Files
coopertron/tests/fixtures/sample.review.md
T
solo.istandClaude Fable 5.1 05f4e88867 synthesize: anonymous braindump CSV → themed review doc via the co-op gateway
Stdlib-only pipeline for the pre-meeting braindump: CryptPad Form CSV in,
review.md + synthesis.json out. Two model stages (tag atomic points with
verbatim quotes, cluster into themes/novel/dissent) wrapped in deterministic
Python that verifies every quote against the source and computes counts from
item ids, so the model never writes the final document. Defaults to
tinfoil/glm-5.3 with reasoning_effort=low: without it the model spent the
whole 8000-token budget on hidden reasoning and returned nothing (probed
2026-10-19; findings in README). Refuses non-TEE models without a flag.
Content-hash cache so prompt tuning reruns are free. Offline unittest suite
with a fake gateway; smoke-tested live on 5 synthetic responses for ~$0.03.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-10-19 02:56:50 -04:00

10 KiB

Braindump synthesis

Generated 2026-10-19 from sample.csv · 12 responses · 71 points · 2 themes · model tinfoil/glm-5.3

Counts below are computed from the extracted points, not estimated by the model. Quotes are verbatim from the responses. The full responses are in the appendix.

Priorities (pick up to 3)

12 of 12 responses answered.

  • ████████████ 4 · privacy
  • ████████████ 4 · governance and participation
  • █████████ 3 · pricing and fairness
  • █████████ 3 · docs and onboarding
  • █████████ 3 · API and developer tooling
  • █████████ 3 · model choice
  • █████████ 3 · community and chat
  • ██████ 2 · owning our own infrastructure
  • ██████ 2 · green energy

Themes

# Theme People Points For / Against
1 Own the stack 7 36 36 / 0
2 Institutional access 6 35 35 / 0

1. Own the stack

7 of 12 people · 36 idea

Self-hosting versus buying tokens.

Points raised:

  • point about Q6 from r06
  • point about Q1 from r07
  • point about Q2 from r07
  • point about Q3 from r07
  • point about Q4 from r07
  • point about Q5 from r07
  • point about Q6 from r07
  • point about Q1 from r08
  • point about Q2 from r08
  • point about Q3 from r08
  • point about Q4 from r08
  • point about Q5 from r08
  • point about Q6 from r08
  • point about Q7 from r08
  • point about Q1 from r09
  • point about Q2 from r09
  • point about Q3 from r09
  • point about Q4 from r09
  • point about Q5 from r09
  • point about Q1 from r10
  • point about Q2 from r10
  • point about Q3 from r10
  • point about Q5 from r10
  • point about Q6 from r10
  • point about Q1 from r11
  • point about Q2 from r11
  • point about Q3 from r11
  • point about Q4 from r11
  • point about Q5 from r11
  • point about Q6 from r11
  • point about Q1 from r12
  • point about Q2 from r12
  • point about Q3 from r12
  • point about Q4 from r12
  • point about Q5 from r12
  • point about Q7 from r12

"I disagree with adding providers that" (r06, Q6) "I run a small nonprofit and" (r07, Q1) "Governance nerd. I barely use the" (r08, Q1) "Needed an API for a side" (r09, Q1)

2. Institutional access

6 of 12 people · 35 idea

Group and org accounts.

Points raised:

  • point about Q1 from r01
  • point about Q2 from r01
  • point about Q3 from r01
  • point about Q4 from r01
  • point about Q5 from r01
  • point about Q6 from r01
  • point about Q1 from r02
  • point about Q2 from r02
  • point about Q3 from r02
  • point about Q4 from r02
  • point about Q5 from r02
  • point about Q6 from r02
  • point about Q7 from r02
  • point about Q1 from r03
  • point about Q2 from r03
  • point about Q3 from r03
  • point about Q4 from r03
  • point about Q5 from r03
  • point about Q6 from r03
  • point about Q1 from r04
  • point about Q2 from r04
  • point about Q3 from r04
  • point about Q5 from r04
  • point about Q1 from r05
  • point about Q2 from r05
  • point about Q3 from r05
  • point about Q4 from r05
  • point about Q5 from r05
  • point about Q6 from r05
  • point about Q7 from r05
  • point about Q1 from r06
  • point about Q2 from r06
  • point about Q3 from r06
  • point about Q4 from r06
  • point about Q5 from r06

"Privacy. I wanted an AI I" (r01, Q1) "Green energy mostly. I teach and" (r02, Q1) "Curiosity about co-op governance. I use" (r03, Q1) "A friend invited me. I mostly" (r04, Q1)

Standout ideas

Raised by one person, specific enough to be worth a look.

  • point about Q1 from r01 (idea)
    "Privacy. I wanted an AI I" (r01, Q1)

Notable dissent

Places where people pull in different directions. Worth time in the meeting.

  • Own the stack: Cost doubts.

    "Nice work on the dashboard." (r12, Q7)

Open questions for the group

  1. Institutional tier? (from r01)
  2. Bare string question

Appendix: all responses, verbatim

What drew you to the co-op, and what do you use it for today?

  • r01: Privacy. I wanted an AI I could use for client work without feeding a big vendor. Mostly chat, some API from scripts.
  • r02: Green energy mostly. I teach and wanted something I could recommend to students with a straight face.
  • r03: Curiosity about co-op governance. I use the API from a small writing tool I maintain.
  • r04: A friend invited me. I mostly use the chat for drafting emails.
  • r05: Wanted to support a cooperative alternative. I use it daily for coding help.
  • r06: Privacy and the TEE story. Chat only.
  • r07: I run a small nonprofit and we needed shared AI access we could justify ethically.
  • r08: Governance nerd. I barely use the models; I joined for the experiment.
  • r09: Needed an API for a side project and liked the values.
  • r10: Climate reasons.
  • r11: Wanted to try open-weight models without setting up my own box. Mostly chat and a bit of API.
  • r12: Someone mentioned it at a Metagov call.

Ideas. What should the co-op build, try, or offer next?

  • r01: A shared prompt library members can contribute to. Also a Matrix bot that answers questions about the docs.
  • r02: Classroom or group accounts: one payment, several seats, shared budget.
  • r03: An institutional or app-partner membership so projects like mine can buy inference for their users through the co-op instead of each user joining separately.
  • r04: Voice mode on the phone app.
  • r05: Self-host at least one open model on our own hardware so we are not just reselling other people's APIs.
  • r06: Private web search inside the enclave instead of our own SearXNG.
  • r07: Group accounts with a shared budget and an admin who can see spend per seat.
  • r08: Publish a monthly financial and usage report and discuss it on Loomio.
  • r09: An Anthropic-compatible /v1/messages endpoint that is actually documented and tested, so Claude-style clients just work.
  • r10: Default every new member to the greenest model, not the fastest.
  • r11: Image generation in the chat picker, not hidden behind an integration button.
  • r12: A quarterly braindump like this one, and a public roadmap page built from it.

Wants. What would make membership more valuable to you personally?

  • r01: A cheaper tier for light users; I barely touch my allowance most months.
  • r02: Being able to see the energy and emissions per conversation, not just tokens.
  • r03: Stable model names. When tinfoil/glm-5-3-flash disappeared my app broke.
  • r04: Honestly it is fine as is.
  • r05: Higher allowance for people who pay more; the flat $15 cap makes paying extra feel pointless.
  • r06: Nothing specific.
  • r07: An invoice I can hand to our bookkeeper instead of an Open Collective receipt.
  • r08: Real decision rights, not advisory votes.
  • r09: Per-key spend limits so one script cannot burn my whole month.
  • r10: A badge or summary of emissions avoided versus a mainstream provider.
  • r11: Longer memory across chats.
  • r12: Knowing what other members are working on; a members directory people opt into.

Needs. What is missing, broken, or in your way right now?

  • r01: The chat loses my place on mobile when I switch apps. Also the model picker names are confusing (dotted vs dashed slugs).
  • r02: Onboarding email landed in spam for two of my students. The email-must-match rule tripped one of them.
  • r03: No deprecation notice period for models. No status page.
  • r05: Tool calling is flaky on some models and there is no doc saying which ones support it.
  • r06: The chat is slow to start on voice sometimes.
  • r07: The dashboard shows my usage but not my organization's.
  • r08: The charter still says decisions are not binding.
  • r09: Budget exceeded errors arrive mid-stream with no warning beforehand.
  • r11: Image models in the picker caused 500 errors and then got removed with no explanation.
  • r12: Matrix invites arrived before I knew what Matrix was. A one-page explainer would help.

Goals. Where should the co-op be a year from now?

  • r01: Fifty members, real elections, and at least one model we host ourselves.
  • r02: A recognized option for universities and nonprofits, with an institutional tier.
  • r03: A documented SLA-ish commitment, even an informal one, and a changelog members can subscribe to.
  • r04: More members so it keeps existing.
  • r05: Our own GPU box, even a small one, and a path to more.
  • r06: Keep it small and trustworthy. Growth is not a goal for me.
  • r07: An institutional tier with clear terms and pricing.
  • r08: A functioning member-elected board and a budget members approve.
  • r09: A developer-friendly gateway people choose on merit, not only on values.
  • r10: Carbon accounting published per month.
  • r11: More models, including vision, at the same price.
  • r12: A community that knows itself, not just a service.

Dissent. What worries you, what do you disagree with, or what are we getting wrong?

  • r01: I worry we are spreading across too many providers. Three providers is already hard to explain to a new member.
  • r02: Not dissent exactly, but the Loomio threads are long and I cannot keep up.
  • r03: I think the terms forbidding any redistribution of access will block exactly the kind of mission-aligned projects we say we want.
  • r05: Self-hosting will cost more than buying tokens and we should be honest about that before voting for it.
  • r06: I disagree with adding providers that are only policy-private. It dilutes the one thing that made me join.
  • r07: The pay-what-you-can model is lovely but I suspect it is not covering costs. Show us the numbers.
  • r08: We keep adding features before we have governance. Build the decision process first.
  • r10: Tinfoil is not green and we should say so more plainly on the models page.
  • r11: I do not think self-hosting is worth it at our size.

Anything else?

  • r02: Thank you for doing this.
  • r05: Happy to help with ops.
  • r08: Could we do these braindumps every quarter?
  • r12: Nice work on the dashboard.

Processing notes

3 extracted point(s) were discarded because their quote could not be verified against the source text or they referenced an unknown response.