Hanko, hand-lettered over a pink brush stroke
判子 · A seal you cannot take back

An agent that files receipts.

Hanko reads what crypto traders and news sites are saying, decides whether to act, and seals its reasons before the market answers. Three days later it checks itself in public, wrong calls included.

  • It spots copied posts, so five sites repeating one story count once.
  • Every decision says, in advance, what would prove it wrong.
  • Nothing can be edited later: each run is fingerprinted and timestamped on Bitcoin.
See the live record Try it live
Latest real decision loading

This card is read from Hanko's live record, which the agent updates on its own every few hours. Every decision is on the ledger.

66
runs sealed and timestamped
204
decisions graded against what happened
37 of 48
positions it took that held up
65
runs anchored in Bitcoin blocks
Misses count too: 103 decisions have been proved wrong, and every one is published. Open the ledger
How it works

It runs by itself every few hours.

Each step leaves a record you can open and check.

STEP 01

Listen

Hanko reads public sources: news sites, Telegram channels, Bluesky accounts, and RYO's market data. Every page it reads is saved exactly as it arrived.

See what it read
STEP 02

Count real voices

A press release repeated by five sites counts as one voice. Copies and reposts are set aside, and the reason is written down.

How often voices copy
STEP 03

Decide and seal

It decides to enter, pass or wait, sizes the position, and writes down what would prove it wrong. Then the whole thing is sealed with a fingerprint and timestamped on Bitcoin, so nobody can quietly change it.

Change the inputs yourself
STEP 04

Grade itself in public

Three days later it checks its own words against what actually happened. Every result is published, and nothing is deleted.

See the scorecard
Swipe sideways
The most recent grade

It marks its own homework, and shows you.

This is the last decision Hanko graded, pulled from the live record. It appears exactly as graded.

Loading the latest grade

What it said would prove it wrong
    Why you can trust it, 1 of 3 · example with made-up names

    Three voices, two opinions.

    The easiest way for an agent to be confidently wrong is to mistake an echo for agreement. Here one account reposts another, so Hanko counts two opinions.

    α
    voice_alpha
    The original post. Bullish, and fairly sure.
    COUNTED
    β
    voice_beta
    A repost of voice_alpha. Same opinion, second account.
    SET ASIDE AS AN ECHO
    γ
    voice_gamma
    Its own reasons, posted at its own time.
    COUNTED
    Independent voices
    3
    out of 3 posts
    Why you can trust it, 2 of 3 · example with made-up data

    When it knows less, it bets less.

    Here one safety check could not be reached. No special rule handles this. The missing check lowers Hanko's confidence, and lower confidence means a smaller position.

    Everything available
    3.28%
    of the book
    evidence complete: 4 of 4 checks
    + safety check: passed
    Safety check unreachable
    3.28%
    of the book
    evidence complete: 3 of 4 checks
    ? safety check: no answer, so the size shrinks

    The gap is also written into the decision, so you can see exactly what was missing and why the position was smaller.

    Why you can trust it, 3 of 3 · example with made-up data

    The promise is sealed before the answer arrives.

    Changing one word of a sealed decision changes its fingerprint, so any later edit shows.

    When it decidedSealed
    Enter TOKENA
    size 3.28% · confidence 0.66
    ! wrong if the price falls 15% within 3 days
    ! wrong if trading depth drops 30% within 3 days
    ! wrong if fewer than 2 voices still agree in 3 days
    Three days laterGraded
    Enter TOKENA
    x the price fell more than 15%
    + trading depth held up
    + 2 voices still agreed

    One failed condition is enough: the decision is graded as proved wrong even though the other two held.

    What you can do with it

    Use it without setting anything up.

    For traders

    Check an exit before you trade

    Pick a token and an amount. Hanko asks three live sources how much it would really cost you to sell, and answers in plain words.

    Try it live
    For skeptics

    Audit any decision

    Open the live record: what it decided, why, what would prove it wrong, and how each call turned out.

    Open the ledger
    For builders

    Plug it into your tools

    The same checks are available to other agents and apps, in the same shape as RYO's own tools.

    Read the docs
    For the curious: the full machine receipt behind the example above
    DECISION TRAIL · TOKENA EXAMPLE · MADE-UP INPUTS 
    
        
      
    
      
    Real output from the real engine, run at build time on fixed, made-up inputs (three fictional voices and a fictional token) so the example stays the same on every build. The live decisions are on the ledger.