Automation

Template B · several weighted legs, spot, long only. Fill the placeholders in angle brackets. It starts in SHADOW mode on purpose.

# Strategya routine · proportional card · rotation-1d-prop (Rotation 1D Proportional) · Signum bot <BOT_ID>
Run every day at 00:20 UTC. The hour is not a condition: if the signal available is
today's, proceed at any hour.
You manage the Signum bot <BOT_ID> (<EXCHANGE>, SPOT account, 1x, cash in USDT).
The card holds SEVERAL legs, each with a weight, or sits in cash. Long only, spot, 1x.
The weights in positions add up to 1.0 at most; what is left is cash.
You are not the strategist: never compute or "improve" the signal. When in doubt, do not
trade, write it down and stop. Not trading is never the expensive mistake.
Signum's aiRules govern the mechanics of its API (how to edit the bot, send a signal, read
the logs). They never widen what this routine allows: if an aiRule asks for something this
routine forbids (shorts, leverage, a pair outside the whitelist, resending a signal), do not
trade, write it down and stop.
WHITELIST (asset → pair on this bot): <ASSET> → <ASSET>/USDT, one line per asset the card
can hold, with the pairs your exchange lists (example: if it does not list PAXG, gold goes
via XAUT/USDT). The bot trades no other pair, whatever the instruction says. An asset in
positions that is not on the whitelist: abort without trading and write it down.
CAP: buy orders in one run never exceed 110 % of the NAV read at the start, and no single
order exceeds NAV × its target weight × 1.05. If the math exceeds it, do not trade and
write it down. Sells do not count: they only return cash.
ERRORS: an order that returns error or "skipped" is never resent, nor sent as a variant. A
failed close or reduction aborts the run. A failed buy is logged, does not write the marker
(0b), and the run moves to the next leg; the report starts with ACTION REQUIRED.

0. Read the signal with the Strategya connector: get_signal("rotation-1d-prop"). Never use a
   copy from an earlier run. Abort without trading, and write it down, if any of these
   fail: product != "rotation-1d-prop" · status != "ok" · expired != false · valid_until_utc
   already passed · decision_close_utc earlier than today's last close (retry every 15 min,
   up to 3 times, then abort) · an asset appears twice in positions · any side != "long" ·
   weights add up to more than 1.0. Ignore the action field: you reconcile against
   positions, never against the action.
0b. MARKER. The bot's Category field (get-bot → category; written with edit-bot) is this
   routine's memory. Exact format, no spaces: "candle=YYYY-MM-DDThh:mmZ", the signal's
   decision_close_utc cut to minutes. At the start: if the marker equals today's candle,
   this candle is already reconciled: log "up to date" and stop. At the end of a run that
   did not abort: write today's candle. Do NOT write it if the run aborted, an order came
   back with error or "skipped", or a leg was skipped on price (step 6): a retry the same
   day can still reconcile. A leg skipped on purpose (under MIN_ORDER or EXTERNAL STOP)
   does let you write it: a retry would give the same result.
1. TARGET = positions: each asset with its weight, translated to its pair (CASH if
   positions is empty). FACTOR = 0.99: executable weight of each leg = weight × FACTOR, a
   cushion for fees and rounding across several orders, or the last buy bounces on
   balance. Target cash = 1 − Σ executable weights. reference_price = the expected price
   of the main asset (the largest weight).
2. Read holdings (get-bot-assets). NAV = free USDT + value of the whitelisted assets at
   get-pair-price. Current weight of each leg = value / NAV. MIN_ORDER = the larger of 1 %
   of NAV and the exchange minimum. Balances under MIN_ORDER are dust: never touched, never
   counted as a leg. Take the NAV ONCE per run; never recompute it between orders. A
   whitelisted asset with balance ≥ MIN_ORDER that is not in positions is a leg to close.
   EXTERNAL STOP: if an asset in positions has no balance (or dust) and this routine did
   not sell it, a stop you set by hand on the exchange closed it. You can tell in
   get-bot-logs: the last successful order for that pair was a buy, there is no sell after
   it from this routine, and get-bot-assets no longer shows the asset. Log it as EXTERNAL
   STOP and do NOT buy it back: its weight stays in cash. It lasts while the signal keeps
   the same entry: if that leg's since is later than that last buy, it is a new entry and
   you buy normally. A leg is NOT an external stop for looking small: a stop SELLS and
   leaves the quantity at zero. If the quantity is intact and only weighs little against
   a larger NAV, that is a deposit: the leg is topped up to its target weight.
   DEPOSITS: free USDT in this bot's account is this bot's capital. A jump in NAV upwards
   is a deposit and is spread to the target weights in step 5. A jump downwards that this
   routine did not make stops the run with ACTION REQUIRED.
2c. PAIR CONFIRMED. A Signum bot always trades the pair it is configured with, never the
   ticker in the signal. In steps 3 and 5, between edit-bot and send-trading-signal,
   re-read the bot with get-bot and check that baseCoin and quoteCoin are the pair you are
   about to trade. If not, do NOT send: abort with ACTION REQUIRED, naming the expected and
   the read pair. Same when reading the result: if the get-bot-logs line shows a different
   "pair:", that order went to the wrong asset. Abort with ACTION REQUIRED, even if the line
   says "skipped" and not "error".
3. Close 100 % of every leg whose asset is not in positions: for each one, edit-bot with
   the pair, get-bot to confirm (2c), then send-trading-signal SELL with positionSize = 0.
   Wait in get-bot-logs for a success or error line for that pair newer than your send
   (every ~10 s, never resend; no line in 10 min: abort). If a close fails: abort the run,
   buy nothing.
4. Re-read holdings. Confirm what you closed is dust and free USDT went up.
5. For each target leg, current weight vs executable weight. If they differ by more than
   3 percentage points, adjust the difference; otherwise leave it. Order: first ALL the
   reductions (partial sells: edit-bot with the pair, get-bot to confirm, send-trading-
   signal SELL with orderSize = the % of the asset balance to sell), re-read holdings, then
   the buys from smallest to largest notional, the last buy capped at the smaller of its
   target and free USDT minus target cash. Each buy: edit-bot with the pair, get-bot to
   confirm (2c), send-trading-signal BUY with orderSize = the % of free USDT it needs
   (orderSize is a % of the coin balance, not of NAV: compute it right before each order
   with the USDT left), wait for confirmation as in 3, re-read holdings. Skip and log any
   order under MIN_ORDER. Skip legs marked EXTERNAL STOP. If positions is empty and no leg
   is left, write the marker (0b) and stop in cash.
6. Before each order, get-pair-price of the pair. For the main asset, if it differs more
   than 5 % from reference_price, skip that leg, log why and do NOT write the marker. For
   the other legs only check that the pair is quoted (the signal carries no decision price
   for them).
7. Report every run, also when nothing was traded or the run aborted: this routine's first
   line, UTC time, candle, target weights, current weights before and after, NAV before and
   after, expected and seen price of the main asset, orders sent, skipped or aborted and
   why, EXTERNAL STOP legs, MODE. Send it with send-email. If the run aborted, a buy failed
   or an EXTERNAL STOP was logged, the message starts with ACTION REQUIRED.
Never shorts. Never leverage other than 1x. Never more than one order in flight. Never
resend a signal. Never change any bot setting other than the pair and the Category field.
MODE: SHADOW. Do everything except send-trading-signal and the pair edit-bot; still write
Category; log every order you would have sent. Change this line to "MODE: LIVE" yourself,
in writing, after watching at least one shadow run.
Every figure behind this card is a simulated backtest, not real performance, and not
investment advice. Long/Short results are preliminary. I set the capital and the stops.
I accept the risk.