> For the complete documentation index, see [llms.txt](https://doc.lockerprotocol.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.lockerprotocol.com/hyper-agentic-trader/calibrate.md).

# Measuring the method

Measure a method, or any strategy by its kind, on past candles, before you trust any of it.

`lpa pilot calibrate` walks past candles and reports what the method would have done. No model is involved, nothing is signed, and nothing is sent anywhere but Hyperliquid's own public data.

This is where the figures of the [charter](https://doc.lockerprotocol.com/hyper-agentic-trader/charter.md) and of a [strategy](https://doc.lockerprotocol.com/hyper-agentic-trader/strategies.md) come from.

```sh
lpa pilot calibrate --symbol BTC --interval 4h
lpa pilot calibrate --symbol BTC,ETH --steps 0.5,0.75,1 --ladders 1-1-2-3-5,1-1-2-2-2
lpa pilot calibrate --symbol BTC,ETH --strategy grid-long --levels 6,8,12 --starts 6
```

Two ways to call it:

* **without `--strategy`**, it measures the positions by steps: every combination of candidate step, ladder and entry zone, each entry walked on its own, which says how often the method's target comes and how long it takes;
* **with `--strategy <name>`**, it measures that strategy as it plays, by its kind of method: by steps with its own figures (as one account with `--account`), or a grid, a range or a trade with a stop, each by its own measure. The strategy's figures are the measure's defaults, under the charter in use, and a flag you give still says what it says. The words and the file carry its name.

## What it measures by steps

Over the span you give it, it opens one candidate entry every four hours, long and short, for every combination of candidate step, ladder and entry zone. It then walks the candles that followed with the venue's own funding, filling the steps the price reached and closing on the exit limit, and counts what happened.

For each combination it reports:

| Figure                        | What it says                                                               |
| ----------------------------- | -------------------------------------------------------------------------- |
| Share of targets reached      | within 7, 30, 90 and 365 days                                              |
| Median delay                  | how long a position took to reach its exit                                 |
| Median steps                  | how many steps of the ladder filled                                        |
| Worst excursion               | how far the price went from the average                                    |
| Capital blocked               | in dollar-days: what the method ties up                                    |
| Mean result, funding and fees | the money side                                                             |
| Net per attempt               | with a resting first step, what one attempt made on average, filled or not |
| Entries still open            | the ones the span never closed                                             |

The table is sorted by the share of targets reached **within 30 days**. Its columns are `SIDE STEP LADDER ZONE FIRST EVENTS N FILL% 7d% 30d% 90d% 365d% MED d STEPS WORST% BLOCKED $d NET $ NET/TRY $ FUND $ FEES $ OPEN`.

The figures are also written to a file:

```
~/.lpa/pilot/calibration/<market>-<date>.json          with hourly candles
~/.lpa/pilot/calibration/<market>-4h-<date>.json       with four-hour candles
```

A resting first step, a margin through the level, an event split, `--account` and a strategy's name each add their own tag to the name, such as `BTC-4h-thru5-<date>.json`, so two measures of a day do not overwrite each other.

## The flags

| Flag           | Default                                                                                 | Meaning                                                                                                                                                                                                                              |
| -------------- | --------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| `--symbol`     | required                                                                                | one market, or up to four separated by commas (`BTC,ETH`)                                                                                                                                                                            |
| `--days`       | `365` with 1 h candles, `730` with 4 h candles                                          | how far back to measure, 7 days at least; 730 at most with 1 h, 900 with 4 h                                                                                                                                                         |
| `--interval`   | `1h`; `4h` for a strategy with a stop, and `1h` when that strategy reads the hourly RSI | `1h` or `4h`. These are the two intervals                                                                                                                                                                                            |
| `--steps`      | `0.5,0.75,1`, or the strategy's                                                         | candidate distances between steps, in daily ATRs, from 0.1 to 3, separated by commas                                                                                                                                                 |
| `--ladders`    | `1-1-2-3-5,1-1-2-2-2`, or the strategy's                                                | candidate ladders, each written with dashes, 1 to 8 steps of 0.5 to 10                                                                                                                                                               |
| `--zones`      | `100,33`, or the strategy's                                                             | candidate entry zones, in percent of the seven-day range, from 10 to 100. 100 is no zone                                                                                                                                             |
| `--first`      | `0`                                                                                     | where the first step goes: `0` is at market, `0.25` or `0.5` rests that fraction of a step away from the price                                                                                                                       |
| `--wait`       | `24`                                                                                    | hours a resting first step waits before the attempt is abandoned and counted unfilled, 1 to 168                                                                                                                                      |
| `--through`    | the charter's `throughBps`                                                              | basis points the price must go **through** a level before it counts as filled, 0 to 100                                                                                                                                              |
| `--events`     | `0`: no split                                                                           | with a number of hours, 0 to 72, each combination is measured twice: the entries a high-impact USD release follows within that window, and the clear ones. The span then starts in December 2025, where the official schedule starts |
| `--every`      | `4`, and `24` for a strategy with a stop                                                | hours between candidate entries, 1 to 24                                                                                                                                                                                             |
| `--equity`     | `5000`                                                                                  | the equity the ladders are sized on, 10 to 1,000,000                                                                                                                                                                                 |
| `--account`    | off                                                                                     | run one account through the span instead of independent entries (below)                                                                                                                                                              |
| `--policies`   | the three, or the strategy's                                                            | with `--account`: parts `long-any`, `long-zone`, `short-zone` and `long-none`, joined with `+`, separated by commas                                                                                                                  |
| `--concurrent` | `1,2`, or the strategy's                                                                | with `--account`: positions at once, one per market, 1 to 4                                                                                                                                                                          |
| `--strategy`   | none                                                                                    | measure that strategy by its kind of method (below)                                                                                                                                                                                  |
| `--levels`     | the grid's own                                                                          | with a grid: counts of levels to compare, 2 to 20, separated by commas                                                                                                                                                               |
| `--starts`     | `1`                                                                                     | with a grid, a range or a stop: starting days thirty days apart, up to 12                                                                                                                                                            |

## About the span

The exchange keeps about 5000 candles of each interval, so:

* **1 h candles** reach about **207 days** back. The words of the answer say how many days were really measured.
* **4 h candles** reach about **two years**, which matters: hourly candles never held a falling market, four-hour candles do. With `--interval 4h` the entries sit on a candle's end, the funding and the capital blocked count the candle's four hours, and the week's range reads the same hours.

A span is read in snapshots of the exchange's ceiling, so a year costs only a few requests.

The funding is read hour by hour to the end of the span, and a page the exchange refuses as too frequent is asked again. When a market's funding still cannot be read to the end, the answer says so under the table, in `pilot.log` and in the measure's file:

```
funding of HYPE read until 2025-12-22 08:00 UTC (HTTP 429): its last rate counted after it
```

Read a measure with such a line as an approximation for that market: its funding after that date is the last rate read, not the real one.

## Reading `--first`, `--through` and `--events`

These three flags answer questions the method cannot answer by guessing.

**`--first`: at market, or resting?** A resting first step pays the maker's fee instead of the taker's and often gets a better price, but it is not always filled. Each row says its fill rate and its net per attempt. Measured on two years of BTC and ETH on 3 October 2026: a resting first step saves the taker's fee and a few hours, but loses 17 to 39 % of the attempts, and the net per attempt is best at market everywhere. That is why the first step goes at market unless the trader names a limit.

**`--through`: how honest is a fill?** A maker's order sitting at a level is not surely filled when the price merely touches it. `--through 5` counts a fill only once the price went five basis points through the level. Measured on the same two years: five basis points lower the share of targets reached within a week by about one point (BTC, long in the lower third: 96.96 % to 95.79 %; ETH: 98.59 % to 98.29 %) and within thirty days by half a point, raise the capital blocked by a fifth, and leave a few more positions open. Ten basis points double that. Every measure counts a fill with the charter's `throughBps`, five basis points by default, a stop's target included (its stop counts at the touch, a market order); the replays and the brain's judgement of a ladder, a grid and a range count it with the margin the position was opened with, and a stop's replay and judgement count the touch.

**`--events`: does a release hurt?** `--events 12` splits every entry into the ones a high-impact USD release followed within twelve hours and the clear ones. Measured on BTC and ETH to October 2026: the longs opened near a release reached their target as often as the clear ones, while the shorts stayed stuck 6 % of the time against none. That is where the charter's `eventSides` comes from.

{% hint style="info" %}
**Read the share of targets reached as a share, not as a profit.** A method that reaches 98 % of its targets can still lose money on the 2 % that never arrive, which is why the worst excursion and the capital blocked are in the table beside it. A measurement on past candles is not a promise about the next trade.
{% endhint %}

## The account table

With `--account`, the measurement stops being about independent entries and becomes about one account:

```sh
lpa pilot calibrate --symbol BTC,ETH --interval 4h --account --policies long-any,long-zone --concurrent 1,2
```

It runs the span hour by hour on every market named, opens when a market is free and the policy allows, holds at most one or two positions at once (one per market, the room before liquidation shared between them), sizes each ladder on the equity of that day with the first step and the first ladder you named (`0.5` ATR and `1-1-2-3-5` without a flag), follows the steps and the exits on the candles with the venue's funding, and reinvests what the charter says.

It then reports what a statement would, in the columns `POLICY AT ONCE DAYS CYCLES L/S OPEN STUCK RETURN% MAX DD% WORST LAT% IN MKT% MED d MAX OPEN FEES $ FUND $ BLOCKED $d EQUITY`:

| Figure                         | What it says                                    |
| ------------------------------ | ----------------------------------------------- |
| Cycles                         | how many positions were built and closed        |
| Return                         | what the account made over the span             |
| Worst fall of the equity       | including the latent result of open positions   |
| Worst latent of an hour        | the deepest an hour ever looked                 |
| Time in the market             | how much of the span the account held something |
| Median cycle                   | how long a position lasted                      |
| Fees, funding, capital blocked | what it cost to be there                        |

These are the independent entries' figures turned into what one account would have earned and endured. The two tables answer different questions: the first says whether the method's exits arrive, the second says whether a person could have held it.

## Measuring a strategy

```sh
lpa pilot calibrate --symbol BTC,ETH --strategy <name> [--starts <n>]
```

`--strategy` names a shipped strategy or one of yours, resolved under the charter in use, and the measure follows its kind of method. A strategy that follows the leaders is measured without that filter, and the answer says so: the venue keeps no past positions of the top accounts, so the entries do not wait for their side.

`--starts <n>`, up to 12, runs the same accounts from n starting days **thirty days apart**, each to the end of the span, and a second table says the spread of what they gave. The last start must leave sixty days at least, and a count that does not is refused with the figure: *`--starts 8` over 200 days leaves 30 days to the last start, under sixty.* Read a single start as a draw, never as a result.

A flag that measures another kind is refused by its name rather than taken in silence, with the line that measures this one instead:

```
$ lpa pilot calibrate --strategy grid-long --symbol BTC --steps 0.5
Error [BAD_ARGUMENTS]: --steps measures positions by steps, not grid-long: a grid is measured as one account a
market and a count of levels.
Hint: `lpa pilot calibrate --strategy grid-long --symbol <BTC[,ETH]> [--days <730>] [--interval 4h]
[--equity <5000>] [--through <bps>] [--levels 6,8,12]`.
```

`--starts` outside a grid, a range or a stop is refused the same way: a measure by steps already takes an entry every `--every` hours.

### By steps

The strategy's step, ladder and zone become the measure's defaults, and with `--account` its sides and its count of positions at once. A DCA (`ladderMode` by the time or by the value) and an exit by parts (`exitTiersPct`) are always measured as one account, because the entry-by-entry measure knows neither moments nor parts. The tables are the two above.

### A grid

```sh
lpa pilot calibrate --symbol BTC,ETH --strategy grid-long --levels 6,8,12 --starts 6
```

One account a market and a count of levels. Each grid is set on the range of the hours before it, read over the horizon the day reads: the dossier's nearest to the strategy's `gridRangeHours` (the last day, seven days or thirty), a strategy of 336 hours being measured, as it is played, on seven days, and the line saying which. Neither the regime nor the events are judged by this measure, and the line says so too: a grid that opens in some regimes only is measured on every day. Each grid is then walked to its end on the candles with the venue's funding, its result added to the equity the next one is sized on, and set again when the strategy restarts it. An account is **liquidated**, and then keeps nothing, when it no longer covers the maintenance margin of what it holds, half the initial margin at the market's highest leverage. Without `--interval` the walk reads hourly candles, which the venue keeps for about 207 days; `--interval 4h` reaches two years, more coarsely.

Columns: `MARKET BIAS LEVELS OUT DAYS GRIDS STOPPED LIQUIDATED REFUSED PAIRS RETURN% WORST GRID% MAX EXPO% OUT OF RANGE% FEES $ FUND $ EQUITY`: the grids set, the ones that stopped out of their range, the ones liquidated, the ones the figures refused, the round trips, the return, the worst a grid stood, the exposure, the time out of range, the fees and the funding. With `--starts`, a second table gives, for each market and count of levels, `STARTS WORST% MEDIAN% BEST% LIQUIDATED WORST GRID%`.

Written to `~/.lpa/pilot/calibration/grid-<markets>-<strategy>-<date>.json`.

### A range

```sh
lpa pilot calibrate --symbol BTC,ETH --strategy range-reversal --starts 5
```

One account a market, on the hourly candles the venue keeps, about 200 days, with seven days read before the first arming so the regime can be judged. A session is armed every day at the strategy's hour, as if the trader named that market each day, and walked to its exit.

Columns: `MARKET DAYS SESSIONS ARMINGS OFF REGIME OFF WIDTH PASSED OFF TARGET OFF SIZE TRADES WON% AVG R RETURN% WORST DD% HELD% ORDERS/DAY FEES $ FUND $ EQUITY`: how many sessions and how many armings, how many were not armed because of the regime, the range's width (or no range read), an entry the price had already passed on every side, a target too near its entry or that pays less than its fees, or the size, then the trades, the share won, what a trade made over what it risked, the return, the worst fall of the equity, the time held, the most orders a day and the fees. With `--starts`, a second table gives `STARTS WORST% MEDIAN% BEST% WORST DD% TRADES WON%`.

A range is read on hourly candles only: `--interval 4h` is refused, since six hours would be one candle and a half. Written to `~/.lpa/pilot/calibration/range-<markets>-<strategy>-<date>.json`.

### With a stop

```sh
lpa pilot calibrate --symbol BTC,ETH --strategy swing-trail --every 24 --starts 6
```

One account a market, one trade at a time, sized from its risk. The trader cannot be replayed over two years, so the entry is mechanical: its rule's when the strategy has one (the breakout at its level, the hourly RSI's excess, the pullback near the EMA 20), otherwise one every `--every` hours, 24 by default, on the side of the daily EMA 50. The stop sits 1.5 daily ATR away and the target at 2.5 times the risk; under `decider: "code"` the code's own trigger arms it, with a stop of 2 ATR and a target of 3 R. The exit is the strategy's own: the target, the trailing stop, or the part first. Neither the regime nor the events are judged, so a strategy that opens in a range only is measured in every regime.

Four-hour candles over two years, hourly ones over the two hundred days the venue keeps when the strategy reads the hourly RSI, which then refuses `--interval 4h`.

Columns: `MARKET DAYS SIGNALS TRADES WON% AVG R RETURN% WORST DD% HELD% STOP TARGET TIME OPEN FEES $ FUND $ EQUITY`: the signals the rule gave, the trades taken, the share won, the average result in R, the return, the worst fall, the time held, then how the trades ended (on their stop, on their target, on the time stop, still open), the fees and the funding. With `--starts`, the spread table of a range's.

Written to `~/.lpa/pilot/calibration/stop-<markets>-<strategy>-<date>.json`. To play the same strategy with the model deciding, use [`lpa pilot replay --strategy`](https://doc.lockerprotocol.com/hyper-agentic-trader/replay.md).

## The whole venue

A charter whose `markets` are `["*"]`, the one entry alone, considers every pair of the venue instead of BTC and ETH. `lpa pilot universe` lists the markets whose liquidity passes the bounds (volume, open interest, funding, the spread of the book, and the complete ladder's share of the market), names every one left out with its reason, and with `--measure` scores each kept market over two years with the charter's own figures. A thin market is the one way a method without a stop gets caught, so the bounds are strict by default and each one is a flag. The command is described in [lpa pilot universe](https://doc.lockerprotocol.com/hyper-agentic-trader/commands/pilot.md#lpa-pilot-universe). Once the universe is measured, a charter whose markets are `["*"]` trades it: each day a screen without a model puts the best few markets on the table, and names the others with their reason. See [The universe](https://doc.lockerprotocol.com/hyper-agentic-trader/charter.md#the-universe).

## Learn more

* [The strategies](https://doc.lockerprotocol.com/hyper-agentic-trader/strategies.md)
* [The charter](https://doc.lockerprotocol.com/hyper-agentic-trader/charter.md)
* [Replaying the past](https://doc.lockerprotocol.com/hyper-agentic-trader/replay.md)
* [lpa pilot calibrate, in the command reference](https://doc.lockerprotocol.com/hyper-agentic-trader/commands/pilot.md#lpa-pilot-calibrate)
