A thought you can be wrong about¶

Epic #560 — the fields (#570), the resolution (#571), the mark on the timeline (#572) and the wall (#573).

The manifesto says thinking should help you understand better and live better. Everything ZettelFlow records is on the first half of that sentence, and all of it is internal: a note changed, a link appeared, a claim was rewritten, a verdict was given.

Nothing in this product had ever been checked against what actually happened. The timeline made it visible: every event it could draw was in the past. There was nothing here you could be wrong about — and something you cannot be wrong about is something you cannot learn from in the strongest sense of the word.

A wager is the smallest object that changes that:

a claim  +  what you expect to see  +  by when
        ↓            (time passes)
   "what actually happened?"        ← asked before you are shown what you predicted
        ↓
   expected / happened, side by side
        ↓
   the claim stands, or it changes

No calendar, no sensor, no service, no integration. You are the instrument.

Naming what you expect¶

The claim door has two more optional lines: and what you expect to see, and the day you expect to know by. Skipping them is the normal path — most claims are not wagers, and a form that insists is a form nobody uses twice.

They are plain frontmatter on the note, beside the claim:

claim: microservices move complexity rather than add it
expect: two teams stop waiting on each other
by: 2026-12-01

A prediction you cannot open with your own tools is not yours, so you can write, edit or delete them by hand and the product reads them exactly the same way.

Both halves, or it is not a wager. An expectation with no date cannot be resolved, and a date with no expectation is a reminder — which is the thing this epic refuses to become. It also means a note whose by: means an author is left alone.

A day you typed is a day where you are. The horizon is read as the start of its own local day. 2026-12-01 parsed as UTC midnight would come due the evening before for everyone west of Greenwich, and never before its horizon is the promise a user would notice breaking.

What actually happened¶

When the day arrives, Home offers it — the same single line a returning claim uses, because at most one thing at a time holds across both. Open it and the question comes first, with what you predicted nowhere on screen:

What actually happened? Answer from what you saw. What you expected stays shut until you do.

You write. The observation is saved as a thought about the note — it survives first, the same ordering a withdrawn claim has — and then both sentences appear together:

You expected — two teams stop waiting on each other What happened — they still waited, but on a different thing

No tick, no cross, no colour of success or failure, and no word about which one you wrote. Below them the claim's own three answers, unchanged: it still says this · it says this now · I no longer hold this. Choosing one clears the horizon in the same write, so a wager resolves once.

It says this now asks for the new sentence in its own box, because here the text you typed is an observation — committing it would write "they still waited" into the note's claim.

One mark in the future¶

With the evolution timeline on, the horizon is the first thing this axis has ever drawn that has not happened:

Expected — you said you would know by then two teams stop waiting on each other

The axis is an ordered list, not a scaled one, so a date two years out costs nothing and moves nothing. The row is muted and says the same thing at every distance, before and after the day arrives.

What it will never be¶

• No accuracy, no hit rate, no calibration, no correct or incorrect. A hit rate is an invented score about the user, which §XI refuses as a metric and §XII refuses as a judgement — and this project deleted its telemetry on purpose. A self-scoring dashboard is telemetry pointed inward. A scan over all of src/ and both locale files holds the line, and carries its own planted strings so the scan is proved rather than trusted.
• No habits, no streaks, no repetition. A wager resolves once.
• No reminders and no notifications. One line in Home, the same one a claim return uses.
• No list of open wagers with a count. There is no aggregate over wagers anywhere in the model, so there is nothing a surface could render.
• No external data. No calendars, wearables, trackers or imports. The observation is a sentence you write, and that boundary is not negotiable here.
• No AI judging the outcome. A model deciding whether you were right is the single most §XII-hostile thing this product could do.

The two sentences beside each other are the entire feedback, and they are enough: nobody needs to be told which one they wrote.

Honest limits¶

• Nothing records that a wager was resolved. It stops being due because its two fields are gone, not because a flag was set — and a reader looking for a resolved field should not add one.
• A wager needs a thinking-space folder, because the observation is written as a thought. With none set, wagers are not offered at all and the hand-opened return says why in one sentence.
• A horizon that passed unresolved changes nothing. It is offered the same way, once, whenever you next look: no escalation, no repetition, no ageing.
• A renamed note keeps its wager, because the fields live in the file — while the moves and judgements about it deliberately do not follow the rename. That asymmetry is correct, and it is stated here rather than fixed.
• This was shipped on a design hypothesis. No vault anywhere contained a wager when it was written, so unlike the neighbourhoods or the gaps, its premise could not be checked against data that already existed. The first real evidence is someone setting a horizon two days out and resolving it.

README vocabulary for this page: Wagers.