Experiments with no way back
A trial that quietly became the arrangement — because nobody wrote down when it ended or what would count as it failing.
What it looks like
- Something introduced as a trial is still running, and nobody set a review date.
- Nobody can say what would have counted as the experiment failing.
- Reverting would now cost more than the change did, and that argument is winning.
- Members disagree about whether it was ever agreed permanently.
Why it is hard to see
Because calling it a trial is what got it adopted, and everybody meant it at the time.
“Let’s try it for six months” is the single most useful sentence in community governance. It lowers the stakes, it converts an argument about principle into a question of evidence, and it lets a group act while it is still uncertain. Communities that cannot say it get stuck; communities that can, move.
The failure is not in the trial. It is that the sentence contains a promise nobody wrote down — six months, and then what? Reviewed by whom, against what, and reverting to what if it has not worked?
Without those, the experiment does not end. It accretes into the arrangement, and by the time anybody raises it the situation has changed underneath: people have rearranged their lives around it, a rota has been rebuilt, somebody has bought equipment. Reverting is now a real cost, and reverting is expensive becomes the argument that keeps it — an argument that would have carried no weight in month one and is decisive in month twenty.
The second problem is subtler and worse. An experiment with no failure criteria cannot fail. Whatever happens, somebody can read the outcome as success, because there was never a stated definition to check it against. So the community has run the trial, learned nothing testable from it, and kept the change anyway.
What to check
List what is currently “being tried”. Then, for each: what date was set, what would count as it working, and who is reviewing it. Communities usually find several items on the list and almost no answers to the three questions.
Ask what reverting would cost now versus at the start. If the gap is large, the trial has already become permanent regardless of what anyone intends.
Ask what was learned. A trial that produced no answer to the question it was meant to settle was not an experiment. It was a change with softer framing.
If it is already happening
Do not extend by default — decide. For each live trial: adopt it properly, end it, or set a real review with a date and criteria. Any of the three is fine. Continuing without choosing is the only bad option, and it is the one that happens automatically.
Write the review before you write the trial, from now on. Four fields, and they take five minutes: what is changing, until when, what would count as it working, and what we return to if it does not. A community that cannot answer the third has not designed an experiment.
Prefer the reversible version. Where there is a choice between a trial that can be undone and one that cannot, the reversible one is worth real cost — because the value of a trial is the option to stop, and an irreversible trial has spent that option before it starts.
And be honest about which changes cannot be trialled at all. Selling an asset, changing a legal structure, admitting a member: these are not experiments regardless of how they are framed, and framing them as trials to lower the bar is the version of this pattern that does lasting damage.
What prevents it
Experiments with a stated duration, stated success and failure criteria, and a rollback plan — plus an expiry that is automatic rather than requiring somebody to argue for ending it. The asymmetry matters: continuing should require a decision, not silence.
Sources & further reading
- 1 RCOS stress test — Irreversible Experiments — the specification this page puts into plain language
- 2 RCOS Core — Layer 6: Evolution & Adaptation — what an experiment must define, and why its expiry is non-extendable
- 3 Proposals & ratification — the phases a decision travels through, and the two most often skipped
Written by EcoHubs members, with AI assistance for drafting and editing, and reviewed by a person before publication. Facts are checked against the sources listed; anything we could not verify is marked. How this is written
The specification
This page is the plain-language version. RCOS publishes the stress test itself — the failure pattern, what is being tested, and what a compliant community would have in place.
RCOS stress test — change-emergencies/irreversible-experiments
Terms used here
- Agreements
- The written rules a community has consented to. Distinct from values: agreements say what happens, and what follows when it does not.
- Ratification
- A hold between a proposal passing and it taking effect, during which the result is confirmed before it becomes real.
- Cooling-off period
- A fixed delay between a decision passing and it taking effect, requiring no one's approval to expire.
- Decision tiers
- Sorting decisions by what they cost to get wrong, and giving each class its own method and timeline.
These travel together
Failures are rarely alone. If this one is familiar, check the others from the same lesson before concluding you have found the whole problem.
- The emergency measure that stayed
Decided fast because it had to be, never reviewed, and now it is what the community does — with a precedent attached.
- A community that forgot why
The rule survives and the reason does not — so nobody can tell whether it still earns its place, or which version is even in force.
- Sub-groups with undefined autonomy
A community grows into parts, and nobody says what the parts may decide — so each one improvises a constitution and discovers later they do not match.
- Governance with no path to change it
The rules can no longer be changed, so people stop trying — and the community loses its members quietly instead of loudly.
- The invariant nobody protected
The founding principle everybody assumed was permanent — changeable, it turns out, by an ordinary majority on an ordinary evening.