Find Your Summit POC v2
bnimSummit Method
Greenbuild 2026 · Invest for Impact

The Summit Method · a one-minute primer

Start with your mission. We’ll find the question.

Most building decisions start by asking what it costs. The Summit Method starts by asking what the project is for, names the summit your mission points to, and finds the path that gets you there.

“Value engineering” cuts the mission to fit the budget.
Engineering value designs toward the mission until the numbers work.

Doing good and doing well overlap at the summit Doing good: serves the mission, supports people, builds community. Doing well: costs less to own, lowers risk, opens financing. Doing Good • Serves the mission • Supports people • Builds community Doing Well • Costs less to own • Lowers risk • Opens financing
1 2 3 4 5
  1. Getting startedKnow your starting point: buildings, bills, and condition.
  2. Where we’re headedTurn your mission into one Summit Question.
  3. Name the summitSet the goal and map the paths to it, including standing still.
  4. Find your pathTest each path on cost, risk, financing, and mission, then phase the one you choose.
  5. Climb and learnTake action, measure, and learn by doing.
  6. Find your Summit · this tool: steps 1 to 3, with a first look at the paths
  7. Summit Sprint and Delivery · steps 4 and 5, with your real data
For the web team: how this proof of concept works
  • Screen and Phone views. The toggle at the top right switches between Screen and Phone. Screen gives the current step most of the width and collapses the others to narrow rails; finishing a step (or clicking a rail or step tab) opens the next, and “Show all steps” puts all three side by side. Phone shows one step at a time. The choice is remembered in the browser. Windows narrower than 1,100 px always get Phone and hide the toggle. In Screen view, column 3 previews the paths from the draft until Step 3 is built.
  • Summit Brief. A one-page, Letter-size brief built for printing. Until Step 3 exists, the cost, carbon, financing, and fit rows show placeholders marked “Pending completion of Step 3 development.” If a long draft would spill past one page, fitBrief() scales the printed page down to fit.
  • Scope. Primer, Step 1 (Tell us), Step 2 (Find the question), and a Summit Brief of Steps 1 and 2 are live. Step 3 (Map the paths) is a placeholder that shows a “coming soon” message.
  • Reading a document. “Add a file” reads PDF, Word (.docx), and text files in the browser (pdf.js and mammoth.js load from a CDN only when a file is added), keeps the most relevant pages up to about 40,000 characters, and makes one AI call (mode: "extract") that fills empty Step 1 fields and lists key facts. Facts are sent with every later draft. The file itself is never uploaded or stored. Links are not fetched; bid portals usually need a sign-in.
  • Save and resume. No accounts, nothing stored on a server. Progress autosaves in the visitor’s browser and offers “Pick up where I left off” on return. Visitors can also download a progress file (.json) and open it later on any device, or copy a link that carries their answers in the address (the part after #, which never reaches the server). A later phase could add “email me my brief” with server storage.
  • One file. Markup, styles, and script are in this single HTML file with no build step and no libraries. Fonts load from Google Fonts (Figtree and IBM Plex Mono) as stand-ins for BNIM’s MarkOT; swap in the site’s fonts. Brand colors are CSS variables at the top of the style block (orange #F15A24, charcoal #404041).
  • AI drafting. “Find our Summit Question” and “Refine the draft” each make one call and expect one JSON object back. Set SUMMIT_API (top of the AI section of the script) to the address of the Summit API Worker, which holds BNIM’s Anthropic key and builds the prompt on the server. The page sends only the visitor’s inputs. Left empty, the page falls back to Claude’s artifact runtime, which only works inside Claude. Never put an API key in the page.
  • Response shape. summit_question, why, reframe {instead_of, ask}, outcomes [{outcome, measure, unit, basis, fit}], five_c [{c, message}], ladder, candidates, next_questions, assumptions. Step 2 uses the first six; the rest are kept for Step 3. normalize() cleans every response before it renders.
  • Without AI. The page opens on a worked example (a fictional housing nonprofit) so it never shows an empty state. If the AI call fails, the example stays and a plain message explains why.
  • Website field (unchanged). The preview can’t open other sites. On bnim.com a server-side fetch of the organization’s About and mission pages could feed the prompt. Store nothing a visitor enters without consent.
  • Layout. Single column, one step at a time, with step tabs and Back/Next. Works from phone width up. Expandable sections use native <details>.
  • Disclaimer. The planning-tool disclaimer in the footer should stay on any public version; have BNIM counsel confirm the wording.

The instruction sent with the next first-draft request