Teaching your agent

Strategies and reasons

Tagging each trade with a strategy and a written reason, and how reasons stay private.

A trade on its own says what you did. A strategy tag and a written reason say why. Kinfield asks for both on every order and keeps them with the trade for good, so the record holds your intent alongside the market context and the outcome.

Strategy tags

Every order carries exactly one strategy. Pick the one closest to your idea:

TagUse it when
MomentumYou are following a move that is already under way.
Mean reversionYou expect a stretched price to come back toward its average.
BreakoutPrice is leaving a range or level and you expect it to keep going.
RangeYou are trading between support and resistance in a sideways market.
News / eventA headline, listing, unlock or other event is behind the trade.
HedgeThe trade offsets risk from another position.
OtherNone of the above. The agent's own trades are tagged Other.

Written reasons

The reason field is free text, up to 1,000 characters, and optional. A sentence is plenty: what you saw, what you expect, and what would prove you wrong. For example:

Momentum: broke above the morning high on rising volume.
Expect a run to 104k. Wrong if it closes back under 101.5k.

You can't edit a reason after placing the order. That is on purpose: a reason written after the outcome is known is not the same data.

Reasons stay private by default

Your written reasons are private unless you choose to share them. When an order is placed, Kinfield picks a random salt for it and records only a commitment to the reason on the ledger:

reason_commitment = sha256(salt + "\n" + reason)

The commitment proves the reason existed when the trade was placed and was never changed, without revealing it. The random salt matters: a plain hash of a short reason could be guessed by hashing common phrases; a salted one can't.

  • Sharing reasons off (the default): the open dataset publishes the commitment only.
  • Sharing reasons on: the dataset also publishes the reason and its salt, so anyone can recompute the commitment and confirm the text is exactly what was recorded at the time.

The agent's own trades carry machine-written notes (its confidence, stop and target), which are always published.

What they are used for today

Your agent currently learns from the market context and outcome of each trade. Strategy tags and reasons are recorded, kept on the ledger and published if you share, but not yet used in training. They are there so future models, and anyone using the dataset, can learn from intent as well as action.

Read on: Market context describes what is recorded about the market, and Your agent how it is used.