Fundraising Pipeline

ONE JOB: review the finished Jenga Fund deck, make one tiny live edit to prove the slides are native HTML, and point anyone who wants to build a deck from scratch to the Learn Claude recording. FRAME (set this early): this is NOT a build-a-deck chapter. The deck is already done. The point is to show what you are working with, a deck that is native HTML, not a binary PowerPoint, so it is editable as code, version-controllable, and exports clean later. - Jenga Fund is a demo NGO we invented to make the workflow realistic: it backs East Africa's AI builders (jenga = build in Swahili). - Stage: pre-pilot, idea stage. No live program or traction yet, by design. - Building a deck from scratch is its own skill. Send them to the Learn Claude webinar for that, do not teach it here. MUST LAND (proof gates): - Start from the V0 (starting) download of the starter files, sitting in a real folder on your machine - Codex opened in that folder and asked to survey what is there - The finished Jenga Fund deck open in the live app preview, five slides walked - One tiny edit made live (a word, a number, an accent color) and the slide updates instantly - Said out loud: these slides are HTML, not a .pptx binary, and we built this deck with Codex - A clear callout to the Learn Claude recording for building a deck from scratch ANCHOR FACTS (real, pull in if useful): - Get the files: download the V0 set from the Starter files panel (the V0 / Final toggle), drop it somewhere comfortable (Desktop works) - Deck: Jenga Fund overview, five slides (Fellows, the Gap, the Model, What Fellows Ship, Funding Logic) - It is a native Astro page at /codex/fundraising-toolkit/jenga-fund-overview, edited as code - Native HTML means: live preview, version control, and a clean PPTX export when you need to send it outside the repo - This is the same "source of truth, many shapes" idea the budget packet uses later OPEN (decide before recording): - Which tiny edit to demo (change the cohort number, a headline word, an accent color) - Exact title and link for the Learn Claude recording to call out - Show the V0 download happening on camera, or start with the folder already on the Desktop - FLAG: beat 5 dictation was cut off ("Add some small ..."), confirm the live-edit beat and whether a closing Learn Claude beat should change BEATS (~4 min, riff freely): 0:00 Download the V0 starter files, drop them somewhere comfortable (Desktop for me) 0:30 Open Codex in that folder, ask it to survey the files, let it orient 1:15 Focus on the Jenga deck, open it in the app browser preview, frame the demo org (Jenga Fund, East Africa) 2:00 Walk the five slides fast, then say how we built it with Codex 2:45 Add one small edit live (a word, a number, an accent color), watch the slide update, "these are HTML, not a binary" 3:15 Callout: want to build one from scratch? Watch the Learn Claude recording 3:40 Close: deck is ready, now the real work begins, funders, Step 02 TOTAL: recorded at ~10 min (matches the card's "Watch 10 min")

ONE JOB: find real, current funders that fit Jenga, turn them into a targets workbook, capture the actual portal questions, then let the agent interview you and draft the high-signal answers straight into the workbook, all in one continuous pass. FRAME: deck is done, now the money. This used to be three chapters (research, then workbook plus question index, then drafting); we run it as one flow because it is one file growing. Find who funds this, organize them, capture what each portal asks, then draft the answers that count. You capture and draft here; cheap questions get parked. - Jenga's profile is specific (East African AI builders, youth livelihoods, ~$60K pilot, ~$100K raise, pre-pilot). A good search respects eligibility walls, it does not just list big names. - The drafting is the opposite of "write my application for me": the agent pulls your real thinking out one question at a time, grounded in your design. MUST LAND (proof gates): - Funders match the profile, not a generic top-ten list; sizes and deadlines captured; sources cited (live links) - Eligibility walls flagged out loud (do not pretend everything fits) - Targets workbook: focus funders on top, status and deadlines color-coded, addressable pool assessed honestly - Questions captured from live portals, not guessed, tagged by funder, high-signal vs cheap split - Agent interviews you one question at a time; high-signal answers drafted into the bank, grounded in your real design, not invented - Workbook saved into the source pack ANCHOR FACTS (real): - Strong fits: Anthropic Economic Futures (research grant + Claude credits), DRK (up to $300K unrestricted, post-pilot), PJMF (invitation-only), Mastercard Young Africa Works (bullseye geography) - Walls to name out loud: OpenAI People-First Fund is US 501(c)(3) only (Jenga not eligible now); PJMF invite-only; DRK post-pilot; Echoing Green 2026 cycle closed, reopens ~Sept 2026 - Open and takes pre-launch: D-Prize Global Competition (register Jun 7 2026, extension Jun 28) - Two tiers: FOCUS (Anthropic, DRK, PJMF, Mastercard, OpenAI) and Other (Google.org, NextLadder, Skoll, Echoing Green, D-Prize, Tony Elumelu); the truly open-now set is small - One file, three tabs: Targets, Portals and Questions, Question Bank (jenga-fund-2026-fundraising-targets.xlsx); DRK's live form is verified (safest to open on camera) - High-signal examples (DRK): the social problem, the solution, how you improve on existing efforts, what is unique and systemic. Low / cheap: org age, greater-good yes/no, geographic-impact yes/no - Draft answers grounded in the real design: NEET youth with laptops but no on-ramp; self-paced cohort measured by a verified first $100 earned (not attendance); merit-tiered seats ($20 / $100 / $200), capital-efficient and self-reinforcing OPEN (decide before recording): - How many funders to surface live vs reference the prepared list; show the citations on screen - Which portal to open live (DRK is verified, safest); color scheme for status (open / verify / closed) - Run the interview live (more honest, riskier) or pre-run one; how many answers to draft on screen BEATS (riff freely, this is the long one): 0:00 Pivot: deck is done, now who funds this; state Jenga's profile so the search has a target 2:00 Run the search, narrate the first real hits, read sizes and deadlines (D-Prize, Echoing Green) 4:00 Flag an eligibility wall out loud (OpenAI US-only); show the citations, why sourcing matters 6:00 Turn the shortlist into the workbook: focus funders up top, color the status, name the gated ones 8:00 Switch tabs: open a live portal (DRK), pull the real questions, tag by funder, split high-signal from cheap 11:00 Frame the drafting: it interviews you, it does not invent you 12:00 First high-signal question, agent asks, you answer; watch it draft into the bank grounded in your words 15:00 Second high-signal question, show reuse across funders; park the cheap questions 17:00 Close: targets, questions, and drafted answers in one file, Step 03 compiles the funder packet and closes the workflow TOTAL: recorded at ~12 min (collapsed from the old steps 02-04: research + workbook/index + drafting)

ONE JOB: set up the file workspace, have Codex compile the funder-shaped narrative and budget into the outputs folder, reconcile the packet for a human to sign off, then show the payoff by running it again for a second funder, cheaper not harder. FRAME: this is the back half of the workflow as one continuous flow. The setup is fast (separation of piles, one rule), then two compiles (narrative, then budget), then the human comes back in to reconcile, and the close is where the whole point lands: the source pack is reusable, so the second funder is nearly free. Narrative and budget are both shaped to D-Prize's concept note. - Do not linger on setup; the separation of piles plus the one rule is the whole frame, then go straight into compiling. - The reconcile is the human's job: the agent saved the typing and the reshaping, not the judgment. MUST LAND (proof gates, then the close): - Three input piles present, outputs folder empty on purpose; the one rule said out loud: Codex writes only to outputs/, never edits source - Narrative draft created (narrative-draft.md), claims reference source files; open questions separated, not guessed - Budget draft created (budget-draft.md), categories match funder rules, assumptions visible and marked as estimates - Narrative and budget agree (same $19K), trace table covers the key claims, packet ready for a human to sign off - Then: run it again for a second funder, show funding-portfolio.docx, land the line ANCHOR FACTS (real): - 3 piles: always-true (master source pack) / this-grant (grant-brief, budget-assumptions) / this-funder (D-Prize rules) - D-Prize Round 1 is a two-page concept note, six prompts: distribution + barriers + the one thing to prove, marginal impact, outcome table, rough budget, one-sentence vision, team table - The one thing to prove: a vetted fellow, given tools, community, and a goal, ships work that earns a verifiable first $100 - Budget: $19,000 total, under the $20K cap, above the ~$16K average award; four lines (tiered agent seats ~$8,000, live sessions + one Demo Day ~$3,500, revenue-verification setup ~$3,000, stipended TAs ~$4,500) - Re-scope: full pilot ~$60K down to the ~$20K proof slice; estimates marked (blended seat ~$53/fellow/month, TA stipend ~$750/month, $3K verification not yet a quote); founder time in-kind, no travel (fully virtual) - The 4-file packet: narrative-draft, budget-draft, trace-table, open-questions; trace table is line to source plus confidence - Eight open questions are the real review agenda (e.g. can graduate TAs exist pre-pilot, is the $3K verification real, is the blended seat right) - Second funder: Echoing Green (idea-stage eligible, fiscal sponsor allowed, ~$100K matches the ask); funding-portfolio.docx, five funders against one $100K truth (D-Prize / Echoing Green / Anthropic / PJMF, same facts, different shapes) - The economics: the first funder was the expensive one (it built the source pack), each next one is cheaper, not harder. Zoom out: any repeated doc (board memos, donor reports, compliance), one source of truth, many shapes, full trace - Record LIVE, files appearing on screen is the proof OPEN (decide before recording): - Show the folder tree or just name the piles; show the full prompt to Codex or summarize it - How deep on the re-scope math; which assumption to flag as "this is an estimate, confirm it" - Which open question to read aloud (graduate-TAs-pre-pilot is sharpest); build the second funder live or cut to the finished portfolio; how long to hold the closing silence BEATS (riff freely, this is the long one): 0:00 Show the starter files open, name the three piles; outputs empty on purpose, state the one rule 2:00 Give Codex the inputs and the D-Prize shape, run the narrative live; narrative-draft.md appears, walk the six prompts 6:00 Pick one claim, trace it back to its source file; show a parked open question, confirm no source edited 9:00 Ask for the budget, run it live; budget-draft.md: $19K, the four lines 12:00 Walk the re-scope ($60K pilot down to the $20K slice), point at the visible assumptions marked as estimates 15:00 Reconcile: lay out the 4-file packet, confirm narrative and budget agree on $19K, read one open question as the agenda; a person signs off, not the agent 18:00 Turn the corner: run it again for Echoing Green, funding-portfolio.docx on screen, five funders one $100K truth 20:00 Zoom out: any repeated doc, one source of truth many shapes; land "cheaper, not more expensive", hold the silence TOTAL: recorded at ~6 min (collapsed from the old steps 03-05: setup + narrative + budget + reconcile/close)