hyperliquid-math

English

简体中文

English

简体中文

Appearance

Precision Contract ​

Status: M1 verified Last verified: 2026-07-19

Official sources: HL.DOC.TICK_LOT.2026-07-19, HL.DOC.SIGNING.2026-07-19

Fixture sources: HL.LIVE.MAINNET.M1.2026-07-19, HL.LIVE.TESTNET.M1.2026-07-19

hl.precision.decimal.canonicalize v1 ​

Source ID: HLM.SPEC.PRECISION.CANONICAL_DECIMAL.V1

  • Input is exactly { value: string }; unknown fields are invalid input.
  • Grammar is ^-?\d+(?:\.\d+)?$.
  • The raw decimal spelling is limited to 256 characters by the package-wide resource budget.
  • Reject exponent notation, +, whitespace, non-finite tokens, .5, and 1..
  • Normalize leading integer zeroes, remove insignificant fractional zeroes, emit no exponent, and map every signed zero to 0.
  • No protocol rounding occurs; rounding is empty.
  • Authority is local-exact; maturity is stable.
  • External Hyperliquid/Python/live oracles are not-supported because this is the package's own serialization contract.

hl.precision.price.quantize v1 ​

Source ID: HLM.SPEC.PRECISION.PRICE.V1

  • Input is exactly { value, marketKind, szDecimals, rounding }.
  • value is a positive plain decimal string. marketKind is perp or spot; rounding is explicitly down or up. Math does not infer an order side or execution strategy.
  • szDecimals is a non-negative safe integer no greater than 6 for perps or 8 for spot.
  • First round to at most MAX_DECIMALS - szDecimals decimal places, where MAX_DECIMALS is 6 for perps and 8 for spot. If the result is not an integer, round it to at most five significant figures using the same explicit direction. Integer results are exempt from the significant-figure limit.
  • Because prices are positive, down is toward zero and up is away from zero. A result that quantizes to zero is invalid input because it cannot represent a positive protocol price.
  • The public result is MathResult<{ value: string; precisionChanged: boolean }>:
    • valid input returns { status: "ok", data: { value, precisionChanged } };
    • invalid shape, enum, integer bound, decimal grammar, non-positive input, or a result rounded to zero returns { status: "invalid-input", issues };
    • this total deterministic operation never returns not-applicable or indeterminate.
  • value is canonical, has no exponent or trailing zeroes, and precisionChanged compares numeric value rather than source spelling.
  • Trace uses formula ID hl.precision.price.quantize, version 1, the two spec/official source references, and the DECIMALJS.10.6.0 implementation source. A changed output records each applied decimal-place or significant-figure decision in trace.rounding; unchanged output records none.
  • Authority is local-exact; maturity is stable. The acceptance limits are official protocol facts. Rounding direction is an explicit caller-provided local calculation, not a claim about server rounding.

hl.precision.size.quantize v1 ​

Source ID: HLM.SPEC.PRECISION.SIZE.V1

  • Input is exactly { value, szDecimals }.
  • value is a positive plain decimal string. szDecimals is a non-negative safe integer no greater than 8. This is the v1 supported-range ceiling observed in the dated mainnet/testnet fixtures, not an official protocol-wide maximum. Metadata above that ceiling remains unsupported until a versioned contract extends the bound; the local guard prevents unbounded Decimal quantization.
  • Quantize to szDecimals decimal places using down. Size is unsigned, so this never increases the requested exposure.
  • A result that quantizes to zero is invalid input.
  • The public result is MathResult<{ value: string; precisionChanged: boolean }> with canonical trailing-zero-free serialization:
    • valid input returns ok with that data;
    • invalid shape, integer bound, decimal grammar, non-positive input, or a result rounded to zero returns invalid-input with issues;
    • the operation never returns not-applicable or indeterminate.
  • Trace uses formula ID hl.precision.size.quantize, version 1, the spec/official source references, and DECIMALJS.10.6.0. A changed output records one down decision; unchanged output records none.
  • Authority is local-exact; maturity is stable.

Oracle boundary ​

  • The pinned official Python SDK canonicalizes valid numeric order values to wire strings, strips trailing zeroes, and rejects values that would require wire rounding, but it does not own a full protocol tick/lot validator. The M1 harness covers valid fixture values only, so it is partial wire-string evidence rather than a quantization oracle.
  • Live L2/meta fixtures prove real szDecimals and accepted response spelling. They do not prove how the server would round an invalid submitted order, so live precision coverage is partial.

M0 trace budget ​

For hl.precision.decimal.canonicalize v1, the median public-facade/internal-kernel ratio is at most 35 and the serialized valid result is at most 640 UTF-8 bytes under scripts/benchmark.mjs. The repository records portable ceilings, not machine-specific timings. Increasing either ceiling requires a reviewed architecture rationale and fresh benchmark evidence.