Skip to main content
Every Halyrd agent needs to know how hard it has to work before it can touch real money — and how much rope it gets when things go wrong. Evaluation presets are the answer: each preset is an internally coherent combination of risk thresholds and eligibility settings, pre-tuned so the numbers work together sensibly. Instead of sliding individual values yourself and accidentally creating a configuration that doesn’t make sense (like a daily drawdown limit that’s higher than your total drawdown ceiling), you pick a posture and Halyrd wires it up correctly.

​
When you choose a preset

You set a preset in two places:
  • During agent spawn — the Strategy Builder asks you to pick a preset before your agent starts forward-testing.
  • In Settings — you can change the preset later, which creates a new config version and starts a fresh evaluation window on that version.
Balanced is the default. If you’re not sure which preset to use, start there.

​
Preset comparison

The table below covers spot mode, which is the only mode available now. Spot is long-only: your agent buys tokens with USDT and sells back to USDT — there’s no shorting, no leverage.
FieldConservativeBalanced (default)AggressiveWhat it means
Max drawdown15%20%25%Hard ceiling on total portfolio drawdown. A breach demotes your agent to paper trading until you manually approve re-promotion.
Daily drawdown5%8%12%Per-UTC-day loss limit. A breach locks live trading until the next UTC day — your agent keeps paper-trading in the background during the lockout.
Evaluation window14 days7 days5 daysMinimum proving period. The window extends automatically if the trade floor hasn’t been met by the time it expires.
Trade floor15108Minimum number of closed trades required before expectancy is considered statistically meaningful. Below this count, the agent stays in “Evaluating — not enough trades yet” regardless of how good the results look.
Max consecutive losses345Stand-aside trigger: after this many losses in a row, the agent pauses new entries, reassesses market conditions, then resumes.

​
One-line rationale per preset

Conservative

Prove it slowly, fail safe. Longer window, tighter leash, larger sample before eligibility.

Balanced

The honest middle. Sensible defaults that work for most starting configurations.

Aggressive

Reach live faster, accept shakier evidence and more risk. Shorter window, looser drawdown limits, smaller trade sample.

​
Custom mode

If none of the three presets fit, select Custom to expose every field individually. Each field has inline helper text explaining what it controls. A built-in sanity check warns you before you save if your combination is incoherent — for example:
  • Daily drawdown ≥ max drawdown (your per-day limit would be looser than your total limit)
  • Trade floor set so low that expectancy is statistically meaningless
Custom mode doesn’t change what “profitable” means — it only changes the thresholds. The eligibility definition is fixed across every preset and both market modes: positive expectancy after fees, over the evaluation window, with at least the trade floor of closed trades, and no max-drawdown breach. Presets move the thresholds; they never redefine what the goal is.

​
A note on account size

Halyrd makes no recommendation about how much capital you should trade. Presets set risk thresholds as percentages, not dollar amounts, so they work the same way regardless of account size. If your capital is small relative to transaction costs, the fee filter will naturally reduce trade frequency — that’s the correct behaviour for your situation, not a bug. The evaluation criteria are not tuned to make any particular account size pass; they’re there to produce honest evidence.

​
Promotion mode

After your agent meets the eligibility criteria, promotion to live is manual — you review the results and flip the switch yourself. This is the only wired mode right now. Auto-promotion (where the agent promotes itself once eligible) is a planned future feature; the setting exists in the schema but does nothing yet.

​
Full schema reference

The full EvaluationConfig your agent carries looks like this: 
EvaluationConfig:
  market_mode: spot              # spot | futures (futures presets not yet available)
  preset: balanced               # conservative | balanced | aggressive | custom

  max_drawdown_pct: 20.0         # balanced default
  daily_drawdown_pct: 8.0
  max_consecutive_losses: 4

  eval_window_days: 7            # extends if trade_floor unmet
  trade_floor: 10

  promotion_mode: manual         # manual | auto  (auto = planned future feature)

  version_id: <uuid>
  created_at: <timestamp>
Every time you change your evaluation config — whether you switch presets or adjust a Custom field — Halyrd creates a new config version. Your previous evaluation track record is preserved under the old version so you always have an honest audit trail of what each configuration actually achieved.