Houston Bloom · First matchmaking event · Virtual
Everyone shares one thing. Everyone leaves with their own card.
A virtual matchmaking night for Houstonians in tech. You tell us what you want other people here to be able to find. You leave with your own contact card and the people your work meets. Afterward, you choose whether what you shared joins the commons.
01 · The deal
Give one thing. Leave with two.
You give
What you want other Houstonians to know about One organisation, space, resource or community. In your own words. Before the event.You leave with
Your contact card Yours to hand out, forever, with no account behind it. Your matches The people whose work meets yours, from what you said you have and need.The card is the reason to show up. After the event, you decide whether what you shared joins the commons, where it persists across events.
02 · End to end
Four steps, and only one of them is an event.
Detail
Why the survey runs first. It gives the matching something to work with on the night and the room is not building from nothing. It also means the ten minutes people spend answering happen at home rather than eating the event.
Why the commons comes after. Contributing is a choice each person makes once the event is over, not a condition of attending. What someone chooses to add persists across events; what they keep back leaves with them.
Why the card is handed over at the event rather than emailed. It converts a survey response into attendance. Sent early, it is a file. Given on the night, it is the reason to be there.
Source: 06-program-design v0.1; 02-murmurations-findings v0.1 §6 and §9.03 · Before the night
The survey has to survive being forwarded to a stranger.
It arrives from someone they trust, about a thing they have never heard of, and they decide in about four seconds. So the ask is a question they can answer, not a project they have to understand.
The ask
What do you wish more Houstonians knew about?
The line under it
Houston is bigger than any of us can see.
Detail
Do not say Murmurations at first contact. The word costs people at the exact moment you can least afford it. It gets introduced later, once they have the card in hand and the idea has something to attach to.
Keep it short. A comparable pilot promised two minutes and took ten. Seven required questions, one of which is an email address. Everything that makes a usable map entry is the name, the link, the tags, and the person's relationship to it.
Ask in both directions. What you need and what you can offer. That pairing is what produced the stronger reciprocal matches in the pilot.
Likely vehicle: a Harmonica session rather than a static form. Not settled.
Source: houston-bloom-survey-1.md; 02-murmurations-findings v0.1 §7; 06-program-design v0.1 §2.04 · What they walk away with
A contact card every phone already knows how to open.
Standard vCard, plus one line pointing at their profile. The person lands in Contacts. The data stays at the link.
Detail
A vCard saved as a .vcf file opens natively on every phone and files itself into Contacts with nothing installed. That property is the whole reason to use it.
The card carries the person and the address of their profile. It does not carry the profile. Contact apps discard custom fields they do not recognise, so structured data placed inside the card should be assumed lost on import. Keeping the profile at the other end of a link means the human half lands where a person will find it later and the machine half survives intact.
It also fixes the size problem. A QR code holds roughly a couple of kilobytes at best, so encoding a short card rather than a whole profile keeps it scannable at ordinary print size.
Peer exchange does not route through us. Two people swap cards, each ends up with a contact and an address, and either can pull the other's full profile afterward and evaluate it however they like. Houston Bloom aggregates for the map. It is not the channel through which people know each other.
Source: 03-handoff-and-custody v0.1 §4 and §6.05 · On the night
Graham's software reads what people gave and recommends the groups.
Breakouts are assigned before the call starts, so running them live is one button.
Detail
Matching on inferred need beats keyword matching. In the pilot this ran against, groups built by reading what people meant produced markedly better conversations than groups built by matching terms. That is the argument for letting the software read the free text rather than forcing people into categories at intake.
It will leave people unmatched, and that is expected. The pilot placed seven groups of three and two pairs, then seated the remainder by hand. Someone needs that job on the night, and it should be treated as routine rather than as a failure of the tool.
Cold start is real. A matching tool has little to work with when the corpus is small, so a first event is partly there to seed it. Judge the first run accordingly.
Source: 06-program-design v0.1 §2, from the Madeira Impact Collective pilot account.06 · If you choose to contribute
The commons is opt-in, after the event, and what you add stays yours.
And when several people name the same place independently, that agreement is the quality signal. No scoring, no reputation model, no moderation queue.
Detail
Same subject from two contributors is not an error to be silently merged. It is corroboration, and the entry carries both sources and a count. Three people naming the same East End makerspace is one of the more valuable signals in the set, and surfacing it costs nothing to build.
Same subject from the same contributor is a different case entirely. That is an update, settled by newest timestamp.
Nobody has to host anything to take part. Houston Bloom holds a copy of what each person chose to add. Anyone who later publishes their own supersedes ours automatically, because aggregators settle conflicts by address. No migration, no permission, no coordinated switch.
Open: what the commons artifact concretely is for the people who opt in. The working documents specify a folder of files plus a single offline viewer. Confirm that is still the shape before this goes in front of anyone else.
Source: 05-portable-map-delivery v0.1 §4; 03-handoff-and-custody v0.1 §2.07 · Format
Virtual, and not as a compromise.
- Closing a breakout is instantMoving people between conversations mid-sentence was the hardest part of the pilot that ran this in a room.
- The map is a screen artifactPointing at it together works better shared than projected.
- Houston is enormousA survey forwarded cold converts to a link far better than to a weeknight drive across town.
What we give up is doing it in front of each other. That is what event two is for.
Detail
What virtual costs. The attestation moment, people seeing who claimed what and being positioned to vouch for it, is weaker through a screen. That matters when cards are changing hands and vouches are being made about people. For a first event whose job is matching people, it is not what we are optimising for.
There is a design behind that moment for when it comes back in person: ownership computed by holding the contribution set next to the people present, surfacing while they are in the room and dissolving when they leave. It is written up and it is not being built for run one.
Source: 04-ephemeral-attribution v0.1; format decision recorded 2026-08-05.08 · What the group who ran one told us
The room works. The follow-up is where it dies.
finished the pre-event form, even at ten minutes instead of the two promised.
submitted commitments afterward. This is the number we design against.
facilitator running it alone was the risk they named themselves.
So commitments get captured live, in silence, before anyone leaves the call. And there is a co-host, because you cannot facilitate and operate breakouts at once.
Detail
From the Madeira Impact Collective matchmaking pilot, reported to us directly. They ran matchmaking alone and still found the transitions hard. We are stacking more outputs onto one event than they were, which is the reason the co-host is not optional.
The commitment mechanism. Drop the link in chat, three quiet minutes while everyone is still on the call, and do not move on until it is done. Asking afterward is what produces five out of twenty one.
Also worth carrying forward: asking what you need alongside what you offer produced stronger reciprocal matches, and a follow-up channel kept collaborations moving after the event.
Source: 06-program-design v0.1 §2; call summary, 2026-08-03.09 · The thing we cannot get wrong
Say what is true about anonymity, once, in plain words.
Not this
Your submission is anonymous.
This
Your name will not appear publicly, and Houston Bloom will know the entry is yours.
If a submission can be reconnected to a person, it is pseudonymous, not anonymous. Getting that wrong once would cost more trust than the whole map is worth.
Detail
People choose a visibility level at submission time: the resource appears and they do not, or Houston Bloom participants can see their name attached, or it may appear on the open map and be discoverable by anyone.
Only the third can ever be published outward, because a public index is globally readable and cannot express the other two.
If we want the stronger promise, it has to reach back to collection: split the submission on receipt, store content against a random token, and keep a separate table mapping token to email. That table becomes the only re-identification path, and it can be restricted or destroyed. Without that split, the honest sentence is the one on the right of this slide.
Source: 04-ephemeral-attribution v0.1 §3; houston-bloom-survey-1.md section D.10 · Scope for run one
Four real things we are not building yet.
- ×
A public index - ×
An aggregator - ×
Hosting anyone's profile - ×
A venue upload page
All four are legitimate. None is needed to test what this run is testing, and each adds something that can fail in public while everyone watches.
Detail
An index and an aggregator exist so strangers who never met can find each other. For a first cohort whose profiles came from a survey we already hold, the files are in hand and both steps are unnecessary. They become useful the moment reach beyond the room is wanted, and that is a later decision rather than a prerequisite.
What run one tests: whether people can describe what they are part of without being taught a schema, whether Houstonians describing Houston organisations use vocabulary divergent enough to break naive matching, whether holding your own card changes how you relate to the data, and whether the assembled commons beats a shared document.
Source: 02-murmurations-findings v0.1 §6; 03-handoff-and-custody v0.1 §5.1.11 · What I need from you
Three things I should not decide alone.
- 01Co-host the nightFacilitating and operating breakouts at the same time is the failure the last group named. This is the one thing that cannot be solved with better software.
- 02What does the software need from the survey?The intake format has to match what the matching expects, and that constraint should shape the questions before they go out, not after.
- 03Who gets the first send?It only works forwarded through people who are already trusted. Whose lists, and who does the forwarding.
Still open behind these
- The date, and whether the survey window is long enough for the matching to work on the night.
- Whether the survey ships as a Harmonica session or a plain form.
- What the commons artifact is for the people who opt in.
- Whether we build the intake split that makes the stronger anonymity promise keepable.
- Whether cards are generated ahead and handed over, or generated live on the call.
- Where the collection lives so it survives a change of organiser.
Houston Bloom
People find each other. The commons is what falls out.
Run one is a prototype and it is built to be judged as one. It publishes either way.
Sourcing
Built from seven working documents, all version 0.1 dated 5 August 2026, and from the 3 August call summary in which the Madeira Impact Collective described running their own matchmaking pilot. The documents sit in MatchMakingEvent/_context/.
Every figure here is quoted from those sources. The pilot numbers are theirs, not ours. Nothing on these slides is a measured result of a Houston Bloom event, because the first one has not happened yet.