Shared layer 1 · Buyer or deadline 1 · Outcome data 2 · Three questions 2 · Fit 1

Open in name only, and what not to build

Publishing code is the easy part. A repository with one maintainer, no decision records, no tests a stranger can run, and a sponsor who holds the only deploy keys is open in name only. The pattern worth building runs the other way. The rule is a public file with a reviewer. The decision is written down before it is final. The core belongs to its members, and each one can leave with its records.

05An ordinary day

A Thursday in spring 2027, when shared tools work as they should

In Raleigh a policy analyst at a benefits nonprofit drafts next month's change to a state rule with an AI assistant. She opens a decision record marked proposed, cites the manual section, and runs forty test households. A reviewer on a partner team in Denver reads the change line by line that afternoon, catches one wrong threshold, and approves the fix. By evening two screeners in two states have pulled the new version. In Pittsburgh a 211 call specialist learns that a pantry closed. She marks the record, and the correction reaches the county directory and a clinic's referral app before the next caller. In a small Oregon town a volunteer chapter launches the shared resource tool for its county with a configuration file and a starter kit. The federation's chapters voted on this year's roadmap a month ago. In Berlin a funded maintainer fixes a flaw in a library a dozen civic tools rely on, and each county's bill of materials shows whether it needs the update. At a nonprofit in Fresno, staff finish a month-long trial of an AI note-taker, record what it got wrong, and publish their decision to wait. Nobody waits on a vendor, and nobody's work disappears when one grant ends.

Signs we are getting there

  • Each state's main eligibility rules exist as an open, versioned file with a citation and a named reviewer, and at least three screeners pull from the same file.
  • The number of community resource directories publishing HSDS 3.x feeds is counted publicly, and stale records by county fall year over year.
  • Fewer than a third of new decision records in civic repositories are opened already accepted, against 63 percent in the 2026 open-source study.
  • Every shared civic tool in the catalog ships a bill of materials and a security policy, and has a named steward, before 11 December 2027.
  • No civic tool closes without a published handoff plan, and the register of tools at risk names a successor for each.
06Where you could start

A first step, sized to the time you have

A weekend

  • Check ten local food and shelter listings against your 211 and your county's directory, and file the corrections in HSDS format (B2).
  • Run the Open Source Security Foundation (OpenSSF) Scorecard on three libraries that a civic tool you use depends on, and add the results to the dependency map (B5).
  • Write the first decision record for a project you maintain, explaining one choice already made and when to revisit it (B3).
  • Test one state rule in PolicyEngine against a real case you know, identifiers removed, and file an issue where the answer disagrees (B1).

A season

  • With one legal aid office or benefits nonprofit, convert ten eligibility rules into open, versioned files with citations and forty test households, reviewed by a policy analyst (B1).
  • With two small civic teams, run a review exchange for ninety days on every AI-drafted change, and measure defects found and days to review (B4).
  • With one 211 and three community directories, publish HSDS feeds and measure duplicate and stale records before and after (B2).
  • With one county procurement office, adapt the model open-code clauses into one live contract and publish the result (B7).

A career

  • Convene the federation charter and shared core for a family of county programs, and carry it to a second and third state (B8).
  • Lead a steward organization that holds the security policy, bills of materials, and disclosure duty for a dozen civic projects ahead of December 2027 (B6).
  • Found a member-owned maintenance cooperative among civic teams that pay jointly for the libraries they share (B5, B9).
  • Run the register of civic tools at risk and broker successors before closures (B9).

Who is already doing it

  • PolicyEngine maintains an open rules engine for US and UK taxes and benefits, with an API that powers MyFriendBen and Amplifi.
  • MyFriendBen runs a benefits screener built on PolicyEngine in Colorado and North Carolina.
  • Open Referral stewards HSDS, the open format for community resource directories.
  • Inform USA, formerly the Alliance of Information and Referral Systems, endorsed HSDS in 2018 as its industry standard.
  • The Sovereign Tech Agency, a German public agency, funds the maintenance of open digital infrastructure and pays maintainers through its fellowship.
  • The Open Source Security Foundation runs Scorecard and works with projects on the Cyber Resilience Act.
  • The Digital Public Goods Alliance verifies open digital public goods and counted more than 220 in February 2026.
  • The Beeck Center at Georgetown runs experiments on rules as code and the Digital Government Hub.
  • NTEN supports nonprofit technology staff and surveyed nonprofit AI use with Bridgespan in 2026.
  • OpenStreetMap US supports the American mapping community and publishes an annual report on it.
07Scorecard

Ten territories on one page

Score key: SL shared layer nobody owns, BD buyer or deadline, OD outcome data accrues, 3Q the three questions, FT fit for a builder who convenes. Each is scored 0 to 2, so the total is out of 10.

CodeTerritoryShared layerBuyer or deadlineOutcome dataThree questionsFitTotalHorizon
B2One format for every resource directory2122292027 cohort with one 211
B7Public code and procurement that buys open221229The next contract cycle
B8Many small programs on one core212229Charter in 2027
B1Open rules that every helper can run211228Now, with existing engines
B3Decisions written down in public211228Now, in every repository
B5Paying the people who keep shared code alive2211282027 funding rounds
B6The steward duty and the bill of materials221128Before 11 December 2027
B9When the funder leaves211228At every launch
B4Reviewing code a machine wrote111227Now, between two teams
B10The nonprofit's AI decisions, made together1122172027 with nonprofit networks
08Picks and declines

Six plays to pursue first, and five to decline

  1. Many small programs on one core (B8). The catalog's tools will be run by many small groups in many places. Write the federation charter and the core and configuration design before the second chapter launches, so nobody repeats the brigade network's dependence on one sponsor.
  2. One format for every resource directory (B2). The standard exists, the industry body endorsed it in 2018, and the gap is a validator and a correction loop. Start with one 211 and a handful of small directories.
  3. Public code and procurement that buys open (B7). Federal and European law already lean toward shared code. Model clauses that a county can paste into its next contract carry the same rule to the places most civic software is bought.
  4. Open rules that every helper can run (B1). The engines exist. Shared test households and a public review record for each rule change make them trustworthy enough for legal aid and agencies to rely on.
  5. Decisions written down in public (B3). The cheapest practice in the chapter. A template, a proposal-first habit, and a shared index turn private reasoning into something the next team can inherit.
  6. The steward duty and the bill of materials (B6). The European deadline of 11 December 2027 gives small projects a reason to adopt a security policy and a bill of materials now, and a shared steward makes that affordable.
A closed rules engine licensed state by state. Eligibility rules are public law. Selling them back to agencies and nonprofits as a hidden service makes every helper dependent on one vendor's reading of the statute.
An AI code generator that ships civic changes without a second reader. Developers already distrust these answers more than they trust them. Speed without review moves the cost onto the household that gets the wrong result.
A national directory that copies local records and sells referrals. It pulls the work of 211 lines and volunteers into a private database. The shared format and correction loop keep records with the places that maintain them.
A repository dump at shutdown. Publishing code the week a team closes, with no documentation, tests, or decision records, looks open and helps nobody pick the work up.
A federation whose core holds the chapters' money and data. The brigade network showed what happens when one sponsor changes course. Chapters need their own records, a vote, and a way to leave.
Sources for this chapterChecked 5 October 2026

Assumptions and limits. Scores are judgments. Figures were checked on 5 October 2026 against web search results, because this environment could not open source pages. Each figure is attributed to the organization that reports it, and readers should confirm it on the linked page. The chapter rests on 19 web searches across two sessions, 15 earlier and 4 in this one. Several figures are organizations' own accounts. These include MyFriendBen's families helped and benefits, Open Referral's list of adopters, the Sovereign Tech Fund's totals, and GitHub's Octoverse numbers. The Secure by Design signer count comes from a security firm, the PESOSE amounts from a grants summary site, and Microsoft's 20 to 30 percent from a secondary site quoting its chief executive. The Sovereign Tech Fund's 2025 budget of about €19 million is as reported by heise. Release dates for HSDS 3.0 and 3.1 could not be verified in this pass and are not stated. An industry-wide share of code written by AI could not be verified, since company figures use different definitions. Counts of projects at the Apache and Linux foundations, OpenFisca's government users, U.S. DOGE Service staffing after 2025, and the number of registered open-source stewards under European law could not be verified and are omitted. The nonprofit survey reports some figures for executives and some for all respondents, and the split was not fully clear from search summaries. An independent spot-check on 5 October 2026 used 7 more searches and relied on search results, because source pages could not be opened. It corrected 18F's staff to about 90 and its span to nearly 11 years. MyFriendBen's counts of 65,000 families and $40 million could not be found in search results, which quoted higher, undated counts of more than 120,000 families and $60 million. A figure of 57 percent of executives with no AI budget could not be verified, so the chapter gives the survey's 63 percent with no AI-specific funding instead.