How JemirUp works
Watch the walkthrough
Read it step by step
Ten minutes to read. For each step of the loop, the guide below says what the step is and which buttons you press.
What is JemirUp?
JemirUp is a workspace for finding out whether your idea is true. You write down what you believe. JemirUp works out which belief is most dangerous to leave untested, designs the cheapest way to test it, reads what you bring back from the field, and updates the record.
The thing it produces is not an answer. It is a record of what you have actually proven, and how that record changes.
One consequence worth stating early, because it surprises people: evidence that proves you wrong is good news here, and the numbers will move in your favour when it happens. More on that below.
The loop, step by step
Each step below is its own page, and the sidebar on a venture marks which one the venture is waiting on you for with a red ▶. It is a loop, not a checklist: the point of finishing it is that the next most dangerous belief is now on top — usually a different one from where you started.
Before the loop starts
You describe what you want to build in your own words — at least 40 of them.
1Thesis
/v/…/thesis- What this step is
- Your idea broken into eight pillars — who the customer is, what problem they have, what they do about it today, your solution, the value, how you make money, how you reach them, and what has to be true for any of it to work. Together they are what you are claiming.
- What you actually do
- Read all eight. Click Edit on any that is wrong, rewrite it in your own words, and save.
2What has to be true
/v/…/assumptions- What this step is
- Every claim the venture rests on, ranked. The ranking is ARS — how much it would cost you to be wrong, times how little you currently know. The top one is what the whole product then points at.
- What you actually do
- Change either number from its dropdown and the ranking recomputes immediately. Click Add one for a claim the decomposition missed — you will be asked for its impact and uncertainty, because an assumption cannot exist without them.
3Experiments
/v/…/experiments- What this step is
- A test of one assumption, with the thresholds written down before you run it. Proposals come at different rungs of the Evidence Ladder, cheapest first, so you can pick what you can actually afford this month.
- What you actually do
- How do I test this? → read the candidates → Choose this test. Then answer how will you collect?: do it yourself (optionally What do I ask? for the questions), or, on the right rungs, let JemirUp host a survey or a landing page. Doing it yourself: Start collecting, go and collect, and press I've finished collecting when you have. Each test shows a five-step track, and only the step it is on has a button. If the criteria are not right, stop the test and choose another: they cannot be edited. Suggestions stay until you choose one, so you can leave and come back; Suggest different tests asks again and uses one of your monthly analyses.
4Evidence
/v/…/evidence- What this step is
- What you collected, read against the criteria you locked in. You get a verdict, a plain-language finding, and every claim quoted back to the exact words in your own document — each quote located and checked character by character before it counts. A quote that cannot be found is rejected, and you are told.
- What you actually do
- Record what you found → paste it in, or upload
.txt.md.csv.pdf. Read the verdict and the quotes. Then Apply or Not this on each suggested change to what you believe.
5Decisions
/v/…/decisions- What this step is
- What you are doing about what you found: continue, iterate, pivot or stop — recorded with your reasoning, at the time, before you know how it turned out. That record is the part that is worth most later.
- What you actually do
- Optionally Where does this venture stand? for a read of what you have proven and what it suggests. Then Log a decision → pick one of the four → write why in your own words.
Off the loop — the case against your venture
On demand, never automatic: the strongest honest argument that this fails, across what people already do, whether the money works, whether you can reach buyers, and what might not be allowed. Every objection comes with something you could go and check this month.
How to tell which step you are on
There is only ever one ▶, and it does not march downward: it jumps to whatever is blocking, and it can move backwards. These are the rules it uses, in order — the first one that is true wins.
- 1No thesis accepted yet→ Thesis — review what was drafted
- 2A suggestion from your evidence is unresolved→ Evidence — apply or dismiss it
- 3A finished test has nothing read against it→ Evidence — add what you collected
- 4A landing page or survey is live→ Experiments — send the link; JemirUp records who visits and answers
- 5A test is running→ Experiments — you are out collecting
- 6A test is planned but not started→ Experiments — choose how you'll collect, then start; the criteria lock
- 7New evidence, and no decision about it→ Decisions — say what it means
- 8Your own evidence contradicts a pillar of your thesis→ Decisions — the most useful thing that can happen here
- 9Assumptions went stale when your thesis changed→ What has to be true — re-test them
- 10Your riskiest assumption has no evidence→ Experiments — design a test
A paused or stopped venture is waiting on nothing, so it shows no arrow at all.
The five words you need
- Assumption
- One claim that could be false and would matter if it were. “Retailers will pay 1,000 birr a month” is an assumption. “We will build a great product” is not — nothing could prove it wrong.
- ARS — Assumption Risk Score, 0–100
- How dangerous it is to leave a claim untested.
- The Evidence Ladder: Desk research, Verbal, Behavioural, Commitment
- Not all evidence is worth the same, so tests come in four kinds, weakest first. Desk research ●○○○ is what others published about others. Verbal ●●○○ is what customers say: interviews, surveys. Behavioural ●●●○ is what they do: clicks, sign-ups, pilot use. Commitment ●●●● is what they pay for: deposits, pre-sales, signed letters.
- Pre-committed criteria
- Before an experiment starts, you fix the number that decides it: “succeeds if 5 of 15 agree to pay”. Once it starts, those numbers are frozen and cannot be edited.
- Evidence Score, 0–100
- How much of what matters you have tested — not how good your idea is. A weak idea tested rigorously scores high, and should. A brilliant untested idea scores near zero, and should.
Why being proved wrong makes your risk score go down
Risk here means not knowing — not bad news.
One venture, all the way through
A real one, with the numbers it actually produced. A founder wants to build a stock alert app for small shops in Addis Ababa — shopkeepers track stock in a paper notebook, so they only notice something has run out when a customer asks for it and leaves.
Describe the idea
They write a paragraph in their own words, including the part they are least sure of: “I think they would pay around a thousand birr a month.”
Get a thesis back, and correct it
Eight pillars come back. Business model: “Flat monthly subscription of 1,000 ETB per shop, by mobile money.” Each one is editable — this is the founder's thesis, not the system's, and nothing is saved until they accept it.
See what is riskiest
Ten assumptions are extracted and ranked. Top of the list: “Independent retailers will pay 1,000 ETB per month” — ARS 100, 0% evidence. High impact (the whole business model rests on it), high uncertainty, nothing gathered yet. That is the definition of dangerous.
Design a test — and lock the numbers
Three options come back at different rungs. The founder picks the Verbal one: visit 15 shops, state the price plainly, write down the first reaction.
Succeeds if 8 of 15 accept the price without objecting. Fails if 5 or more object on hearing it. They press Start collecting — and those two numbers freeze. They cannot be edited again.
Go and do it
Nothing happens in the app during this step. The status says you're collecting because you are — walking Merkato, talking to shopkeepers, writing down what they said. This is the part no software can do for you, and it is the only part that produces evidence.
When the fifteen visits are done, press I've finished collecting. That records that you finished collecting. It does not record that it worked — there is deliberately no button for that.
Have the evidence read
They paste in their twelve write-ups. The verdict: contradicted. Seven of twelve objected to the price against a pre-committed failure threshold of five, so the test fails on its own terms.
Every claim is quoted: “A thousand? No. That is what I pay a boy for a week of work.” — and clicking it opens the transcript at that exact sentence. The footer says 8 of 8 quotes were found in your documents and checked word for word. If a quote could not be found in what you uploaded, the whole analysis is refused rather than shown to you.
Watch the numbers move
The pricing assumption goes 0% → 35% evidence, and its ARS falls 100 → 65 — down, because the unknown is smaller now. Business model turns partly validated. The venture reaches stage evidenced.
And the riskiest assumption becomes a different one. That is the product doing its job: you now have a new most-dangerous belief, and the loop starts again on that.
Decide, on the record
The evidence also pointed somewhere useful: the owners who already pay an accountant did not object. So the founder logs an iterate decision — narrow the segment to shops that already buy bookkeeping, and re-test at a lower price.
That decision is permanent and dated. In six weeks, when someone asks why the price changed, the answer is one click away with the twelve interviews behind it.
Who does what — you, or the system
This is the thing people get wrong most often, and the confusion is reasonable in a product full of AI. The short version: it proposes, you decide, and the fieldwork is always yours.
| Step | Who acts | What that means |
|---|---|---|
| Drafting a thesis | System proposes | You edit and accept. Nothing is saved until you do. |
| Ranking assumptions | System computes | Plain arithmetic, not a model. Same inputs, same answer, always. |
| Designing an experiment | System proposes | You choose one of the suggestions. To change its criteria, stop it and choose another. |
| Running the experiment | You | Nothing is happening on our side. This is the fieldwork. |
| Saying you've finished collecting | You | Records that collecting finished — never that it worked. |
| Reading the evidence | System | Quotes are verified against your documents word for word. |
| Changing a belief | You | No AI output ever changes your thesis on its own. |
| Logging a decision | You | The recommendation is advisory. Disagreeing is recorded as such. |
What it costs you (nothing)
Founders never pay for JemirUp. There is no subscription, no trial, no credits to buy. Institutions — incubators and accelerators — pay for the cohort view, and that is the entire business model.
If you're in a programme
An incubator can invite your venture to a cohort. They cannot add you — you accept, or you do not, and you can leave at any time. There is deliberately no way for an organisation to enrol you or to change what you share.
- Scores only
- They see your stage, score and velocity. Nothing about what the venture is.
- Thesis and scores (the default)
- They also see your pillars and assumptions — but never the evidence behind them.
- Everything, including evidence
- They also see your uploaded documents and the quotes drawn from them — except anything you marked sensitive. Recorded interviews contain words a customer gave to you, not to your incubator, and marking a document sensitive keeps it and everything drawn from it out of their view. Your scores do not change either way.
Who else can see your work
Other founders: nothing. A programme: only what you chose above, and only after you joined. The people who run JemirUp: how the platform is used (your name and email, when you were last here, your venture's name, stage and Evidence Score, and which programme it is in), never what you are building. Your idea, assumptions, experiments and evidence are not shown on any of our screens.