{"path":"research/blobs-community-sandboxes-and-the-advice-process.md","content":"# Blobs: testing Deliberus inside a community that decides by advice process\n\n**Date**: 2026-09-19 · **Status**: design thinking, Claude's, **nothing ruled**. The founder's question,\nfrom an unconference of a participatory community he belongs to: *\"think deeply about how Deliberus\ncould carefully be tested in a sandbox (or multiple, let's call them blobs) for ways to enrich and\ndeepen the decision-making and deliberation process … and how it could interface with the existing\ndo-ocracy/democratic elements of Advice Processes etc and the formal board etc.\"*\n\nThis doc is community-agnostic on purpose. The application to a named community is strategy and lives\nin `.private/`, per the born-private rule.\n\n## 1. The fit, stated once\n\n**The advice process is a way of deciding that has no counting step, so its whole quality rests on\nhow reasons are handled, and it has no tooling for reasons.**\n\nThe rule, in Laloux's words: *\"Any person can make any decision but must seek advice from affected\nparties and people with expertise.\"* Loomio's guide gives the steps: notice a problem or opportunity\nand take the initiative · seek input · clarify and strengthen the advice through discussion · decide,\nand publish an **outcome** — *\"a clear statement of the decision made and what will happen next\"*,\nwhich should *\"explain how their advice influenced the final decision.\"* Nobody votes. One\nself-appointed person decides and is accountable for it.\n\nEvery mainstream civic tool counts something: polls, votes, clusters of agreement. The advice process\nneeds the opposite: a way to hold many people's reasons, show the decider what they actually bear on,\nand show the advisers afterwards what happened to what they said. Deliberus refuses verdicts by design\n(*the map records reasoning; it never endorses conclusions*) and treats a retelling of someone's view\nas theirs to rule on (*only you can say whether a retelling of your view is fair*). That is the same\nshape from the other side. It is a rare structural fit, and worth testing carefully because of it.\n\n## 2. Where the advice process hurts, and which part of the map answers it\n\nKnown weak points of the method, from its own literature and from communities that run on it:\n\n| The pain | What the map offers | Status here |\n|---|---|---|\n| **Advice as theatre**: the decider asks, then does what they meant to do anyway. Accountability is only social. | The outcome becomes a *weighing*: which advice moved the decision, which was outweighed and by what, attributed to the decider. | weighing questions shipped; an outcome helper is unbuilt |\n| **\"Was I heard?\"** An adviser cannot tell what happened to their point. | Their point is a thing with an address, and they can call its retelling fair or unfair. | addressing shipped; the fair/unfair act is unbuilt |\n| **Volume**: a contested thread runs to a hundred comments and one volunteer must hold it all. | The same advice from twelve people becomes one claim with twelve names on it, sorted by kind: what can be checked, what a word means, what ought to be done, what matters to someone. | sameness merging is ruled, unbuilt |\n| **The same fight every year.** Reasons live in old threads and in heads. | The reasons are kept. Round two starts where round one stopped. | the core claim of the project; untested on a community |\n| **Newcomers cannot join in** because deciding well needs the history. | The history is readable as reasons, with dates. | dated readings shipped |\n| **A fight hangs on a sentence nobody said.** | Finding the unwritten premise is the map's most valuable move. | strongest by hand, weakest in the pipeline |\n| **Decisions never expire.** *\"We do this until we make another decision\"*, with no memory of what would justify revisiting. | Premises carry dates; stale ones get flagged; a decision can name what would reopen it. | staleness flags shipped; revisit triggers unbuilt |\n| **Informal power.** With no formal structure, those with time, history and persistence carry the day (Freeman, *The Tyranny of Structurelessness*). | Nothing is overwritten and the last word has no privilege, so persistence buys nothing. | shipped as a principle; **and it cuts both ways, § 6** |\n\n**Checked against a real room the same day (2026-09-19).** The table above was written from the\nmethod's literature. That afternoon the founder photographed the whiteboards of two workshops a\nparticipatory community ran on its own decision-making and onboarding (transcribed privately). The\ndecision board names the same pains, several nearly word for word: initiators who do not wrap up, a\nquestion about how long a decision stays valid, twenty hours of comment handling, the same voices\nheard, big topics tangled with specific decisions, and, at the centre of its own mind map, *\"What\nare we actually discussing\"*. **Three things the table missed, each of which changes the design:**\n\n1. **The first tools a community reaches for are the two the evidence warns about**: comment\n   summaries and a poll were both on the board. The room flagged the poll itself (*popularity\n   contest*). Nobody flagged the summary, which is the one measured to steer.\n2. **Platform fatigue is a named pain** (too many competing platforms, digital overwhelm), so a\n   trial that adds one more place to go makes the named problem worse. A map has to live beside the\n   community's existing forum and never be somewhere people must go.\n3. **The room's own fear is a tool that becomes control**: it drew the move from power handed to\n   someone else toward power each person carries. Anything that ranks people or announces a result\n   reads as the wrong side of that drawing.\n\nOne more datum, about the pitch rather than the design: the room wrote Deliberus down as *a new\nsocial media platform*. That is a real stranger-test result for the spoken explanation.\n\n## 3. What a blob is\n\n**A blob is a compartment of the graph with its own membrane and its own lifetime: the same code,\nseparate contents.** The idea is already in the corpus as a finding from biology\n([convictions-across-scales.md](convictions-across-scales.md) § One Graph): life shares one genetic\ncode and separates its cytoplasm aggressively, so *share the code, compartmentalise the contents*.\nA blob is that finding with a name.\n\n*(Word collision, noted once: elsewhere the corpus says \"blob\" for an undivided lump of prose. This\nis the other sense: a bounded cell. \"Prose blob\" and \"blob\" stay distinguishable by context.)*\n\nA blob's membrane answers five questions, written down before it opens:\n\n1. **Who may read it, and who may write in it?** Never wider than the source the words came from. If\n   the thread was members-only, so is the blob.\n2. **What comes in from the commons?** The code: the kinds of claim, the argument patterns and their\n   critical questions, the strength arithmetic. And shared *facts* with sources, which communities\n   need constantly (what does the permit law actually require?).\n3. **What may leave?** Nothing by default. A claim leaves only with its author's consent. Patterns\n   without content (*the crux was unwritten on both sides*) may leave with the community's consent.\n4. **How does it end?** Every blob has a lifetime and a way to dissolve it that anyone inside can\n   trigger. **This is where a blob differs from the commons**: in the commons nothing is deleted; in\n   a sandbox the right to leave outranks permanence.\n5. **Who keeps it, and what is their interest?** Said out loud. Whoever frames the map frames the\n   debate, so the keeper is never presented as neutral.\n\nWhy compartments at all, instead of the one public graph: consent in a small community where everyone\nknows who said what · reversibility · local meanings (a community's \"consent\" or \"dream\" is its own\nsense, which the concept layer should record as a difference, never flatten) · containment of failure\n· and honesty about the adversary: **a community with real stakes is the first time any instrument\nhere meets a motivated stakeholder** ([fractal-scales-and-temporal-frame.md](fractal-scales-and-temporal-frame.md)\n§ Three classes of adversary). First contact should happen where the blast radius is small.\n\n**Built 2026-09-19, the same day, on the founder's choice of a separate store over a hidden flag.**\nHe weighed three shapes out loud — *\"Just some hidden-or-not flag for entire extractions? Or yeah we\ncould create bubble universes of argumentation within the existing database, a scoped part of the\ndatabase, or simply a different database maybe is simplest, idk? … (This would be a separation both\nfor Postgres and for FalkorDB I guess?)\"* — and the separate store is what was built, for one reason\nthat outweighs the rest: **a hidden flag holds only while every reader remembers the filter.** The\npublic graph is read by a dozen surfaces (the front-page list, the feed, the sitemap, the agent index,\nsimilarity search, related claims, concept pages, the stats page, the question-answering layer), and\nthe linking pass would wire hidden claims to public ones, so hidden evidence would move public\nstrengths. This project has twice been bitten by a guard that held only by accident. With a separate\ngraph and a separate database the public app cannot return a blob's claim at all, because it is not\nconnected to the store that holds it, and dissolving a blob is two commands instead of an untangling.\nHis middle option and his last one turned out to be the same thing: FalkorDB holds many named graphs\nin one server and Postgres many databases, so a blob is *a scoped part of the same servers* and *a\ndifferent database* at once.\n\nWhat now exists (`deliberus/blob.py`, `scripts/blob.py`, `scripts/store_hand_extraction.py`):\n\n- **One switch, never two.** `DELIBERUS_BLOB=<name>` selects both stores, and both take the same\n  derived name. The hazard this removes is the half-configured blob: a graph pointing at the blob\n  while the database points at the public one would write a community's words into the public\n  database. The switch wins over whatever the environment file says, and pointing a single store at\n  a blob by hand is a startup error.\n- **The membrane is a precondition, in code.** `scripts/blob.py create <name>` refuses until the\n  five answers above are written down in a private file. `list` shows each blob with both halves and\n  whether its membrane is on file. `dissolve <name> --confirm <name>` removes both stores and leaves\n  the membrane file as the record that the blob existed.\n- **A hand-made extraction names its own destination.** The by-hand method (a frontier model or a\n  person doing the extraction, stored through the production path) had a new one-off script per use.\n  It now has one loader, and every data file carries a `blob` field that must match the active\n  switch, so a community's notes cannot be stored in the public graph by a forgotten variable.\n  Verified on the first use: with no switch set the loader refused; with it set, 54 claims went into\n  the blob, and the public graph, the public database and the live site were each measured to hold\n  none of them.\n- **Showing a blob without hosting it.** `scripts/snapshot_landing.py --path=… --inert-links` takes\n  a still picture of one page of the locally running app, with its links made inert, since a blob's\n  pages exist nowhere else. A keeper can carry that on their own device and show it in person, which\n  is as visible as the room the words came from.\n\n**Still unbuilt**: members-only hosting and the access gate it needs · thread ingestion with\nper-comment attribution · the fair/unfair act · export. And one thing the first use surfaced about\nthe product itself: the extraction page's relationships view showed no attribution, so a source's own\nclaims and claims added by an extraction were indistinguishable there. Fixed the same day.\n\n## 4. The blob ladder — ordered by exposure, each with the instrument that can fail\n\nThe rungs follow the growth path already in [islands-of-coherence.md](islands-of-coherence.md) § 5c\n(solo → one friend side by side → a facilitated pair → wider), located inside one community. **A rung\nis earned only by an instrument that can fail there.**\n\n**Blob 0 — my own advice process.** The keeper runs an advice process they would run anyway, as its\ninitiator, and uses a private local map to weigh the advice and write the outcome. Nobody else meets\nthe tool; they meet a better outcome statement. *Tests*: does a map help with the method's hardest\nstep? *Can fail*: slower and no better than reading the thread; or the thread's humans saw a crux the\nmap missed. *Courtesy owed*: say in the opening post that a mapping tool will help weigh the advice,\nand that anyone can ask for theirs to be left out.\n\n**Blob 1 — two chairs.** Two members who really disagree on a live community question, side by side,\nstarting from nothing, with a facilitator and the designed break\n([live-election-test-design.md](live-election-test-design.md)). The blob dissolves afterwards unless\nboth keep it. *Tests*: the one signal that would redirect the whole build, **do people point at\nclaims**, plus the four failures of deliberating groups, with a two-minute before-and-after reading\nof each person's position, since the pair leaving *more* polarised is a real possible result. A\ncommunity rich in listening practice staffs the attunement half for free.\n\n**Blob 2 — the shadow map.** With the initiator's consent, a map is kept *beside* one live advice\nprocess. The thread stays the process; the map replaces nothing. Visible to the initiator and to\nadvisers who opt in. It shows structure and asks questions. **It never says where the community\nstands**, because a summary is the one intervention measured to steer (arXiv:2605.14097).\n*The control that makes it a test*: the same thread handed to a plain language-model summary, which\nis what communities will reach for anyway, and afterwards each adviser judges both: *was my advice\nretold fairly? what did each miss? did either lean?* *Can fail*: the summary is judged as fair and\nmore useful, and typed structure did not earn its cost here.\n\n**Blob 3 — the perennial.** One question that returns every cycle gets a standing members-only map,\nand each new round starts from it. *Tests the memory claim with a countable measure no model judges*:\nof the points raised in the new round, what fraction repeat a claim already on the map, and what\nfraction are new? If repetition does not fall between rounds, the memory is not being used. *Horizon*:\nat least two cycles.\n\n**Blob 4 — the board's reasons.** For one decision the formal board must take itself, the board\npublishes its reasoning as a map: what was weighed, what outweighed what, what would reopen it. In a\ncrisis the map is written *afterwards*: the record, never the process. *Tests the institution rung*:\ndo members then attack particular premises, or still the board's legitimacy? *Risk, at full weight*:\na legible record of reasoning is a record that can be used against its authors\n([peer-review-and-the-reasoning-layer.md](peer-review-and-the-reasoning-layer.md)). Opt-in, and a\nsmall decision first.\n\n**Blob 5 — the charter.** Where a community is about to fix something nearly irreversible — a\npurpose clause, a land-use principle — a slow standing map of what the thing is *for*, feeding the\ndrafting over years. This is where *the graph never has to decide now* fits best. Only if the earlier\nrungs have earned the trust.\n\n## 5. The interfaces, one by one\n\n- **The advice process**: the map is advice management plus the outcome as a weighing. It adds no\n  step and no veto. People keep writing in the thread as they do now; structure is the machine's job.\n- **Do-ocracy**: doing is deciding, and most decisions never touch a process. So a blob is started\n  the do-ocratic way, by whoever wants one, and **introducing blobs at all should itself go through\n  an advice process**, never arrive by a board's decision.\n- **The board** (legal and financial backstop, deciding mainly in emergencies): when a question is\n  pushed back to the community, the board gains a way to *see what the community's reasoning was*\n  without polling anyone.\n- **The general meeting** (majority votes; supermajorities for the rules): a motion arrives with its\n  map as pre-reading, and the map outlives the vote as its rationale. One shipped instrument matters\n  here: a vote on a conclusion can contradict majorities on its own premises, and the map can show\n  when that has happened.\n- **Shared budgeting**: for the few large or contested proposals, the considerations that funders\n  are actually weighing (safety, permanence, cost), mapped once.\n- **A consent culture**: *only you can say whether a retelling of your view is fair* is that culture,\n  applied to arguments.\n\n## 6. What could go wrong, and the honest answer to each\n\n- **\"Paperwork kills the vibe.\"** A community built on immediate experience will rightly reject a\n  tool that asks anyone to fill anything in. Nobody is asked to. Whoever does not want the map never\n  sees it.\n- **Legibility moves power.** Making reasons visible takes influence from whoever holds it informally.\n  Expect resistance from exactly the people whose buy-in matters. **And the mirror: the map-keeper\n  becomes a new informal power.** Every structure on the map is a proposal; anyone can say the whole\n  picture is framed wrong; a second keeper is better than one.\n- **Steering.** No summaries of where people stand, ever. Structure and questions only.\n- **Exposure.** Attribution is what the map runs on and what a small community has most reason to\n  fear. Inside a blob, a claim is exactly as visible as the words it came from, never more.\n- **Other people's words through a model.** Extraction sends text to a model provider. Say so before,\n  and let people opt out.\n- **The keeper becomes infrastructure.** A blob that needs its keeper forever is a job. Blobs end by\n  default, and none is designed to depend on one person's standing labour.\n- **One more platform.** A community already tired of its tools will experience a new one as a cost\n  before it experiences any benefit. The map is a view beside the forum people already use, reached\n  only by those who want it; if it ever needs its own notifications, it has failed this test.\n- **\"They liked it.\"** People preferred the facilitation that steered them. Liking is worth nothing\n  as evidence here; every rung above names something countable.\n\n## 7. Sequencing\n\nNothing here jumps the ruled order: the reviews, then one friend, then the interface overhaul, then\nthe two-person rehearsal. Blob 0 needs no readiness beyond what exists. Blob 1 **is** the rehearsal,\nheld inside a community. Blobs 2 and up wait on the rehearsal, on thread ingestion, and on the\nfair/unfair act. What can happen before any of it costs nothing: listening for which questions a\ncommunity names as recurring, and for whether people point at particular sentences when they argue.\n\n**State on 2026-09-19, below the first rung**: one blob exists, kept by the founder alone, holding two workshop whiteboards mapped by hand. Nobody but the keeper reads it, no advice process has been run with it, and no instrument has been pointed at it, so it has earned no rung. What it proved is the mechanism: the switch, the refusal to store in the wrong place, and the measured absence from every public surface.\n\n**See also**: [islands-of-coherence.md](islands-of-coherence.md) § 5–5c (the workshop, the matching\nrule, live use from scratch) · [incentives-analysis.md](incentives-analysis.md) (when being legible\npays) · [legibility-under-power-red-team.md](legibility-under-power-red-team.md) ·\n[what-human-judgment-is-for.md](what-human-judgment-is-for.md) (consent versus ratification) ·\n[the-ratification-record-and-the-triage-feed.md](the-ratification-record-and-the-triage-feed.md) ·\n[convictions-across-scales.md](convictions-across-scales.md) § One Graph.\n\n**Sources outside the corpus**: F. Laloux, *Reinventing Organizations* (the advice process) · Loomio\nuser guide, *Advice process* (steps and the outcome statement) · J. Freeman, *The Tyranny of\nStructurelessness* (1972) · arXiv:2605.14097 (facilitation that steers).\n"}