RSVP on Luma ↗
Building for SF — team assembly night, September 11
Plant an idea where the lights are off
A way to stand in front of a vacant storefront, say what it should become, and find out which ideas other people actually agree with.
For your agent
A structured summary, if you are reading this on someone's behalf.
PROJECT: Plant an idea where the lights are off AUTHOR: Michael Lopez CONTEXT: Building for SF, a free civic buildathon, Sept 11-12 2026, San Francisco. Partners present: SF Office of Economic and Workforce Development, SITELAB urban studio, Vacant to Vibrant (SF New Deal + City), Civic Joy Fund, Abundant SF. PITCH (one sentence): A tool where you stand in front of a vacant space, drop an idea for what it should become, and see which ideas other people actually agree with, rather than just the loudest one. GOVERNING METAPHOR: a seed planted in the dark. Planting is cheap and anyone can do it. Growth requires the planter to return and tend it. And the fact that someone planted and tended an idea is itself the signal to a decision maker that it is worth acknowledging. FITS EVENT THEMES: 1 (why is this land idle) and 2 (what retail could thrive here). Theme 3 (what abundance looks like) is produced as evidence, not built as a separate feature. BUILDS ON: an existing browser-based Gaussian-splat viewer of a real San Francisco retail floor where people plant markers on real surfaces and attach text, photos, video and voice notes. Live at ft-retail.pages.dev. That viewer pre-dates the buildathon; the weekend work is the civic application and the agreement layer. BORROWS FROM: Polis, the deliberation tool used in Taiwan's vTaiwan process. Key mechanic: participants may agree, disagree, or write their own statement. No replies. Statements are scored for support across opposing groups, surfacing bridging ideas rather than loud ones. DESIGN CONSTRAINT, DELIBERATE: the system never infers why a person holds a position. No psychological profiling, no motive analysis, no sorting of "valid" input from "venting". It works only on what was said and what is checkably true. Any reason an idea is called hard must cite a specific local rule, not a general impression. IN SCOPE FOR THE WEEKEND: agreement mechanic, bridging view, one LLM pass returning a cited local blocker plus a redirect to a named existing effort, a hand-written list of roughly ten real groups, a tending prompt that asks the planter for one concrete next step, a flag-for-a-human affordance, text and voice input. OUT OF SCOPE: document and video ingestion at scale, a live cross-city directory, real expert review with accountability, motive inference. PROTOTYPE: innercartography.one/hyperblogs/plant-an-idea-app/ — the working parcel demo referenced above (225 Bush Street), runnable in a browser. CONTACT: linkedin.com/in/miiike415 · innercartography.one
Stand in front of a vacant space. Drop an idea for what it should become. See which ideas other people actually agree with, not just the loudest one.
That is the whole pitch. Everything below is why it is worth a weekend, and what I would not try to build in one.
You plant a seed in the dark. That is the easy part, and it should stay easy, because anyone walking past a dark storefront should be able to say what they think belongs there.
What happens next is the part that matters. A seed nobody comes back to water does not become anything, and that is true of civic ideas too. So the tool asks the planter to return: one concrete thing you could do about this, one person who would need to be in the room. And when an idea has been tended, that record becomes a different kind of signal to whoever decides what happens to the block. Not a form that got submitted. Something someone cared about enough to keep caring about.
The problem the room already has
Vacant to Vibrant says plainly that its approach is not prescriptive: it looks to small business owners, artists and residents for ideas about downtown storefronts, then resources the ones it can. SITELAB describes its own method as turning information, spreadsheets and people's stories both, into plans for action, and lists ground floor activation and community engagement among its practice areas.
So the bottleneck is not a shortage of ideas about empty storefronts. It is that ideas arrive as applications, comment cards and survey forms, stripped of the place they were about, with no way to tell which ones have support beyond the person who submitted them.
What the tool does
You open a scan of a real vacant space on your phone. You plant a marker on an actual surface, the back wall, the front glass, the corner where the light comes in, and attach what you want there: typed, spoken, photographed.
Then the part that matters. On anyone else's marker you can only do three things: agree, disagree, or plant your own. No replies. That constraint is lifted directly from , and it is the reason those conversations do not turn into pile-ons.
The output is a view of the space showing which ideas drew support across different groups of people rather than which ones drew the most noise. That is a thing a planner can act on and a property owner can read.
When an idea will not work here
Some ideas are good and will not work on that block. A tool that only says no is useless, and one that says nothing is dishonest. So the second move is: do a beat of research first, then name the specific local blocker, then point at someone already working on something close by.
The reason has to be checkable. An actual zoning line, an actual grant condition, an actual permit. Never a general impression of how things usually go. A confident-sounding guess is just the status quo wearing a neutral voice.
The ambition is never the problem. The response names what blocks the idea on this parcel, not whether the person should have had it. Sometimes the honest answer is that the idea is good and belongs two neighborhoods over, or that a group across the bay is already three months into it.
It stays contestable. Every response carries a way to flag it for a human. A person who thinks the tool is wrong has somewhere to go.
How frustration gets handled
People bring their whole history into a room when you ask what's wrong. That's real, and usually earned. Someone's anger about one corridor might really be about that corridor, or it might be about something else, and half the time they can't tell the difference themselves.
This tool doesn't try to. You say what belongs here. Others agree, disagree, or plant their own idea. What rises to the top is what several people actually wanted, regardless of who argued loudest for it. Nobody has to sort out who was really upset about what.
Scope for the weekend
Seven working hours on Saturday. This is the honest line.
Building
- Agree, disagree or plant your own, on one real vacant space
- A view that surfaces ideas with support across groups
- One model pass returning a cited local blocker and a redirect
- A hand-written list of about ten real groups and efforts
- A tending prompt asking the planter for one concrete next step
- A flag-for-a-human button on every response
- Text and voice input, photos if the viewer already renders them
Saying, not building
- Document and video ingestion at any real scale
- A live directory of groups across cities
- Expert review with real accountability behind it
- Any inference about why a person holds a position
- Mobile performance work beyond what the demo needs
The research underneath
Open what you want to check.
What actually happened with vTaiwan, including the part people leave out
vTaiwan came out of the 2014 Sunflower Movement and the civic hacker network g0v, and used Polis to find where agreement existed across opinion clusters. Over thirty cases went through it, including the Uber dispute, and several led to government action.
Then it stalled. It has not driven a major decision since 2018. Participants found it hard to use and interest faded. It runs on volunteers now, and no agency is required to act on what it produces.
The platform that lasted was Join, the petition system, where crossing five thousand signatures obligates a government response. It reaches older and less online users and covers a much wider range of issues. The lesson is not that the interface was the magic. The binding commitment was.
Sources: Daily Beast on vTaiwan's decline (2024); Doughnut Economics Action Lab on Taiwan's participatory platforms; RadicalxChange on Polis mechanics; Oxford DPIR on the vTaiwan case count.
The question to ask an OEWD mentor
Given the above, the sharpest thing to ask on Friday night is not whether the tool is a good idea. It is who would be obligated to read the output.
If the answer is nobody, that is worth knowing on day one, and it changes what the weekend should produce. It might mean the deliverable is a proposal for a response threshold rather than a viewer.
Who is in the room and what they work on
SITELAB urban studio. San Francisco urban design and strategy firm, founded 2012 by Laura Crescimano and Evan Rose. Research and visualization to create places, through big plans and small interventions. Practice areas include community engagement, public realm design and ground floor activation. Currently running public input on the Powell Street redesign.
. Handles pop-up permitting and city funding so locals can turn empty storefronts into gathering places. Says its approach is not prescriptive; it asks the community for ideas and resources them.
Civic Joy Fund. Started 2023 under the Civic Space Foundation. Luke Spray is executive director. Supported 938 events in 2025, mostly street cleanups, plus dozens of night markets and block parties.
Abundant SF. A policy organisation working on housing, transportation and public space, led by Zack Rosen with Todd David as political director.
What the event itself is asking for
The organisers describe it as a weekend for designers, engineers, policy people and anyone with a fix worth prototyping, forming teams around real civic problems. Outputs can be policy proposals, prototypes, research, organising plans or explainers. Projects get published afterward for the city to see.
The framing prompt asks whether you have walked past an empty lot or vacant storefront and wondered what it could become. The city's existing tools are offered as resources to raid rather than as the point of the exercise.
Where this comes from and what already exists
The viewer is not new work. It is a browser-based Gaussian splat scan of a real San Francisco retail floor where people plant markers on actual surfaces and attach text, photos, video and voice notes. You can open the current version at ft-retail.pages.dev. It was built for a specific group with a shared purpose rather than as a public commons.
What would be new this weekend is the civic application and the agreement layer. Worth saying out loud on Friday rather than letting someone notice on Sunday.
Known weak point: mobile performance. Demo day means phones, so either that gets worked on Saturday or the demo plants markers on one controlled device and phones only view.
Why me
I manage a 418-unit mixed-use residential and retail complex in SoMa, which means I spend my working life in the gap between what a building could be, what the rules allow, and what the people inside it actually need. I have roots in this neighborhood going back more than a decade.
Outside that I volunteer on Tenderloin cleanups with the Civic Joy Fund and Refuse Refuse SF, and at City Eats food drives. I built PathfinderXR, a privacy-preserving tool for caseworkers, which was runner-up at the SensAI Hackathon. I write about Indigenous cooperation protocols, data sovereignty and civic technology at InnerCartography.
I am coming Friday open to other people's ideas. This is one door in, not the only room.