Theory of change

Internal team document. This page is not indexed and nothing in the app links to it.

• Status: merge draft. Not adopted — see The decision.
• What was merged: the mission choice from THEORY_OF_CHANGE.md (14 June 2026) + the mechanism agreed at the July 2026 meeting.
• Last updated: 16 July 2026.

Why this exists

We had two documents titled "theory of change" with two different frames: the June one picked a mission, the July one picked a mechanism. Two documents about one thing is the three-missions problem in miniature. They are merged here because this is one bet, not two.

This is not a manifesto and not a public promise. It is a bet, written down so it can be tested and, if necessary, conceded.

Table of contents

• The choice everything rests on
• The hypothesis
• The chain, and where our boundary runs
• The load-bearing assumption
• What would falsify the hypothesis
• Who we equip
• What counts as equipment
• What we do not promise
• The third variable: mandate
• Why we think we have been running on this hypothesis all along
  • What this does not prove
• Blind spots
• Signs of progress and of stalling
• Open questions
• The decision
• Nearby
• How to change this document

The choice everything rests on

Mission: civic enforcement — "the laws exist, the enforcement doesn't." Mechanism: lowering the friction of civic action.

Together: we lower the friction of civic enforcement.

The other two missions are not discarded: "the experience of civic action" becomes the result of each cycle, and "self-integration of newcomers" a side effect, not a goal.

Why this is one bet and not two things glued together. The phrase "there are good laws, there is no enforcement" is from December — one of the three missions found inside our own "fixable city" text. The June document chose that one. The July meeting chose the mechanism. Both halves came from the same place: the mission answers "why", the mechanism answers "how". Until now they simply sat in different files.

The hypothesis

If we lower the transaction costs of civic action — hand a person a ready-made route (knowledge: where to go, how, what is legal; action: generate and file) — then an equipped citizen will act more often and more easily than an unequipped one.

Because the barrier to participation is not motivation, it is friction: the person is not passive, they are simply not willing to pay the participation tax.

Provided that friction — and not fear, apathy, or disbelief in the outcome — really is the main restraining factor.

The chain, and where our boundary runs

lower friction → citizen is equipped → acts more often / more competently / more easily ‖ the mass of high-quality civic action grows → the cost of ignoring it rises for the official → … → quality of governance

Everything to the left of ‖ is our zone: we build it and we measure it.

Everything to the right is the democratic-accountability assumption. We hold it as a hypothesis about the world, not as our promise. We do not close the right-hand side ourselves — we hand it to equipped citizens.

That boundary is not rhetorical. It is drawn in code. See below.

The load-bearing assumption

The binding constraint on participation is friction, not motivation.

Everything stands or falls on this. If people do not act because they are afraid, do not believe, or do not care, then removing friction changes nothing and the harness is useless no matter how clean it is. This is the first thing to test in a pilot, before any scaling.

What would falsify the hypothesis

The hypothesis is refuted if equipped citizens do not act noticeably more often and more easily than unequipped ones.

Not "the city did not get better" — that is not our claim. Specifically: we handed over the route and behaviour did not shift. That would mean the load-bearing assumption is false, and the project needs rethinking from the mechanism, not more polish on the tool.

Who we equip

People for whom the motivation question is already settled: they have their own answer to "why", and they do not need convincing that acting is worth it. For them it is binary — do nothing and nothing will certainly happen; try and something might. That difference is enough for them. What they lack is not meaning, it is a route.

We deliberately stay out of motivation. Betting on people having abundant energy and activist stubbornness to push through the rakes is a bad bet: most people are simply not like that. Even a professional citizen is not the same thing as an activist.

This is a question of work order, not of a values hierarchy. There is some unfairness in the choice — why start with those who already have resources? The answer: it is not about fairness, it is about practical survivability. If it does not work with people who are already ready, it will not work with anyone. Once the mechanism runs and has been tested, motivation can come back on the table — and then it will mean something, because there will be somewhere to bring people to. An inspired person arriving at an empty field steps on the same rakes as everyone before them and burns out. Motivational work without equipment spends people.

What counts as equipment

• Knowledge — where to go, how to file, what is legal.
• Action — generate and submit. Not "explain how to submit", but submit.
• Battle-testing — "this axe has been in ten battles, and in eight of them the warrior came back." A tool whose track record is known — how many times it worked and when it did not — is expensive and valuable knowledge in itself. It follows that equipment is versioned: v2 is built on data about how the warriors carrying v1 died.
• Community — solitary action is itself friction. Measuring street trees alone is awkward; in a group it is normal. So rallies, public talks, the network and the chat are also friction-reduction tools, not merely "events".
• Education — on the friction side, not the motivation side. It gives a person the ground to say not "there was a tree here and now there isn't, how awful", but "clear-cutting a whole street is wrong, every other tree is acceptable, replacement should be gradual". It is an instrument of action, addressed to rational agents. Motivation is a different thing: persuading someone that acting is worth it in the absence of any evidence that it works. That is out of scope.

What we do not promise

Point zero of every piece of communication we make.

We give no guarantees. We are not a company selling an outcome for money. The right-hand side of the chain does not depend on us — so we cannot and will not promise it.

The formula: we do not guarantee that this tool will solve your problem. We have tested it on this many cases and we see that it works with this probability. The tool exists; from here it is at your own risk. We are simply handing you the possibility.

The same follows for how we report results: as a whole. Not "they told us to get lost" and not "it all worked", but — what was actually achieved, which hypotheses we tested, what we concluded, and what we do next. An honest post-mortem lowers friction by itself. "They told us to get lost, let's go cry" raises it.

A risk we knowingly accept: people will read us carelessly and hear "they solve my problems" instead of "they hand out tools". This cannot be removed, only lowered to a level we agree to accept. Whoever owns communication sets that level.

The third variable: mandate

This section was in neither the June document nor the July meeting. It matters more than the rest of the page.

The hypothesis above describes the citizen. It does not describe us.

The core team is the most equipped civic actor in this city. A site, bots, typed clients for three government APIs, EKENG signing, response tracking. Zero friction. Motivation present. Over 7.5 months of the org chat: 820 proposals → 40 things done. A 5% conversion rate. Four times more deferred than done.

If friction were the binding constraint, we would be the most active organization in Yerevan. We are not. So for us the binding constraint is something else — and our frame has no slot for it.

SELF_DESCRIPTION.md names it: a vacuum of mandate. Not "who are we", but "who has the right to decide". The right lies on the table and nobody picks it up: Sergey — "not mine to decide"; Alex — "I delegate my vote"; Kira carries the only explicit mandate, and it covers recruiting; Victor refused the role of power source before the start.

The cleanest measurement of that variable is one person over one period:

Same motivation, same equipment, same person. The only difference is whether anyone's permission is required.

Two consequences. First: the hypothesis is true about citizens and incomplete about organizations — an honest limit of its applicability, not a defect. Second: no document, this one included, fixes a mandate vacuum. Which is why the page ends in blank lines rather than conclusions.

Why we think we have been running on this hypothesis all along

The hypothesis was not invented at the meeting. It was reconstructed from what had already been built over eight months. Easy to check: walk the repository history and look for a single step on the motivation side. There is none. Everything built lands to the left of ‖.

Add the Telegram bot, the mini app, short links and QR codes on paper posters — that is access friction. The public API and self-service keys are the arms shop taken literally: the tool is handed to whoever is on the field.

Three reasons to treat this frame as working rather than merely elegant:

1. It is descriptive, not aspirational. It compresses 1102 commits; it is not a plan. We did not adopt a framework — we named a bet we were already making without being able to say it out loud. Frames like that do not collapse on first contact with the work, because the work already went through them. That is what separates it from the twelve imported methodologies before it: those had to be applied to something; this one does not — it is already running.
2. The ‖ boundary is physical. The code stops at "filed and tracked". It cannot make an official reply. "We do not close the right-hand side ourselves" is not a hedge — it describes where the machine ends.
3. The falsification instrument is already in production: the funnel submitted ≥ accepted ≥ responded, a tracker per channel, OCR and translation of the authority's reply. The comment in app/[locale]/reports/page.tsx sets it out: terminal outcomes get null so the renderer does not pretend the timeline is still moving. "Promise no outcome" is wired into the code, not only the tone.

What this does not prove

The repository shows the machine is built and can measure itself. It does not show that an equipped citizen acts more often than an unequipped one: no comparison, no control group, unknown counterfactual. The load-bearing assumption remains untested.

Separately: eight months of consistent movement along the hypothesis is not the achievement of a group decision. It is what one person built while the group discussed who it was. See The third variable.

Blind spots

• The silent ones. Someone looked, did not understand, left — no trace, no thanks, no complaint. We do not know how many. Feedback share can grow but never reach 100%.
• The counterfactual. We do not know whether a person was acting before they got the tool.
• No comparison against the unequipped. Which is exactly what the falsification clause is built on.

The testing agreement: collect artefacts and do not follow the hypothesis blindly — record both what confirms it and what does not. The moment a piece of equipment exists, we work out how to learn whether it worked.

Signs of progress and of stalling

We are moving if, within 90 days: the registry is published with ≥15 cases; ≥1 package of official appeals is filed; a public response track is running; ≥1 attributable reaction from an authority is recorded; the "meeting → decision → execution" cycle has run ≥3 times; ≥1 decision was taken under the rule and not reopened.

We are stalled if: the pilot topic was changed or widened mid-flight; materials settled unpublished; we returned to arguing "who are we"; we launched another frame instead of a campaign; a decision was put on the table and not taken.

This page, until "The decision" is filled in, belongs to the lower list.

Open questions

1. How do you measure the person who did not act? How to separate "the tool shifted behaviour" from "the people who would have acted anyway showed up". Without this, falsification is not operational.
2. What do we do if the hypothesis fails? Default: "rethink from the mechanism" — but what that means in practice is undecided.
3. Cadence of education work. A one-off review changes nothing. The bar is monthly, or every two months; the format is unchosen.
4. Separating the brand from the community, so members' activity does not read as the project's position.
5. Media dependency. Our strongest analysis goes out through someone else's channel. Acknowledged, not formalized.

The decision

This document does not take effect until three lines are filled in. While they are blank it is a description, not a decision, and its fate is the same as every frame before it.

• Owner of the first external product: _______
• Date of first publication: _______
• Decision rule: _______ (e.g.: 3 of 4 in favour; every decision has an owner; silence = consent within 48 hours; an adopted decision is not reopened without a new decision)

The pilot inherited from June, unless replaced: one campaign — air, illegal waste burning. ≥15 documented cases on our own infrastructure, ≥1 package of appeals to the responsible bodies, a public response track, ≥1 attributable reaction. 90 days. The topic may be replaced — but only with one, and carried to the end.

Nearby

• /dashboard/foundations — the team's internal sense-making tool. It answers the motivation question inside the team — the very question we put out of scope on the outside.
• /escalation-graph — a machine-readable map of where to go when a body refuses.

How to change this document

Change it when: the load-bearing assumption has been tested, in either direction; the ‖ boundary has moved; a new class of tools entered the definition of equipment; the mission was replaced by a decision under the rule.

Every change carries a date and a reason. If in six months this document describes something other than what we do, the document is at fault. Fix the document.