Drafts: OpenAI Open Source Fund + Claude for OSS applications, plus funding-program landscape

[tonyketcham]

Two @flatbread/proof DAG runs produced four review-ready Markdown artifacts; a third revision DAG then compressed the roadmap and rebalanced both drafts toward audacious bets. Nothing here is submitted — everything is a draft for maintainer review.

What landed

Path	Purpose
funding-applications/openai-open-source-fund.md	Full Codex OSS Fund application response, form-field-mirrored, with budget table, revised 4-phase roadmap (foundations + Effort Graph compressed into months 1-4; audacious bets occupy months 5-12), and > NOTE TO REVIEWER: callouts on every UNVERIFIED form field.
funding-applications/claude-for-oss-brief.md	Sales-intake brief for Anthropic's Claude for OSS funnel — eligibility checklist, revised monthly token-volume projection (~80-160M input / ~16-32M output, ~2-3x prior estimate) tied to continuous preset/eval workloads, and MCP / Claude Code / HITL / evals positioning.
funding-applications/REVIEW-CHECKLIST.md	Pre-submission self-assessment with a new top-of-file revision summary, updated 1–5 scoring (technical specificity raised; evidence-of-traction due-diligence cost flagged), refreshed UNVERIFIED list, cross-draft consistency check, and recommended submission order.
funding-research/funding-program-landscape.md	Independent landscape research — TL;DR top-5 ranked table, full per-program coverage across AI-lab, foundation, infra/credits, and academic/research categories, eligibility-blocker map, stacking strategy, and ordered application backlog.

Audacious-bet shape adopted in both drafts

• Phase 1 (months 1–2) — Foundations, compressed. Typed defineConfig, ID normalization, relation validation, watch-mode parity — one umbrella PR train under Codex/Claude review.
• Phase 2 (months 3–4) — Effort Graph MVP. Conventions preset, schema-validated Append API, flatbread-mcp server (read + append) for Codex / Claude Code / Cursor.
• Phase 3 (months 5–8) — Bet A: Workflow Presets for Complex Projects. Six shipped presets — schema-cutover, release-train, research-compendium, docs-site-refactor, api-version-cutover, design-system-token-rotation — each a parameterized DAG over the Effort Graph + @flatbread/proof. Seventh slot held for a community-contributed preset by month 8.
• Phase 4 (months 9–12) — Bets B + C in parallel.
  • B. HITL ergonomics. needsApproval boundary on every DAG node, Claude-Code-style plan-review gate on Decision/Plan, LangGraph-style durable pause/resume keyed to a thread_id Session checkpoint.
  • C. Continuous-improvement evals + research loop. fixture-promote CLI, PR-time regression-replay GitHub Action, public Inspect-View-style dashboard, eval-driven preset retuning so the catalog self-tunes.

OpenAI budget rebalanced inside the published $25k cap: Phase 3 presets ($8k, largest line), evals loop ($5k), HITL ($3.5k), Effort Graph + MCP ($3k), foundation Codex toil ($2.5k), docs/cookbook/contributor sponsorship ($3k).

How it was generated

• DAG GraphQL connections #1 (/tmp/dag-oss-fund-drafts.json) — 7 tasks, 3 ranks: 4 parallel research tasks → 2 parallel draft writers → 1 review-checklist task. 7/7 in 5m 59s.
• DAG Only select and read markdown files for query parsing #2 (/tmp/dag-funding-research.json) — 6 tasks, 2 ranks: 5 parallel research tasks → 1 synthesis task. 6/6 in 10m 25s.
• DAG chore: prettier global config & husky enforcement #3 (/tmp/dag-revise-drafts.json, this revision) — 7 tasks, 4 ranks: 3 parallel research/state-summary tasks → 1 redesign synthesis → 2 parallel draft revisers → 1 review-checklist update. 7/7 in 25m 13s.

All three DAGs ran via pnpm exec proof (the @flatbread/proof package shipped from this repo) using the proof skill workflow. Live canvases were rendered to ~/.cursor/projects/workspace/canvases/ for the maintainer to scrub progress in real time.

What the maintainer needs to do before submission

The REVIEW-CHECKLIST file consolidates the action list. High-leverage items:

1. Fill placeholders flagged in the OpenAI draft: LinkedIn URL, primary GitHub handle (org FlatbreadLabs vs personal), confirm ketcham.dev@gmail.com is the canonical contact.
2. Confirm program form-field text against the live forms (the playbooks captured the current published fields; OpenAI in particular updates Typeform fields without notice).
3. Confirm whether Claude for OSS permits API credits for an automated eval/preset DAG harness, or only Max-seat usage — the brief is written to accept either shape but the ask line should be tightened once known.
4. Decide whether to register a fiscal sponsor (Open Source Collective is the recommended on-ramp — it unlocks Sentry / NLnet / thanks.dev / and several rows in the landscape).
5. Stress-test the revised Claude monthly token-volume math against actual recent @flatbread/proof usage; the projection assumes continuous preset/eval workloads.
6. Sanity-check the Phase 3 preset list (schema-cutover, release-train, research-compendium, docs-site-refactor, api-version-cutover, design-system-token-rotation) — these names appear in both drafts so they must read as committed product surfaces.

Out of scope

• No source-code changes. No test changes. No build / lint / typecheck delta — only Markdown additions and revisions under two new top-level directories.
• Applications are not submitted anywhere; this PR exists so the artifacts are reviewable in normal PR flow before the maintainer sends anything.