A continuous governance surface: A Talk to the City-Inspired Experiment

🗳️ Meta-Governance
1

TL;DR

  • Lighthouse Labs adapted the methodology behind Talk to the City, to build a structural map of tensions in ENS governance discourse.
  • We processed the ENS governance forum data (Jan25-Nov25) to cluster recurring tensions and claims that shape this community’s decision-making.
  • Notably, these algorithmically-extracted themes show strong alignment with the governance challenges independently identified in the recently published Metagov Phase 1 results
  • Given the overlap between (manual qualitative vs AI synthesized) we are leaning into the following thesis:

A continous surface could be created from emperical forum data to provide a continuous surface of tensions and resolutions in the DAO.

What is Talk to the City and Why Does It Matter?

For those unfamiliar, Talk to the City (T3C) is an open-source AI tool
built by the AI Objectives Institute, a nonprofit research lab. It’s designed to take large-scale democratic input — surveys, interviews, pools — and distil it into navigable, structured maps of opinion while preserving the nuance and diversity of individual voices.

T3C has been deployed at national scale. Taiwan’s Ministry of Digital Affairs used it to process public input across the 2023 AI Assemblies, and has since stood up its own hosted instance for ongoing policy work. It’s also the champion in modern civics work. https://civic.ai/.

What makes T3C stand out is its bottom up approach: every high-level theme is grounded back to specific participant statements, making the analysis auditable. You can see a broad pattern, then drill down to the exact words behind it.

We believe this kind of tooling would deeply benifit ENS governance which is what led us to this work.

Why We Did This for ENS

As advocates of the retro.

We also had some experiments we wanted to run. Over the last few months we pieced together a pipeline that ingested forum data, iterated on clustering approaches and claim extraction.

With the recent conclusion of Phase 1, we were excited to see how two methodologies had such overlap in their results thus prompting to concretely share this work.

Why fork?

The original T3C pipeline was not really designed for ingesting hierarchical forum data, so we we had to iterate on a few designs to extract signal.

When we ran the base topic-extraction against ENS forum data, we encountered a lot of noise. We would get "EP 5.13 Budget Discussion" as a topic, rather than the deeper structural question.

To solve this, we redesigned the pipeline with several key changes also encoding our lived experience with how discussions in tehcnical communities generally fare.

Consider the following theme. "We need more accountability". While we could map claims to that topic it does not reveal why we need more accountablity.

We believe that nearly most disagreements can be framed across a spectum. So we introduced the notion of Themes and Tensions in our clustering approach.

A structural theme in our view is defined not as a topic label but as a spectrum between two legitimate positions (poles).

Each pole comes with a “typical argument” — the strongest version of that position as actually expressed in forum discourse.

{
  "name": "DAO-Defined Needs vs. Provider-Defined Proposals",
  "tension": "Should the DAO specify what work it needs (RFP model) before soliciting applications, or should service providers define their own scope and let delegates choose among competing visions?",
  "poles": [
    {
      "position": "DAO should define needs first",
      "typical_argument": "Without clear DAO-defined requirements, multiple providers build the same thing, wasting funds. An RFP model ensures funding targets actual gaps and allows meaningful performance evaluation against defined deliverables."
    },
    {
      "position": "Providers should define scope",
      "typical_argument": "The DAO lacks the specialized expertise to write good RFPs, and prescriptive scoping stifles innovation. Open applications let providers propose novel solutions the DAO wouldn't have thought to request, and delegates can sort quality through voting."
    }
  ],
  "instances": [
    {
      "reference": "SPP Season 2 debate on DAO-defined needs vs. provider-defined proposals",
      "expression": "Explicit discussion of whether the DAO should specify needed work before soliciting SPP applications, or continue the open-application model."
    },
    {
      "reference": "RFP model as an alternative to SPP open applications",
      "expression": "Proposals to replace the open SPP application process with formal requests for proposals, with debate about whether the DAO has expertise to scope them properly."
    },
    {
      "reference": "Duplication of provider efforts in SPP",
      "expression": "Concern that multiple funded teams build overlapping products because no central prioritization exists, illustrating the cost of provider-defined scoping."
    }
  ]
}

This framing does two things: it prevents the analysis from becoming prescriptive (“the DAO should do X”), and it acknowledges that most governance disagreements aren’t about right vs. wrong — they’re about trade-offs between competing ideals/methodologies.

Consider the following claim:

{
  "claim": "The Service Provider program should fund individual projects based on merit rather than allocating funding to companies as institutional units.",
  "quote": "That's because you are still thinking in terms of funding companies, rather than projects. The point of this program is to fund projects, not companies.",
  "themes": [
      {
        "theme": "DAO-Defined Needs vs. Provider-Defined Scope",
        "pole": "Providers should define their own proposals",
        "note": "This supports a project-based submission model where providers propose individual projects and the DAO evaluates them on merit rather than bundling them by company."
      }
  ],
  "post_id": "56929"
}

Validation

Two completely independent approaches:

  • 22 human interviews analysed qualitatively by governance researchers,
  • and algorithmic extraction from public forum posts

arrived at a similar picture of the DAO’s governance landscape. This is significant for a few reasons:

  • the forum is a genuine signal source, forum discourse reflects the same structural challenges that experienced stakeholders describe in private interviews.
  • It validates our algorithmic approach, it extracted the same underlying governance tensions that trained researchers identified through qualitative work.

Results

Next steps

Tracking how the community’s position on structural themes is a valuable signal about governance health. Are tensions are being resolved?, are there new ones are emerging? Do stated tensions actually have formal resolutions?

The pipeline is in early stages, but could be run continiously as new forum activity is generated.

We created a (very rough) prototype of how we envisage this infomation being presented: https://v0-ens-tttdao.vercel.app/

References

3 Likes