← flowsniper.ai

The mental model

Each of the four modes reacts to a different kind of signal. The single most useful question you can ask when picking a mode is:

That sentence is the entire product in compressed form. Everything below is the elaboration.

Trigger matrix

Each cell shows whether that exit trigger fires for that mode. ✓ means the trigger is active; dash means the mode does not react to that condition.

Trigger	Sniper	Sentinel	Designate	Shadow
STOP_LOSS Price drops below the per-position floor	✓ configurable %	—	✓ default 15%	—
TRAILING_STOP Locks in profit after position arms	✓ tiered (15/25/40)	—	✓ single tier	—
TAKE_PROFIT Sells at fixed % gain (if configured)	✓ if set	—	✓ if set	—
PARTIAL_LADDER Sells in steps as profit grows	✓ default [+15/20%, +40/30%]	—	—	—
TIME_EXIT Sells after holding too long	✓ default 120h	—	—	—
SIGNAL_EXIT Sells on signal decay	✓	—	—	—
Drift rebalance Trim outperformers to target weight	—	✓ fully-built only	—	—
Universe dropout Slow unwind when subnet fails quality screen	—	✓ rate by reason	—	—
Mirror sell Target wallet reduces position	—	—	—	✓ proportional
Mirror close Target wallet fully exits	—	—	—	✓
User exit Operator-initiated	via menu	via [R] force	via DESIGNATE_EXIT	via menu / promote

Sniper exits in detail

Sniper has the most complete exit toolkit of any mode. It runs every sweep cycle against every open Sniper position, evaluating six potential triggers in priority order.

Priority order

When multiple triggers fire on the same cycle, Sniper resolves in this order (lowest number wins):

1. STOP_LOSS — capital protection trumps everything
2. TRAILING_STOP — lock in gains once armed
3. SIGNAL_EXIT — decay-based exit
4. TAKE_PROFIT — fixed-percent target hit
5. TIME_EXIT — held too long but not bleeding
6. PARTIAL_LADDER — step-based partial profit-taking

STOP_LOSS

Computes a price floor from the entry mark and the configured stop-loss percentage. If the current mark drops to or below the floor, Sniper exits the entire position.

floor = entry_mark × (1 - stop_loss_pct / 100)
trigger condition: current_mark ≤ floor

Default stop_loss_pct is 15%. Configurable per position via custom_risk on Designate-marked positions, otherwise uses the global Sniper default.

TRAILING_STOP — tiered

Sniper's trailing stop is the most nuanced exit trigger in the bot. It tightens as the peak profit grows:

Peak gain (HWM %)	Trail tightness	Default %
≥ 40%	Tightest tier	3% from peak
≥ 25%	Tight tier	5% from peak
≥ 15%	Standard tier	7% from peak
≥ arm threshold (default 8%)	Loose tier	trailing_stop_pct default 8%

The trailing stop arms when peak gain crosses trailing_stop_arm_profit_pct (default 8%). Once armed, it stays armed — even if the position retraces back below 8%, the trail still protects you. This is intentional: a position that hit +30% and dropped to +5% should still get exited if it falls another 5% below the peak, not silently reset.

TAKE_PROFIT

Hard ceiling. If position PnL reaches or exceeds take_profit_pct, exits the full position. Default is 100%, which means most operators never see this fire because trailing stop or partial ladder closes positions before they double.

Set lower (e.g., 50% or 25%) if you want to lock in big winners regardless of trailing stop behavior.

PARTIAL_LADDER

Sells fractions of the position at predefined PnL levels. Default ladder is two steps:

[+15% gain, sell 20% of position]
[+40% gain, sell 30% of position]

After the second step fires, ~50% of the position remains. That remainder is then managed by trailing stop or other triggers. The ladder fires at most once per step (no re-firing if price oscillates back through a level).

TIME_EXIT

Sells positions held longer than max_hold_time_seconds. Default is 432000 seconds = 120 hours = 5 days. Reasoning: Sniper is built for momentum trades. A position held more than 5 days has either stopped trending or already taken profit via another trigger; holding longer adds opportunity cost without expected reward.

Time exit is the lowest-priority trigger. It only fires if no other exit has already been queued for that position on the same cycle.

SIGNAL_EXIT

Fires when a subnet's classification flips to EVADE AND its decay score crosses hard_exit_decay (default 0.68). Translation: the underlying setup that justified entry has rotted enough that holding is no longer rational.

This is the "the bot's own signals say get out" exit, distinct from price-based stops. A position can be in profit when SIGNAL_EXIT fires — the trigger is signal degradation, not pnl.

Sentinel exits in detail

Sentinel has two exit categories: drift rebalance for positions that remain in the qualified universe (trim outperformers back to target weight), and universe-dropout unwind for positions whose subnets have fallen out of the qualified universe (slow exit at a rate determined by the dropout reason). There is no price-based stop-loss, no take-profit, no signal-decay exit.

Drift rebalance — how it works

On every sweep cycle, Sentinel computes:

1. Each held position's current weight in the portfolio (% of total Sentinel value)
2. Each position's target weight from the tier's universe ranking
3. The drift = actual weight − target weight

If drift exceeds drift_threshold_pct (3% Aggressive, 5% Balanced, 10% Conservative), Sentinel may queue a partial trim. "May" because four safeguards stack:

SAFEGUARD A — Build-out gate

If fewer than 75% of target holdings are held, no rebalancing fires. The math on a partial portfolio produces hostile decisions (each held position over-represents the target). Wait until the build is mostly complete before rebalancing.

SAFEGUARD B — Cooldown

Each subnet has a per-tier cooldown after rebalance: 14 days (Aggressive), 30 days (Balanced), 60 days (Conservative). No subnet rebalances more often than this regardless of how much it drifts.

SAFEGUARD C — Position floor

Never trim below max(0.01τ, target_value). And refuse to trim positions that start below 1.5× their target value — drift noise alone isn't worth the trading fees. This prevents the "fee-and-slippage death spiral" on small positions.

SAFEGUARD D — Daily trim cap

At most 1 trim per netuid per UTC day. Prevents cascade failures where one cycle trims, the next cycle sees residual over-target weight, trims again, repeat.

What drift rebalance is NOT

• Not a profit-taker. It trims winners as a side effect of portfolio mathematics, not because they made money.
• Not a stop-loss. A position that loses value relative to others gets more weight added to it, not trimmed.
• Not a complete exit. Trims always reduce position size; never zero it out.
• Not fast. Cooldowns mean an outperforming position gets trimmed at most once every 14-60 days.

If you need any of those behaviors on a Sentinel position, the answer is to promote the position to Designate via Designate → Promote Sentinel. The on-chain alpha doesn't move; only the DB row migrates. Designate's risk profile (stop-loss, trailing stop, take-profit) then applies.

Universe-dropout unwind — how it works

When a held position's subnet falls out of the qualified universe (its quality screen starts failing), Sentinel automatically unwinds the position over time. The unwind rate depends on why it dropped out, modeled on industry practice.

Reason classification

On detection, Sentinel re-runs the universe filter checks against the held position's signal and records the FIRST failing reason. This reason is then locked in for the entire unwind window — the rate doesn't shift mid-unwind even if conditions evolve.

Reason	Speed	Aggressive	Balanced	Conservative
pool	Fast	7 days	15 days	30 days
dereg_risk	Fast	7 days	15 days	30 days
mcap_min	Slow	14 days	30 days	60 days
mcap_ratio	Slow	14 days	30 days	60 days
pool_cv	Slow	14 days	30 days	60 days
flow_14d / flow_7d	Slow	14 days	30 days	60 days
age	Slow	14 days	30 days	60 days
rank	Slow	14 days	30 days	60 days

Industry rationale

Fast unwind — liquidity-driven dropouts (pool, dereg_risk). Industry index funds typically remove on liquidity grounds within 5–10 trading days. Thin liquidity means market impact gets worse the longer you wait.

Slow unwind — quality/valuation dropouts (mcap, flow, cv, age, rank). Active funds typically use 1–3 month unwind windows on outflow-driven removals. The position may recover; rushing the exit costs more than waiting.

How a single dropout plays out

Day 0: SN42 falls out (pool dropped below threshold). Sentinel logs the dropout, classifies pool reason, locks in 15-day unwind window (Balanced tier), records original position value, fires Discord webhook. No trim this cycle — gives a stable baseline.

Day 1: Position trimmed toward 93% of original value (1 day of 15 elapsed = 6.7% trim).

Day 5: Position trimmed toward 67% of original.

Day 10: Position trimmed toward 33% of original.

Day 15: Final exit fires. Discord webhook confirms full unwind. Subnet enters 7-day post-exit blacklist.

Or: any time before day 15, if SN42's pool recovers above threshold AND the subnet returns to the target list, unwind halts. Discord webhook fires recovery message. Normal Sentinel mechanics resume.

Mid-unwind safeguards

• Daily trim cap. Max 1 trim per netuid per UTC day (prevents cascade trims).
• Min trim size. Skips trims below 0.005τ (avoids fee-dominated micro-trades).
• Linear pacing. Trim target is a linear function of time elapsed in the unwind window. If price decay has already brought the position below the target curve, no trim fires that cycle.

Operator overrides

• Promote to Designate — Designate → Promote Sentinel halts the unwind and applies Designate's risk profile. The picker shows a Status column that flags which positions are currently being unwound, so you can prioritize.
• [E] Exit non-target positions in Sentinel Portfolio Preview — one-shot bulk exit of all non-target Sentinel positions immediately.
• Force rebalance in the menu — bypasses cooldowns and gates for one cycle (affects all of Sentinel, not just dropouts).

Designate exits in detail

Designate has a simpler risk model than Sniper: three triggers, no partial ladder, no time exit, no signal decay. The philosophy is "you marked the target — the bot manages it with industry-standard rules until you tell it otherwise."

STOP_LOSS

Same mechanic as Sniper but with Designate-specific defaults. Configurable per position via custom risk settings when you mark the target.

floor = entry_mark × (1 - stop_loss_pct / 100)
default stop_loss_pct: 15%

TRAILING_STOP

Single tier, not tiered like Sniper's. Arms when peak gain crosses trail_arm_pct (default 8%), then sells if price drops trailing_stop_pct below the peak.

TAKE_PROFIT

If take_profit_pct is set non-zero, fires a full exit when PnL reaches that level. Default is 0 (disabled). Set explicitly when marking a target if you want a hard upside exit.

DESIGNATE_EXIT

Operator-initiated. Mark a Designate position for immediate exit through the menu. Routes through the same execution path as automated exits, with the reason logged as DESIGNATE_EXIT user-requested.

Shadow exits in detail

Shadow has no independent risk management. Its exit triggers are entirely target-driven.

Mirror sell (PARTIAL_EXIT)

When a watched wallet reduces its position in a subnet you also hold via Shadow, Shadow mirrors the reduction proportionally:

reduction_pct = (target_alpha_before - target_alpha_now) / target_alpha_before
your_alpha_to_sell = your_position × reduction_pct

Filtered by shadow_trade_delta_min (default 2%) — small target reductions below this threshold are ignored to prevent fee-dominated micro-trades.

Mirror close (full EXIT)

When a watched wallet fully exits a subnet, Shadow closes your corresponding position completely. Reason logged as SHADOW target fully exited — mirror close.

What Shadow doesn't do

• No stop-loss on the position itself. If your target wallet holds through a 50% drawdown, you hold through a 50% drawdown.
• No take-profit. If your target rides a position to 10x, you ride it to 10x — until they sell.
• No drift rebalancing. Each Shadow position is judged purely by what its source wallet does.

If you want stop-loss-style protection on a Shadow position, the workflow is to promote the Shadow position to Designate. The mirror relationship is preserved — if the target sells, ShadowTracker still triggers your exit. But Designate's risk rules now also apply, whichever fires first wins.

Worked examples

Concrete timelines showing how each mode actually behaves. All examples assume default parameters and a 1τ wallet.

Example 1 — Sniper rides a winner through partial ladder + trailing stop

Example 2 — Sentinel holds through drawdown, trims a winner months later

Example 3 — Designate stops out on a manual pick that breaks down

Example 4 — Shadow mirrors a target through entry, partial exit, full close

Parameter reference

All defaults pulled from source. Operators can override any of these per their own configuration via the appropriate menu (Sniper menu for Sniper params, etc.).

Sniper

Parameter	Default	Range	What it does
stop_loss_pct	15.0	1.0 – 50.0	Hard floor for STOP_LOSS. Position exits if mark drops below entry × (1 - pct/100).
trailing_stop_pct	8.0	1.0+	Loose-tier trail. Used when peak gain is between arm threshold and 15%.
trailing_stop_arm_profit_pct	8.0	1.0 – 50.0	Peak-gain threshold above which trailing stop arms.
tiered_trail_15_pct	7.0	1.0+	Trail tightness when peak gain ≥ 15%.
tiered_trail_25_pct	5.0	1.0+	Trail tightness when peak gain ≥ 25%.
tiered_trail_40_pct	3.0	1.0+	Trail tightness when peak gain ≥ 40%.
take_profit_pct	100.0	0+	Hard ceiling. Set 0 to disable. Set lower to lock big winners regardless of trail.
max_hold_time_seconds	432000 (5 days)	any positive int	TIME_EXIT threshold. Set 0 to disable.
min_hold_time_seconds	600 (10 min)	any positive int	Minimum hold before any exit can fire. Prevents whipsaw on volatile early-cycle data.
hard_exit_decay	0.68	0.0 – 1.0	SIGNAL_EXIT threshold. Position exits if classification = EVADE AND decay_score ≥ this.
partial_ladder	[[15, 0.20], [40, 0.30]]	list of [pnl%, frac]	Partial-exit ladder. Each step: at this pnl%, sell this fraction of remaining position.

Sentinel

Sentinel has no exit-trigger parameters per se — its exit behavior is governed by the tier config. The relevant knobs:

Parameter	Conservative	Balanced	Aggressive
drift_threshold_pct	10.0	5.0	3.0
rebalance_min_interval_days	60	30	14
weekly_deployment_pct	10.0	20.0	35.0
max_position_pct	25.0	20.0	18.0
holdings_default	5	8	10

The build-out gate (75%), daily trim cap (1 per netuid), and position proximity floor (1.5× target) are not user-configurable in v1.2.

v1.2 override notes: Eight of the Sentinel tier values are now operator-tunable per tier via Sentinel → [9] Risk Profile → [F] Tier filter overrides: pool_min_tao, mcap_min_tao, mcap_pool_ratio_max, flow_14d_min_pct, flow_7d_min_pct, age_min_days, pool_cv_max_pct, dereg_rank_min. The max_position_pct is also overridable globally via [C] Adjust position cap (5–50% range, with a math floor warning when cap × holdings < 100%). Each tier maintains independent overrides — switching tiers preserves the customizations for the tier you switched away from.

Designate

Designate uses Sniper's parameter names but with its own defaults. These are the defaults applied when a position has no custom_risk dict:

Parameter	Default	Notes
stop_loss_pct	15.0	Same as Sniper default.
trailing_stop_pct	8.0	Single-tier trail (Designate doesn't use Sniper's tiered system).
trailing_stop_arm_profit_pct	8.0	Stored as trail_arm_pct in Designate's effective risk.
take_profit_pct	0.0 (disabled)	Set explicitly per position when marking the target if you want this active.

When marking a Designate target, you can override any of these via the custom_risk dict, which takes precedence over the global defaults.

Shadow

Parameter	Default	What it does
shadow_trade_delta_min	2.0	Minimum % change in target's position to trigger a mirror trade. Prevents micro-trade noise.
shadow_budget_pct	20.0	% of wallet allocated to Shadow when in percentage budget mode.
shadow_budget_tao	0.25	τ amount when in absolute budget mode.

"I want X" decision tables

Profit-taking

What you want	Mode	How it works
Take aggressive partial profits at fixed %s	Sniper	PARTIAL_LADDER fires at +15% (sell 20%), +40% (sell 30%), then trailing stop on the rest
Take all profits at one fixed %	Sniper or Designate	TAKE_PROFIT triggers full exit at threshold (must be configured)
Lock gains via trailing stop after a position runs	Sniper (tiered) or Designate (single tier)	TRAILING_STOP arms after profit exceeds arm threshold, then sells if price drops the trail % below peak
Trim winners to feed laggards (no full exit)	Sentinel	Drift rebalance trims outperformers, redeploys into underweight positions. Only after portfolio is 75%+ built. 14-60 day cooldown depending on tier.
Hold the position no matter how high it runs	Sentinel mostly does this	Sentinel won't sell on profit alone; only on drift rebalance, which is portfolio-relative not absolute. Caveat: massively overweight positions get trimmed.
Mirror your favorite trader's profit-taking	Shadow	If they take profit, you take profit, proportionally

Loss / risk management

What you want	Mode	How it works
Hard stop-loss at fixed % below entry	Sniper or Designate	STOP_LOSS at the configured % (default 15% both)
Sell on technical breakdown	Sniper	SIGNAL_EXIT fires when classification flips to EVADE and decay crosses 0.68
Time-out a position that's just sitting	Sniper	TIME_EXIT fires after 5 days (default)
Manual stop-loss whenever I decide	Designate	Use DESIGNATE_EXIT, or promote a Shadow/Sentinel position into Designate first
No automatic exit on losses	Sentinel or Shadow	Both will hold through drawdowns. Sentinel screens at entry; Shadow inherits risk from your target wallet.
Stop-loss on a Sentinel position	Promote to Designate	Use the menu to migrate. On-chain alpha doesn't move; only the DB row migrates between mode tables.

Edge cases

Situation	What happens
Subnet I hold drops 50%	Sniper stops out at floor. Designate stops out at floor. Sentinel holds. Shadow holds unless target sells.
Subnet I hold pumps 200%	Sniper: partial exits fire, trailing stop closes runner. Designate: take-profit fires if configured, else trailing stop. Sentinel: trims via drift rebalance over weeks if it gets too overweight. Shadow: holds unless target sells.
I want OUT of everything right now	Sniper: stop the daemon, manually unstake. Sentinel: menu [R] force rebalance + manual unstake. Designate: bulk DESIGNATE_EXIT via menu. Shadow: manual unstake or stop watching the source wallet.
Target wallet I'm shadowing rugs	Shadow's mirror-close fires when their alpha drops, but only when they actually sell. If they hold a dead bag, you hold a dead bag. Promote to Designate to apply your own stop-loss.
Subnet gets deregistered	Sentinel's universe filter catches it pre-entry. Once held, no automatic exit; the alpha may go to zero on chain. Manual exit only.
Bot crashes mid-trade	Open positions are safe — they sit on chain as staked alpha. When the daemon restarts, recon sweep reconciles DB to chain. No exits are missed permanently; the next sweep evaluates them all from current state.
Two modes hold the same subnet	Each mode tracks its position independently in its own state dict. Exits from one don't affect the other. Allowed and intentional — Sentinel can hold SN8 while Sniper takes a short-term scalp on the same subnet.

Strategy combinations

Setup	Why it works
Sniper + Sentinel	The classic. Sniper plays short-term momentum; Sentinel builds long-term portfolio. Sniper takes the fast money, Sentinel survives the volatility.
Sniper + Shadow	Active scans for momentum + information edge from wallets you trust. Sniper's rigor + Shadow's signal intelligence.
Sentinel + Shadow	Pure long-bias setup. Sentinel does broad risk-managed accumulation; Shadow concentrates capital around specific wallets you trust.
Sentinel + Designate	Sentinel handles broad portfolio; Designate is your conviction picks with proper stop-loss. Combines automated discipline with discretionary edge.
All four	Maximum strategy diversification. Some operators run this. Requires more wallet capital to give each mode a meaningful budget allocation.

What's deliberately missing

Some behaviors users sometimes expect but the bot intentionally does not implement. These are design choices, not feature gaps.