uniteTOlove.ca

The Toronto Deliberation Stack — Explained Simply

v0.1, 2026-07-19 Date: 2026-07-19 ·

Status: v0.1, 2026-07-19 · RENDER — derived from this library's internal records, this library's internal records, and this library's internal records (all 2026-07-17). DRAFT, pre-ratification — the charter behind this plan is still a proposal. It has not been formally approved. Its own design brief says plainly: no real deliberation, no real recommendation, no real resource decision, and no real vote has happened yet. This page changes none of that. It only explains it.

This is written for anyone who wants to know how Toronto might get better at listening to itself, from a single street to the whole city. You don't need to read three dense research files to find out. It explains the plan, why it matters, how anyone would know if it's working, and where it could go wrong. It uses only research already done for this program. Nothing below is a promise.

What it is

In Toronto's last municipal election, only 29.17% of eligible voters cast a ballot. That is the lowest turnout since 1974, and it's down from 40.9% just four years before. In plain terms: roughly seven out of every ten people who could have voted, didn't. And the people who do show up to City meetings skew older, whiter, wealthier, and more likely to own their home than Toronto really is. The City's own 2020 review found this gap serious enough to act on. It now pays outside community groups to reach youth, people with disabilities, and Black, Indigenous, 2SLGBTQ+, and other under-represented residents — because open meetings alone weren't reaching them.

The Toronto Deliberation Stack is a plan to change how the city has that conversation. Its starting idea is called subsidiarity. Decisions should get made by the smallest group of people close enough to be affected by them. A question only moves up to a bigger group when the scale calls for it. The plan lays out a ladder. It runs from one person, up to a small neighbourhood-sized group called a purok. Roughly 10,000 puroks exist informally across Toronto. Next comes 158 neighbourhoods, the same boundaries the City already uses for planning. Then 25 wards, then the City of Toronto itself, then the wider Greater Toronto Area and Ontario's roughly 444 towns and cities, and beyond. At every rung, the plan spells out how people would talk to each other. It also spells out who checks the result. And it spells out how a small local conversation feeds into a bigger one, without every rung re-arguing what a smaller rung already settled.

AI plays a role here, but a narrow one, by design. The plan draws a hard line between two different things. One is AI-simulated opinion. It is used only as a labeled rehearsal tool, to help organizers prepare. It is never shown to real participants as if it were real opinion from real neighbours. The other is real deliberation: real people, in a real room or a real structured online space. Only the second one is treated as the real thing in this plan. AI can cluster comments, translate language, and summarize patterns. It is never allowed to decide anything. And it is never allowed to fake the appearance of agreement.

The plan doesn't launch as one big thing. It is built in three stages. Each stage only gets attempted once the stage before it proves itself:

a park, a transit route, a public space, nothing invented for the exercise. It would use a panel of residents chosen at random. The approach is modeled on the Toronto-based group MASS LBP. MASS LBP has already run more than 50 of these panels since 2008, and sent random invitations to over 300,000 households across its work. In-person neighbourhood meetings and a simple online comment tool would feed the panel more input.

operated in Belgium's Ostbelgien region since 2019. It has 24 members, with a third replaced every six months, and it organizes larger citizen assemblies. Lawmakers there are legally required to formally respond to what those assemblies recommend. This would pair with a published rulebook. The rulebook would state which kinds of questions get decided at which rung of the ladder.

residents more direct votes, the power to recall an elected official, or decision-making bodies chosen by lottery. These would run only as small, reversible, time-limited experiments. Each one would get an announced end date and review. None would be rolled out as permanent change.

Why it matters

The best evidence available says structured conversation really changes how people think. Researcher James Fishkin has run more than 100 of these sessions in over 30 countries. Each one gathers a random group, gives them balanced information, and lets them talk it through. The pattern holds up well — people learn real things and often shift their views. In one U.S. study, support among Republican participants for deporting all undocumented immigrants fell from about 80% to about 40% after one weekend of mixed, balanced discussion. Worth flagging honestly: that specific study comes from Fishkin's own lab, testing its own method. Studies repeated by researchers with no stake in the result are rarer than the number of published papers suggests. The direction of the finding looks solid. The exact numbers deserve some humility.

But changing minds is not the same as changing policy, and the honest evidence on that second part is far less encouraging. France ran the biggest test of its kind: 150 randomly selected citizens, nine months of work, 149 detailed climate-policy recommendations. Only about 20% were adopted as proposed. Another 51% got watered down. The rest were simply dropped. When the citizens who did the work were asked to grade the government's follow-through themselves, they gave it a 2 to 3 out of 10. This is the single most important fact behind the whole plan: a well-run conversation and real power over the outcome are two separate things. Doing the first well does not guarantee the second happens at all.

There's also a harder piece of honesty built into this plan's own foundation. Its central idea — push decisions down to the smallest workable group, then combine them back up for scale — has a documented failure mode. Local power, wherever it already exists, tends to grab a bigger share of decisions and money that get pushed down. This problem is worst in places that were already unequal before that happened. Pushing power down without real safeguards does not spread power out. It can just hand more of it to whoever already had the most influence in a given neighbourhood. That is exactly why the plan treats its safeguards as the actual product, not an afterthought (see below).

One more honest note. It's tempting to say Torontonians have simply "lost trust" in government, and that's why this plan matters. But the actual survey evidence on that point is contested, not settled. One well-known annual survey finds a real trust crisis, scoring government trust lower than business, media, or nonprofits. A separate, more careful academic survey finds something different. Trust in most Canadian political institutions has held roughly steady over 15 years. The one real decline is mostly among Conservative Party supporters, since their 2015 election defeat. The honest claim is that trust is low compared with other institutions, and split sharply along political lines. It is not that trust is uniformly collapsing.

How would we know it's working

Nothing in this plan is running yet. The charter that would authorize it has not been formally approved. The design brief says plainly: no real deliberation, no real recommendation, no real resource decision, and no real vote has been convened, run, or claimed by anything in it so far. Nine separate legal questions (see below) need real legal review before a single real councillor's office gets contacted about a real pilot.

If the plan does move forward, it is designed to prove itself in stages. Each stage names, in advance, exactly what would count as it failing — not something decided after the fact.

counts as working if the residents' panel's advice visibly changes something real and checkable. That could be a council motion that references it, a changed budget line, or a staff report that cites it. Before any design work even starts, the plan calls for a simple test: ask a councillor's office or a City division whether they'll commit, in advance, to a real response. If nobody will make that commitment upfront, the plan treats that as a fast, cheap sign the approach won't work here. It says to redesign the pilot with a different question, rather than run it anyway on hope.

that real advice can move a real decision. It is not enough that the process ran smoothly. Before building the full rotating council, the plan calls for proposing one small legal piece first. That piece is a rule requiring a City division to formally respond to proven advice within a set window. If City Council won't even adopt that minimal rule, the plan says to stay at small-pilot scale rather than build the bigger system anyway.

lottery — don't get built at all until a lawyer checks each one first. The question each time is narrow: what is the most that's allowed under current Ontario law, without needing the province's permission? If the honest answer for any one of them is "nothing meaningful is legally available," that idea moves to a research-and-advocacy track instead of a real pilot. Each trial, if it goes ahead, needs an end date, a review point, and a stated failure condition, decided before it starts, not figured out afterward. The standing council itself only earns the right to try the boldest ideas after a real track record across more than one cycle. It also needs to survive a real change in who sits on Council, or who is mayor. A system that has only ever run under one friendly government has not really been tested.

What could go wrong

Nobody has actually proven this works at city scale yet. The strongest deliberation research comes from small groups, usually 200 to 500 people talking face to face. Whether that same quality holds when a whole city, or AI-assisted tools, tries to scale it up to millions of people is a real open question. No existing study has settled it. This plan is, honestly, a real test of that question. It is not a proven method just waiting to be copied.

A well-run conversation could still just get ignored. This is named directly, using France's climate convention as the warning case. A program can listen carefully and still decline to act on what it heard, while publicly claiming credit for having listened. The plan's answer is the response-channel commitment built into every stage above. Whether that commitment holds under real political pressure is exactly what the small pilot is designed to test. It hasn't been tested yet.

Pushing power down could just relocate who captures it, not spread it out. Pushing decisions and money down, without real safeguards, makes it easier — not harder — for whoever already holds local power to grab a bigger share. This risk is worst where a neighbourhood was already unequal to start with. The plan's answer is four hard-wired controls at every rung. Every step publishes what happened to people's input. Any standing panel rotates its members, so no one gets a permanent seat. Panel members get chosen by random lottery instead of self-selection, wherever a body claims to represent people. And outside reviewers check the record on an ongoing basis, not just at launch, with minimum resourcing so lower-capacity neighbourhoods aren't simply outvoted by better-resourced ones.

Mixed groups make people more moderate; like-minded groups make people more extreme. Getting this backwards is easy to do by accident. The evidence is clear: people who deliberate only with others who already agree with them tend to end up more extreme, not more moderate. This plan's whole depolarizing effect depends on actively mixing who's in the room and keeping facilitators neutral. If any rung lets people self-sort into like-minded groups, or lets a facilitator bring their own opinion into the room, that is a real risk. The plan says plainly this is the single easiest thing about the whole design to get quietly wrong.

The AI doing the summarizing could quietly become the thing that decides what the neighbourhood said. An AI pass that clusters hundreds of comments into a readable pattern shapes what everyone downstream sees as the group's view. The plan requires every clustering pass to publish the raw input it worked from and the method it used. It must also give people a way to check whether their own comment shows up where they'd expect. No AI summary gets presented as the neighbourhood's view without that trail attached.

Coordinated fake activity could distort an online layer, or stack an in-person meeting to fake agreement. The plan treats this as a real risk, not a made-up one. Nearly half of all political social-media activity during the Brexit referendum came from automated accounts. The plan names that precedent as the real threat level its safeguards need to defend against. A quieter version of the same risk: a panel could look broadly representative while it was actually recruited through a narrow circle of already-connected people. That is exactly why the plan insists on random lottery selection over self-selection wherever a body claims to speak for a population.

"Consultation" could end up meaning nothing changed, even though everyone felt heard. This failure — inviting people in, listening politely, but never connecting what they said to who holds real decision power — is named as the hardest good-faith failure to catch. From the outside, it looks exactly like success.

The boldest governance ideas carry their own documented failure patterns, not just legal risk. The one real-world dataset on delegated ("liquid") voting — the German Pirate Party's platform — shows voting power piling up in fewer hands over time as the system matured. No evidence turned up of that power being abused. But ordinary members simply stopped staying engaged with the routine, boring decisions the system needed them to keep showing up for. No software fix solved that. Randomly selected decision panels have only one real way to be checked. The next randomly selected panel can revisit a bad call. That is slower and weaker than voting someone out. A panel that keeps changing may be more exposed, not less, to whoever controls the briefing materials and expert witnesses.

Toronto's City Council legally cannot grant itself powers that belong to the province. Ontario municipalities are creatures of the province. Most binding versions of recall, direct votes, or randomly selected decision-making bodies would need the province's permission, which this plan cannot assume it will get. Nine separate legal questions are listed as unresolved. One is what a bylaw can require of a future Council. Another is how participant data would be protected, since Ontario has no general private-sector privacy law. A third is what happens if a "vote" run inside this system gets confused for a real, binding municipal vote. The plan states outright that none of these are settled by design work alone. Real legal review has to happen first.

And underneath everything above: the charter for this whole program is still a proposal, not yet formally approved. Its own flagship design document says outright that nothing in it has convened, run, or claimed a single real deliberation, recommendation, resource decision, or vote yet.

Receipts

Everything above comes from three documents. The charter and its flagship design brief are both still marked proposal or paper-only, not yet approved:

is built to follow, including the safeguard against decisions getting relocated rather than truly shared, the requirement that any process actually connect to a real decision, and the line against ever treating AI-simulated opinion as if it were real.

three staged models, the ladder of decision-making from a single street to the whole province, the safeguards, and the nine legal questions still open.

figure above, with contested or mixed findings flagged as such, not smoothed over.

Every number and claim in this explainer traces back to one of those three files. If a figure above doesn't have a citable source in those files, it isn't in this explainer either.

There's no meeting to attend or panel to join yet — nothing here has launched, and no real Torontonian has been invited into a real room. The one honest action available right now is to watch for the plan's own cheapest test of itself: real outreach to a ward councillor's office or a City division, asking plainly whether they'll commit in advance to genuinely responding to what a citizen panel recommends. If that commitment can't even be gotten before a single meeting is designed, that's the plan's own signal that the approach needs rethinking — and it says so, on paper, before doing anything else.