Settings reference / Advanced / research
Advanced / research settings
Watch-only logs, wall-flow analysis and calibration, market-data recording and the disk budget. 73 settings on this tab.
Generated from the code. Defaults, ranges, choices, units and the live confirmation rule on this page are read from the settings schema each time the docs are generated, so they always match the bot.
Summary
All settings
Wall-flow log: compact
wallflow_compact
- Default
- Off
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Compact wall-flow log: only the approach, the sample ~60 s before the first touch, the touch, the wall's end and the 30-minute outcome rows (no confirmation rows, no 5 s samples), each with the three powers (impact_power, wall_power, reflection_power). A few MB per day. On the Trading instance the wall-flow log runs only in this compact form. Analysis only: never changes a trading decision.
What it does
A compact form of the wall-flow analysis log: only the approach (price within 0.2% of a wall), the sample about 60 s before the first touch, the touch, the wall's end and the 30-minute outcome rows are written; confirmation rows and the 5-second samples are left out. Every row carries the three powers: impact_power (2-minute aggressive flow toward the wall divided by its support capacity), wall_power (support capacity = cushion + the wall's remaining notional, USDT) and reflection_power (the 5-minute reflection margin as a share of capacity: 1 = no impact, below 0 = eaten within 5 minutes). On the Trading instance the wall-flow log only runs in this compact form. Analysis only: it never changes a trading decision.
Example
A coin with 20 confirmed walls a day of which 2 are approached and 1 touched writes about 5 rows instead of several hundred; across 10 coins that is a few MB per day.
When to change it
Turn it on (with wallflow_log) on the Trading instance, or on any instance where the full log is too large.
Higher, lower, on or off
On: far fewer rows, same touch and outcome detail. Off: the full log with confirmations and 5 s samples.
Risk notes
None for trading. It uses a little CPU for each wall near the price.
Wall-flow analysis log
wallflow_log
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Analysis only: every 5 s while price is within 0.5% of a tracked wall, and at its confirmation, first touch and end, save the flow measurements (impact, support capacity, reflection margins, bounce estimates, required move) plus each touched wall's 30-minute outcome to data/wallflow/ (one gzip CSV per coin per session). Written by a background thread; never changes a trading decision.
What it does
Analysis only. While price is within 0.5% of a tracked wall, it saves flow measurements every 5 seconds (and at confirmation, first touch and wall end), plus each touched wall's 30-minute outcome, to data/wallflow/. It never changes a trading decision.
Example
On, with BTCUSDT near a bid wall for 2 minutes: about 24 rows (one per 5 seconds) go into that coin's gzip CSV for the session, then one outcome row 30 minutes after the touch.
When to change it
Keep it on on the Research instance if you want data for the flow study. Turn it off to save disk space. The Trading instance never writes it, and the simulator does not either.
Higher, lower, on or off
On: rows are written by a background thread to one gzip CSV per coin per session. Off: nothing is written.
Risk notes
Low: only changes what is logged. It uses some disk space and a little CPU.
Flow filter at entry
flow_filter
- Default
- (empty)
- Range
- Free text
- Type
- text
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Empty = off. Comma-separated conditions on the anchor wall's flow measurements at entry, all of which must hold, e.g. 'reflection_margin_5m>0, est_bounce_15m>=required_move_pct'. Walls that fail (or whose measurement is unknown) are skipped. Entry only: open trades are never affected.
What it does
An extra entry check on the wall's order-flow measurements. You write conditions separated by commas, and every one must hold, or the wall is skipped. A measurement that is unknown counts as a fail. Empty means the filter is off. It only affects new entries, never open trades.
Example
reflection_margin_5m>0, est_bounce_15m>=required_move_pct skips a wall unless it absorbed selling with margin to spare over 5 minutes and its estimated 15-minute bounce covers the move needed to reach target_net_roe. The log then shows e.g. "Skipped ... flow filter: est_bounce_15m unknown".
When to change it
When a flow study suggests that some measurements separate walls that hold from walls that break, and you want to trade only the stronger-looking ones.
Higher, lower, on or off
Empty: no flow check. With conditions: fewer entries, only walls that pass every condition. Allowed names include impact_30s, cushion, reflection_margin_1m/5m/15m, est_bounce_5m/15m, required_move_pct, room_ok, refill_ratio, imbalance_05 and others; operators are >, >=, <, <=, =.
Risk notes
Low for safety: it can only block entries. A typo is rejected when you save ("Flow filter: not understood"). Too strict a filter means the bot rarely or never trades.
Bounce estimate calibration (5 min)
bounce_calibration_5m
- Default
- 1
- Range
- 0 to 100
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Analysis only: est_bounce_5m_cal = est_bounce_5m x this (the median actual / estimated move measured by the flow study).
What it does
Analysis only. A multiplier for the 5-minute bounce estimate in the wall-flow log: est_bounce_5m_cal = est_bounce_5m times this. It lets you correct the estimate with the actual/estimated ratio measured by the flow study.
Example
The estimate says a 0.40% bounce in 5 minutes and the flow study found real moves were about half the estimate. Set 0.5 and est_bounce_5m_cal is logged as 0.20%.
When to change it
After a flow study tells you the median actual/estimated ratio. Leave it at 1 until then.
Higher, lower, on or off
Higher: the calibrated 5-minute bounce is larger. Lower: smaller. 0 makes the calibrated value zero.
Risk notes
Low: only changes what is logged.
Bounce estimate calibration (15 min)
bounce_calibration_15m
- Default
- 1
- Range
- 0 to 100
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Analysis only: est_bounce_15m_cal and room_ok_cal use est_bounce_15m x this.
What it does
Analysis only. A multiplier for the 15-minute bounce estimate in the wall-flow log. est_bounce_15m_cal and room_ok_cal use est_bounce_15m times this.
Example
Estimated 15-minute bounce 0.80%, calibration 0.75: est_bounce_15m_cal = 0.60%. With a required move of 0.62%, room_ok_cal is false even though the raw estimate would pass.
When to change it
After a flow study measures how the 15-minute estimate compares with real moves. Leave it at 1 until then.
Higher, lower, on or off
Higher: larger calibrated bounce, so room_ok_cal is true more often. Lower: smaller calibrated bounce, room_ok_cal true less often.
Risk notes
Low: only changes what is logged.
Coins to record
record_watch_count
- Default
- 15
- Range
- 0 to 40
- Type
- Whole number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: With recording on and the scanner on, record the order book and trades of this many coins: the traded coins plus the next best by the scanner (hygiene exclusions apply, the allowlist does not, since it only limits trading). Recorded-only coins are never traded. 0 = only the traded coins. Binance only (MEXC depth is polled).
What it does
With record_market_data on and the scanner on, how many coins to record: the traded coins plus the next best from the scanner. Hygiene exclusions apply, the allowlist does not. Recorded-only coins are never traded. Movers come on top of this number.
Example
Count 15 with 2 traded coins: the bot also records the 13 next best coins, so 15 in total. Set 0 and only the 2 traded coins are recorded.
When to change it
Raise it to collect backtest data on more coins; lower it if disk space or memory is tight.
Higher, lower, on or off
Higher: more coins recorded, more disk use (busy coins take several hundred MB per day). Lower: fewer coins recorded. 0: only the traded coins.
Risk notes
Low for trading (recorded-only coins never trade). Watch disk space and memory on a small PC. Binance feed only (it does nothing when a MEXC feed is running).
Record market data
record_market_data
- Default
- Off
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Save the live order-book updates and trades of watched coins to backtest_data/live/ in the format the backtester reads (one gzip file per coin per UTC day). Busy coins take several hundred MB per day; disk usage is shown at the top of the dashboard. Binance feed only.
What it does
Saves the live order-book updates and trades of watched coins to backtest_data/live/, in the format the backtester reads (one gzip file per coin per UTC day). Disk use is shown at the top of the dashboard.
Example
On, recording 15 coins: each UTC day adds 15 files to backtest_data/live/. If 3 of them are busy coins at about 300 MB each, those alone add about 900 MB per day.
When to change it
Turn it on on the Research instance to build backtest data. The Trading instance never records, even with this on.
Higher, lower, on or off
On: recording runs for the watched coins (see record_watch_count). Off: nothing is recorded.
Risk notes
Low for trading. It can fill the disk over time; check the disk-use figure. Binance feed only; the simulator cannot record.
Record spot books (study)
spot_record
- Default
- Off
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Research only (combined-book study): keep a compact record of the Binance SPOT order book of the watched / recorded coins that have a spot market, plus the extra coins below: spot vs futures price (basis), spot liquidity per price bucket near the market, and big spot levels appearing / vanishing, in data/spotrec/. Never used for trading; uses spot request weight only. The Trading instance never records.
What it does
Research only, for the combined-book study: keeps a compact record of the Binance SPOT order book of the watched / recorded coins that have a spot market, plus spot_record_extra_coins, in data/spotrec/: spot vs futures price (basis), spot liquidity per price bucket near the market, and big spot levels appearing, changing and vanishing.
Example
On with 12 watched coins that have spot plus 10 extra coins: about 22 coins x 2 MB = about 44 MB per day in data/spotrec/.
When to change it
Turn it on on the Research instance (8787) for the combined-book study. The Trading instance never records it, even with this on.
Higher, lower, on or off
On: the spot book is recorded for the chosen coins. Off: nothing is recorded and no spot connection is opened.
Risk notes
None for trading (it never feeds a trading decision). It uses spot request weight only (a 5000-level snapshot weighs 250, one every 3 s at most), not the futures limit. Disk is capped by spot_record_budget_gb; CPU by spot_record_cpu_guard_pct.
Spot recording: extra coins
spot_record_extra_coins
- Default
- (empty)
- Range
- Free text
- Type
- text
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Futures symbols recorded on top of the watched / recorded coins (comma separated; coins without a Binance spot market are skipped). They come first when the coin limit is reached.
What it does
Futures symbols recorded on top of the watched / recorded coins, comma separated. Coins without a Binance spot market are skipped; 1000PEPEUSDT is mapped to the PEPEUSDT spot market.
Example
HYPEUSDT,ZECUSDT,ENAUSDT: these 3 coins are recorded first, then up to 22 watched coins (limit 25).
When to change it
Add the coins the Trading instance trades most, so its fills can be compared with the spot book.
Higher, lower, on or off
More coins: more disk (about 2 MB per coin per day) and more spot messages (1 per coin per second).
Risk notes
Low: only disk and a little CPU.
Spot recording: at most
spot_record_max_coins
- Default
- 25
- Range
- 1 to 60
- Type
- Whole number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: At most this many coins are recorded (extra coins first, then the watched / recorded ones).
What it does
At most this many coins are recorded: the extra coins first, then the watched / recorded coins in the engine's order.
Example
25 with 10 extra coins and 30 watched coins with spot: the 10 extra coins and the first 15 watched ones.
When to change it
Lower it if CPU or disk is tight.
Higher, lower, on or off
Higher: more coins, more disk and messages. Lower: fewer coins.
Risk notes
Low.
Spot recording: band
spot_record_band_pct
- Default
- 2 %
- Range
- 0.5 to 5 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: The spot book is recorded within this distance of the spot mid.
What it does
How far from the spot mid the spot book is recorded.
Example
2%: with the spot mid at 100, levels from 98 to 102 are summed into buckets.
When to change it
Keep it at least as wide as the farthest wall distance studied (wall_max_dist_pct).
Higher, lower, on or off
Wider: more buckets per row (more disk). Narrower: less.
Risk notes
Low.
Spot recording: bucket width
spot_record_bucket_bps
- Default
- 5 bps
- Range
- 1 to 50 bps
- Type
- Number
- Unit
- bps
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Width of the price buckets the spot liquidity is summed into.
What it does
Width of the price buckets the spot liquidity is summed into, measured from the spot mid. The same bucket sits at futures mid x (1 + k x width) on the futures side (basis-corrected).
Example
5 bps with the mid at 100: bucket 2 holds every ask from 100.10 to 100.15.
When to change it
Match the wall zone width (wall_bucket_bps) so spot and futures buckets line up.
Higher, lower, on or off
Smaller: finer grid, more disk. Larger: coarser, less disk.
Risk notes
Low.
Spot recording: book rows every
spot_record_grid_s
- Default
- 30 s
- Range
- 5 to 600 s
- Type
- Whole number
- Unit
- s
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: How often the bucketed spot book is written per coin.
What it does
How often the bucketed spot book is written per coin.
Example
30 s: 2,880 rows per coin per day.
When to change it
Shorter if the study needs finer timing; big levels are also logged as events with exact times.
Higher, lower, on or off
Shorter: more disk. Longer: less.
Risk notes
Low.
Spot recording: basis rows every
spot_record_basis_s
- Default
- 10 s
- Range
- 1 to 600 s
- Type
- Whole number
- Unit
- s
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: How often spot mid, futures mid and the basis are written per coin.
What it does
How often spot mid, futures mid and the basis (futures mid / spot mid - 1, in bps) are written per coin.
Example
10 s: 8,640 rows per coin per day, about 0.3 MB gzipped.
When to change it
Keep it short enough to follow basis jumps.
Higher, lower, on or off
Shorter: more disk. Longer: less.
Risk notes
Low.
Spot recording: big level from
spot_record_event_usd
- Default
- 50000 USDT
- Range
- 1000 to 1e+08 USDT
- Type
- Number
- Unit
- USDT
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Spot price levels of at least this value inside the band are logged when they appear, change by 30% or more, or vanish.
What it does
Spot price levels of at least this value inside the band are logged when first seen, when they appear, when they change by 30% or more and when they vanish (at most 120 rows per coin per minute).
Example
50,000 USDT: a spot bid of 80,000 USDT at 99.5 appears -> one 'appear' row; it drops to 20,000 -> one 'vanish' row.
When to change it
Lower it to catch smaller spot walls (more rows); raise it on very liquid coins.
Higher, lower, on or off
Lower: more event rows. Higher: fewer.
Risk notes
Low.
Spot recording: CPU guard
spot_record_cpu_guard_pct
- Default
- 40 %
- Range
- 0 to 400 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Spot recording stops by itself (trading is not affected) if this instance's process uses more than this share of one CPU core for 10 minutes in a row. 0 = no guard.
What it does
Spot recording stops by itself if this instance's process uses more than this share of one CPU core for 10 minutes in a row (measured once a minute). Trading is not affected.
Example
40%: the process normally uses about 18%; if it stays above 40% for 10 minutes, spot recording stops and you are alerted.
When to change it
Keep it on: the server is a client's production machine.
Higher, lower, on or off
Lower: stops sooner. 0: no guard.
Risk notes
Low; with 0 a runaway could slow the server.
Spot recording: disk budget
spot_record_budget_gb
- Default
- 1 GB
- Range
- 0.05 to 50 GB
- Type
- Number
- Unit
- GB
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: data/spotrec is kept under this size: the oldest UTC days are deleted first (never today's). Also counted in the total disk budget.
What it does
data/spotrec is kept under this size: the oldest UTC days are deleted first, never today's files. It also counts in the total disk budget.
Example
1 GB at about 45 MB per day: roughly 3 weeks are kept before the oldest day is deleted.
When to change it
Set it to the study length x the measured daily size.
Higher, lower, on or off
Higher: more days kept. Lower: older days deleted sooner.
Risk notes
Low.
Record when the wall-health rules would trigger
wall_health_watch
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Analysis only: when exit_on_weak_refill or wall_exhaustion_protect is off, record once per trade in the live-learning log when it would have triggered, with the price at that moment (a same-signal counterfactual). Never changes a trading decision.
What it does
Analysis only. When exit_on_weak_refill or wall_exhaustion_protect is off, the live engine records once per trade in the live-learning log ("wall_health") the moment the rule would have triggered, with the price, the seconds since the fill, the volume traded at the wall, its refill ratio and remaining share. Comparing that price with how the trade actually ended gives a same-signal counterfactual. It never changes a trading decision.
Example
A live trade gets a record "exit_on_weak_refill would close at market at 1.0612, 8 s after the fill, refill 0.12"; it later stopped out at 1.0590, so the rule would have saved the difference minus the taker fee.
When to change it
Keep it on while evaluating the rules.
Higher, lower, on or off
On: a few extra live-learning records. Off: none.
Risk notes
None.
Correlation refresh
corr_refresh_min
- Default
- 60 min
- Range
- 5 to 1440 min
- Type
- Number
- Unit
- min
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: How often the 15m closes for the correlation guard are refreshed.
What it does
The correlation guard (max_same_direction_correlated) uses 15-minute closes of watched coins and coins with trades. After each scan, a coin's closes are downloaded again only if the stored copy is older than this.
Example
Refresh 60 min: ETH closes were fetched at 09:00. A scan at 09:40 reuses them; a scan at 10:05 downloads fresh ones.
When to change it
Lower it for more up-to-date correlation if you scan often. Raise it to save requests. It has no effect when the guard is off (max_same_direction_correlated 0).
Higher, lower, on or off
Higher: fewer candle requests, correlation can lag behind. Lower: fresher correlation, more requests. Refreshes still only happen after a scan, so it can never be more frequent than your scans.
Risk notes
Low. Stale data only shifts the guard's numbers a little.
Correlation minimum samples
corr_min_returns
- Default
- 96
- Range
- 10 to 1500
- Type
- Whole number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Fewer 15m returns than this and the correlation is 'unknown' (96 = one day).
What it does
The correlation between two coins is only calculated when both have at least this many common 15-minute returns. With fewer, the correlation is "unknown" and the guard does not count that pair as correlated.
Example
Minimum 96 (one day of 15m candles): a coin listed 10 hours ago has about 40 returns, so its correlation is unknown and it never blocks or is blocked by the guard.
When to change it
Raise it for more reliable numbers. Lower it if new coins should be included in the guard sooner.
Higher, lower, on or off
Higher: more stable correlations, but more coins are "unknown" and slip past the guard. Lower: new coins are included sooner, with noisier correlations that can block or allow trades by chance.
Risk notes
Low to medium. Unknown correlation is treated as not correlated, so a high value weakens the guard.
Liquidation learning: samples per coin
liq_learn_min_coin_samples
- Default
- 20
- Range
- 1 to 1000
- Type
- Whole number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: A coin uses its own liquidation-error offset once it has this many live samples; before that the all-coin offset.
What it does
In live trading the bot compares its liquidation estimate with Binance's real liquidation price after each fill and learns a safety offset. A coin uses its own offset once it has this many live samples. Until then it uses the offset learned over all coins.
Example
Minimum 20: SOL has 12 live samples, so it uses the all-coin offset. After its 20th sample, SOL's own offset is calculated; if that is lower than what it used before, it is only proposed on the dashboard for you to approve.
When to change it
Raise it to wait for more evidence before a coin gets its own offset. Lower it if you trade few coins often and want per-coin values sooner.
Higher, lower, on or off
Higher: more coins stay on the shared offset for longer. Lower: per-coin offsets earlier, based on fewer samples (noisier).
Risk notes
Medium. A per-coin offset from very few samples can be too small. Increases apply at once, decreases need your approval.
Liquidation learning: percentile
liq_learn_percentile
- Default
- 95 %
- Range
- 50 to 100 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: The safety offset is this percentile of the unsafe liquidation errors.
What it does
The learned liquidation safety offset is this percentile of the "unsafe" errors (cases where Binance liquidates closer to entry than the bot estimated). The offset is added to the stop-to-liquidation buffer.
Example
95 with 20 unsafe errors: the bot sorts them and takes the 19th smallest (nearest-rank), so the worst error is left out. At 100 it takes the largest of the 20. At 50 it takes the 10th.
When to change it
Raise it toward 100 for the most cautious offset. Lowering it is rarely a good idea.
Higher, lower, on or off
Higher: a larger offset, stops are kept further from liquidation, some entries may no longer fit. Lower: a smaller offset, more entries fit, but more cases where Binance's liquidation is closer than the bot allowed for.
Risk notes
High if lowered: a smaller offset means less room between the stop and liquidation.
Spoof rate window
spoof_rate_days
- Default
- 7 days
- Range
- 1 to 90 days
- Type
- Whole number
- Unit
- days
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Days of wall history used for a coin's spoof rate.
What it does
A coin's spoof rate is the share of its confirmed walls that were pulled (cancelled) rather than traded. It is calculated from the bot's own wall history over this many UTC days, and used by the "Max spoof rate" filter (max_spoof_rate).
Example
7 days: over the last 7 days BTC had 30 confirmed walls, 9 of them pulled, a spoof rate of 30%. With max_spoof_rate at 25, BTC is filtered out of the scan.
When to change it
Shorten it to react to recent behaviour. Lengthen it for a steadier figure. It has no effect while max_spoof_rate is 0 (off).
Higher, lower, on or off
Higher: a slower, more stable spoof rate. Lower: reacts faster to recent spoofing, but fewer walls are counted, so more coins fall under spoof_rate_min_walls and get no rate.
Risk notes
Low. Only affects which coins the filter removes.
Spoof rate minimum walls
spoof_rate_min_walls
- Default
- 20
- Range
- 1 to 10000
- Type
- Whole number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: A coin's spoof rate is only used once it has this many confirmed walls.
What it does
A coin's spoof rate is only calculated once it has at least this many confirmed walls in the spoof_rate_days window. Coins with fewer have no spoof rate and are never removed by the "Max spoof rate" filter.
Example
Minimum 20: a coin with 15 confirmed walls in 7 days, 10 of them pulled (67%), is not filtered because its rate is not used yet. At 20 walls the rate is used.
When to change it
Raise it for a more reliable rate. Lower it to filter suspicious coins earlier. No effect while max_spoof_rate is 0.
Higher, lower, on or off
Higher: fewer coins have a rate, so fewer are filtered, but each rate is more reliable. Lower: more coins are judged, on smaller samples.
Risk notes
Low. Only affects scanner filtering.
Order flow: near-wall range
flow_near_pct
- Default
- 0.5 %
- Range
- 0.05 to 10 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Aggressive trades within this % of a wall count toward its impact rate (analysis only).
What it does
Order-flow analysis only. Aggressive trades within this distance of a wall's near edge count toward that wall's near-wall net flow, which is used for the flow-reversal measurement in the logs.
Example
0.5%: a bid wall's top edge is at 2,000. Aggressive trades between 1,990 and 2,010 are counted for that wall; a trade at 2,015 is not.
When to change it
Only for research. Widen it on fast coins, narrow it to focus on trades right at the wall.
Higher, lower, on or off
Higher: more trades counted per wall, smoother but less specific flow figures. Lower: only trades right at the wall count, noisier figures.
Risk notes
Low. Analysis only; no trading decision uses it.
Order flow: bounce-room range
flow_room_pct
- Default
- 3 %
- Range
- 0.1 to 20 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: How far the bounce room looks for the opposite side's notional (analysis only).
What it does
Order-flow analysis only. The "bounce room" figure sums the opposite side's resting orders between price and the nearest counter wall. When there is no counter wall, it looks this far from the mid instead.
Example
3% with no counter wall and mid at 100: for a bid wall, the ask orders from 100 up to 103 are summed as bounce room.
When to change it
Only for research. Widen it for volatile coins.
Higher, lower, on or off
Higher: bounce room figures cover a larger range and come out larger. Lower: smaller range, smaller figures.
Risk notes
Low. Analysis only.
Bounce estimate cap
bounce_cap_pct
- Default
- 5 %
- Range
- 0.1 to 50 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: The estimated 5/15 min bounce never exceeds this move (wall-flow log, analysis only).
What it does
Wall-flow log only. The bot estimates how far price could bounce off a wall in 5 and 15 minutes from the current aggressive rate and the opposite side of the book. The estimate stops at the counter wall and never exceeds this move.
Example
Cap 5% with mid at 100 and no counter wall: for a bid wall, the 15-minute estimate walks the asks up to at most 105, so it reports at most 5% even if the flow would buy through more.
When to change it
Only for research. Raise it on very volatile coins if estimates are often stuck at the cap.
Higher, lower, on or off
Higher: larger estimates are possible on thin books. Lower: estimates are clipped sooner.
Risk notes
Low. Analysis only.
Watch: skip a coin after a pulled-after-fill trade
shadow_spoof_skip_watch_min
- Default
- 60 min
- Range
- 0 to 1440 min
- Type
- Number
- Unit
- min
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Log only (testnet/live, never changes a trade): after a trade closes with 'wall pulled after fill', the coin would be skipped for this many minutes. Every entry Wallbot places on it in that time is recorded as a would-skip, with its outcome (not filled, or the fill's net result and close reason), in the live-learning log ('spoof_skip_watch'), to test the rule on live data. 0 = off.
What it does
A log-only watch for the idea "when a coin spoofs us, skip it for a while". After a live trade closes with "wall pulled after fill", the coin would be skipped for this many minutes. Wallbot keeps trading normally; every entry it places on that coin in the window is recorded as a would-skip in the live-learning log (category spoof_skip_watch), followed by its outcome: not filled, or the fill's net result and close reason. The one-week report sums what the rule would have saved or cost.
Example
60: HYPEUSDT long closes 10:00 with "wall pulled after fill"; an entry placed on HYPEUSDT at 10:20 is logged as a would-skip, and when it closes at 10:31 with -0.12 USDT, the outcome says the rule would have saved 0.12.
When to change it
Leave it on to collect evidence; 0 switches the watch off.
Higher, lower, on or off
0: off. Higher: a longer would-skip window, more trades recorded.
Risk notes
None: nothing is ever skipped, it only writes log records.
Tag paper-shadow log lines
shadow_log_tag
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Paper-shadow lines in the log file start with [shadow], so they can't be mistaken for live events.
What it does
Lines from the paper shadow (the simulated copy that runs next to a live instance) start with "[shadow]" in the log file, so they cannot be mistaken for live events.
Example
On: "[shadow] BTCUSDT Placed long entry 0.003 @ 64250 ...". Off: the same line appears with no tag, looking exactly like a live order.
When to change it
Leave it on, especially on the live instance.
Higher, lower, on or off
On: shadow lines are tagged. Off: shadow and live lines look alike in the log file.
Risk notes
Low: log text only, but untagged lines can cause confusion when reading the log after an incident.
Log why a confirmed wall gets no entry
log_unused_wall_reasons
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: One line per wall and reason for the otherwise silent cases: the wall is not the nearest one and outside the nearest wall's stack span, or the other side of that coin already has a trade and both sides are not allowed. E.g. 'ENAUSDT support 0.23998-0.24011 not used: not the nearest wall (nearest 0.24388-0.24401), outside its stack span 1.5%'.
What it does
Logs one line per wall and reason when a confirmed wall gets no entry for a reason that is otherwise silent: it is not the nearest wall and lies outside the nearest wall's stack span, or the other side of that coin already has a trade and both sides are not allowed. Not logged by the paper shadow.
Example
"ENAUSDT support 0.23998-0.24011 not used: not the nearest wall (nearest 0.24388-0.24401), outside its stack span 1.5%". The same wall and reason is logged only once.
When to change it
Turn it off if the log is too busy; turn it on when you wonder why a wall you can see was not traded.
Higher, lower, on or off
On: these skips are explained in the log. Off: they stay silent.
Risk notes
Low: only changes what is logged.
Log why a closer wall is not used
log_replan_reasons
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: One line per wall and reason when a closer wall is seen but the re-plan is rejected or delayed, e.g. 'SHORT XRPUSDT: closer wall 0.5412-0.5415 not used: reward/risk 0.80 below 1'.
What it does
Logs one line per wall and reason when a closer wall is seen but the re-plan is rejected or delayed.
Example
"SHORT XRPUSDT: closer wall 0.5412-0.5415 not used: reward/risk 0.80 below 1; keeping the pending entry at 0.5398". Repeats for the same wall and reason are suppressed.
When to change it
Turn it off if the log is too busy; keep it on to understand why the bot kept an entry where it was.
Higher, lower, on or off
On: detailed reasons are logged. Off: only a short generic line in a few cases.
Risk notes
Low: only changes what is logged.
Disk budget guard
disk_guard
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Keeps everything under the disk-budget folder (WALLBOT_DISK_ROOT; on the VPS /wallbot) within the budget: deletes the oldest raw recordings, keeps at most N database backups, alerts, and pauses recording (never trading) when nearly full.
What it does
Keeps Wallbot's disk use under the budget: measures everything under the budget folder every few minutes (in a background thread), deletes the oldest raw recordings (recording instance only) and old database backups, alerts at the warning and high levels and pauses recording at the pause level. Trading is never paused by it.
Example
On the VPS with a 20 GB budget: recordings reach 15 GB at 00:05 UTC, so the Research instance deletes the oldest day's files; the dashboard shows 'Disk 18.2 / 20 GB (91%)'.
When to change it
Leave it on, especially on a shared server.
Higher, lower, on or off
On: the disk stays within the budget. Off: nothing is measured, pruned or alerted.
Risk notes
Low. Off: recordings and logs can fill the disk; on a shared server that can affect other services.
Disk budget: total
disk_budget_gb
- Default
- 20 GB
- Range
- 1 to 100000 GB
- Type
- Number
- Unit
- GB
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Everything Wallbot stores under the disk-budget folder (code, venv, databases, logs, recordings).
What it does
The total disk budget for everything under the budget folder (WALLBOT_DISK_ROOT, on the VPS /wallbot): code, venv, both data folders, logs and recordings.
Example
At 20 GB: usage 16 GB = 80%, a warning notification; 19 GB = 95%, recording pauses.
When to change it
Change it only if the server owner allows Wallbot more or less space.
Higher, lower, on or off
Higher: more room before alerts and pruning. Lower: earlier alerts, recordings pruned sooner.
Risk notes
Medium if set larger than the space really available to Wallbot.
Disk budget: raw recordings
disk_recordings_budget_gb
- Default
- 15 GB
- Range
- 0 to 100000 GB
- Type
- Number
- Unit
- GB
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Above this the recording instance deletes the oldest raw order-book/trade recordings (oldest UTC day first; never today's). The compact study logs are kept.
What it does
Budget for the raw order-book, trade and ticker recordings (backtest_data/live). Above it the recording instance deletes whole files, oldest UTC day first; today's files and files being written are never deleted.
Example
At 15 GB, recording about 4 GB a day: roughly the last 3 to 4 days are kept.
When to change it
Lower it to keep more room for other things; raise it (within the total) to keep more replay history.
Higher, lower, on or off
Higher: more days of raw data for backtests. Lower: fewer days.
Risk notes
Low: deleted recordings cannot be replayed any more; the compact study logs are kept.
Disk budget: warning at
disk_warn_pct
- Default
- 80 %
- Range
- 1 to 100 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Notify when usage reaches this share of the total budget.
What it does
Share of the total budget at which you get a disk-usage warning (once per level change).
Example
At 80% of 20 GB: a 'DISK 80% OF BUDGET' notification when usage reaches 16 GB.
When to change it
Lower it for earlier warning.
Higher, lower, on or off
Lower: earlier warning. Higher: later.
Risk notes
Low.
Disk budget: high at
disk_high_pct
- Default
- 90 %
- Range
- 1 to 100 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Notify again at this share; above it the recording instance also deletes old recordings until usage is under the 'prune to' level, and a paused recording resumes once usage is back under it.
What it does
Second alert level. Above it the recording instance also deletes old recordings until total usage is under 'prune to'; a paused recording resumes once usage falls back under this level.
Example
At 90% of 20 GB: at 18 GB you are alerted and old recordings are deleted down to 17 GB (85%).
When to change it
Keep it between the warning and pause levels.
Higher, lower, on or off
Lower: pruning and resume start earlier. Higher: later.
Risk notes
Low.
Disk budget: pause recording at
disk_pause_pct
- Default
- 95 %
- Range
- 1 to 100 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: At this share recording market data pauses (trading never does) and you are alerted.
What it does
At this share of the budget recording market data is paused (trading is not) and you get an alert. It resumes by itself under the high level.
Example
At 95% of 20 GB: at 19 GB the Research instance stops writing recordings; under 18 GB it resumes, each book starting with a fresh snapshot.
When to change it
Keep it above the high level.
Higher, lower, on or off
Lower: recording pauses earlier. Higher: closer to the hard limit.
Risk notes
Medium if too high: little room left before the budget is exceeded.
Disk budget: prune to
disk_prune_to_pct
- Default
- 85 %
- Range
- 1 to 100 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Above the high mark, old recordings are deleted until total usage is under this share.
What it does
When usage is above the high level, old recordings are deleted until total usage is under this share.
Example
At 85% of 20 GB: usage 18.3 GB triggers deletion of the oldest recordings until it is under 17 GB.
When to change it
Keep it below the high level.
Higher, lower, on or off
Lower: more deleted at once, fewer repeats. Higher: less deleted.
Risk notes
Low: only raw recordings are deleted.
Disk budget: database backups kept
disk_backups_keep
- Default
- 2
- Range
- 1 to 100
- Type
- Whole number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Newest backups (<database>.*.bak) kept per database file in this instance's data folder; older ones are deleted.
What it does
How many database backups (<database>.*.bak, made before each deploy) are kept per database file in this instance's data folder; older ones are deleted at the next check.
Example
At 2: when a third wallbot.db.*.bak appears, the oldest wallbot.db.*.bak is deleted.
When to change it
Raise it if you want more restore points and have room.
Higher, lower, on or off
Higher: more restore points, more disk. Lower: fewer.
Risk notes
Medium if 1: only the newest backup remains to roll back to.
Disk budget: wall-flow logs
disk_wallflow_budget_gb
- Default
- 1 GB
- Range
- 0 to 50 GB
- Type
- Number
- Unit
- GB
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: This instance's wall-flow log folder (data/wallflow) is kept under this size: above it the oldest session files are deleted first (never a file being written). The wall-flow log also pauses with recording when the disk is almost full. 0 = no separate limit (still counted in the total budget).
What it does
The size limit of this instance's wall-flow log folder (data/wallflow). Above it the disk guard deletes the oldest session files first, never a file that is being written. When the disk is almost full (disk_pause_pct) the wall-flow log pauses together with market-data recording, and resumes by itself. 0 = no separate limit; the folder still counts in the total budget.
Example
At 1 GB and about 3 MB per day, roughly a year of compact logs is kept; the oldest file goes first when the folder passes 1 GB.
When to change it
Raise it if you want to keep more history for studies; lower it on a small disk.
Higher, lower, on or off
Higher: more history kept. Lower: older files deleted sooner.
Risk notes
Low. Only analysis files are deleted.
Disk budget: check every
disk_check_interval_s
- Default
- 300 s
- Range
- 30 to 3600 s
- Type
- Number
- Unit
- s
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: How often disk usage is measured (in a background thread).
What it does
How often the disk budget is measured and acted on (a directory walk in a background thread).
Example
At 300 s: usage is checked every 5 minutes.
When to change it
Leave it; lower it only on a disk that fills very fast.
Higher, lower, on or off
Lower: quicker reaction, a little more disk reading. Higher: slower reaction.
Risk notes
Low.
Conviction score (log only)
conviction_score
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Score every confirmed wall the bot considers and every entry it places from graded factors (4h level, stack, refill, wall age, depth share, imbalance, group, stack position), show it on the trade card and in the Activity log, and follow what price did at each scored wall. Log only: it never changes an entry, an exit or a size.
What it does
Log only. For every confirmed wall the bot considers (traded or skipped by an entry check) and every entry it places, a score from graded factors: 4h level (on it 3 / within 0.5% 2 / within 1% 1), stack size, refill of the front wall, wall age, wall size vs the book within 1%, imbalance toward the trade, group and mover strength, stack entry position. Each factor's points and the total are kept; the trade card and the Activity log show them, and every scored wall is followed (touched, held or broke) in data/conviction/.
Example
score 11: 4h 3, stack 1, refill 1, age 2, depth 2, imbalance 1, group 1, position 0 on a core long in front of a wall on a 4h level.
When to change it
Leave it on: it only records. The analysis after 4 weeks is pre-registered (research/2026-11/conviction-score).
Higher, lower, on or off
No trading effect. A few small file writes per minute; the 4h levels come from the S/R filter's cached candles.
Risk notes
None: log only.
Conviction: defended-zone lookback
conviction_zone_window_min
- Default
- 30 min
- Range
- 5 to 240 min
- Type
- Whole number
- Unit
- min
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: The defended-zone factor counts the zone's touch -> refill cycles over this many minutes.
What it does
Conviction score (log only): the defended-zone factor counts the stack zone's touch -> refill cycles at this wall over this many minutes.
Example
30: cycles in the last half hour.
When to change it
Keep it fixed during an analysis period.
Higher, lower, on or off
Changes only the logged score.
Risk notes
None: log only.
Conviction: swing-point history
conviction_swing_lookback_days
- Default
- 30 days
- Range
- 7 to 180 days
- Type
- Whole number
- Unit
- days
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: The swing factor uses every 4h swing point (pivot high / low, unmerged) of this many days, independent of the level lookback and the level tolerance.
What it does
Conviction score (log only): the swing factor uses every 4h swing point (pivot high or low, the same pivot rule and strength as the levels, but not merged and with no tolerance) of this many days. Independent of sr_lookback_days and of the level tolerance.
Example
30: the swing points of the last 30 days. Points: distance from the wall to the nearest swing point on its side of price (on it <= 0.1% 3, <= 0.5% 2, <= 1% 1), swing points within +-1% (1 / 2 / 3+), and its type (a swing low under a long or a swing high over a short 2; flipped 1).
When to change it
Keep it fixed during an analysis period (a change restarts it).
Higher, lower, on or off
Changes only the logged score; one extra 4h candle request per coin per 4 hours.
Risk notes
None: log only.
Conviction log: trendlines
conviction_trend_log
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Log only, not part of the score: with each scored wall, the distance to the nearest active 3-touch trendline on 4h and 1D, a recent break of the trade's action line and its retest, and the 4h safety line.
What it does
Log only, NOT part of the conviction score: with every scored wall and entry, the Activity line and the conviction record also show trendline facts (Tori method, the same rules as the trendline studies): the distance to the nearest active 3-touch trendline on 4h and on 1D, whether price closed through the trade's action line (a long: a falling line through highs; a short: a rising line through lows) within the last few candles and came back to it (retest), and the distance to the 4h safety line (a long: through the two latest swing lows; a short: highs). Built only from candles closed at the time.
Example
'...; trend: 4h line 0.4%, 1D line 2.1%, break 4h 3c ago, retest yes, safety +0.8%'.
When to change it
Leave it on; the 9 Dec conviction analysis compares these as an exploratory extra.
Higher, lower, on or off
No trading effect; two extra candle requests per coin (4h, 1D) per new candle.
Risk notes
None: log only.
Conviction log: trendline break lookback
conviction_trend_break_candles
- Default
- 12
- Range
- 1 to 100
- Type
- Whole number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: A break of the trade's action line counts as recent within this many closed candles of its timeframe.
What it does
Conviction log trendlines: a break of the trade's action line counts as recent within this many closed candles of its own timeframe (4h or 1D).
Example
12: on 4h, a break in the last 48 hours; on 1D, the last 12 days.
When to change it
Keep it fixed during the analysis.
Higher, lower, on or off
Changes only the logged facts.
Risk notes
None: log only.
Conviction log: best-fit trendlines and A/B case
conviction_bf_log
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Log only, not part of the score, in parallel to the trendline facts: best-fit trendlines (most swing points near the line, fewest closes through it, moderate slope) on 4h and 1D with the same distance / break / retest facts, and the owner's safety-line case for the wall (A breakout, B ongoing trend).
What it does
Log only, NOT part of the conviction score, in parallel to the r51 trendline facts (which stay exactly as they are). With every scored wall and entry: best-fit trendlines on 4h and 1D: lines through two of the last 60 confirmed swing points of a kind with at least 3 swing points within 0.25 x ATR(14) of the line, at most 5% of its candles (and at least 2 allowed) closing through it up to its last touch, and a moderate slope (0.02-0.5 x ATR per candle); ranked by most swing points near it, then fewest closes through, then the longer span. Same distance / break / retest facts as the trendline log, plus the owner's safety-line case of the wall: A (breakout: the trade's action line closed through within the break lookback and no line of the trade's own kind yet; initial stop at the breakout candle's extreme) or B (ongoing trend: a line of its own kind is active).
Example
'...; bf: 4h line 0.4% (5 touches), 1D line none, break 4h 3c ago, retest yes; sl: 4h A (stop +1.2%)'.
When to change it
Leave it on; the 9 Dec analysis compares the r51 lines and these best-fit lines (exploratory).
Higher, lower, on or off
No trading effect; computed once per new candle per coin.
Risk notes
None: log only.
Conviction log: Tori-style trendlines
conviction_tori_log
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Log only, not part of the score, in parallel to the other trendline facts: Tori-style action and safety lines on 4h and 1D (wick extremes with a 1-candle swing rule, 3+ touches over at least a week, no close through between the touches, moderate slope, retired at the first close through) and whether the 4h and 1D directions agree.
What it does
Log only, NOT part of the conviction score, in parallel to the r51 and best-fit trendline facts (both unchanged). With every scored wall and entry, on 4h and 1D: Tori-style lines through wick extremes (a swing high / low with 1 candle on each side), at least 3 touches within 0.25 x ATR(14), at least one week from the first to the last touch, no candle closing through the line between its touches, slope 0.02-0.5 x ATR per candle, retired for good at the first close through it. Logged: distance to the trade's action line (falling for a long, rising for a short) and its break (candles ago) and retest; distance to the safety line (the opposite line already there: rising under a long, falling over a short; + = the protected side, where the stop would sit); and whether the 4h and 1D directions agree (top-down).
Example
'...; tori: 4h act 0.4% (3t), safe +0.8%; 1D act none, safe +2.1%; no recent break; top-down agree'.
When to change it
Leave it on; the 9 Dec analysis compares the three line versions (exploratory).
Higher, lower, on or off
No trading effect; computed once per new candle per coin.
Risk notes
None: log only.
Log market context at entry and fill
log_market_context
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Log only: at every entry placement and every fill, store the anchor wall's size and share of the book within 1%, bid / ask depth within 0.5% and 1%, the spread, the coin's 24h volume and change, traded volume and trade count in the last 60 s and 5 min, the stack's walls and the wall's age (trades.mkt_entry / mkt_fill).
What it does
Log only, no trading effect. At every entry placement and every fill (both instances, paper and live, and the shadow), the trade stores a snapshot of the market as JSON: trades.mkt_entry at placement and trades.mkt_fill at the fill. Fields: wall_usdt (the anchor wall's size; at a fill empty if the wall is no longer tracked), wall_share_1pct (its share of bids + asks within 1% of mid), bid_05 / ask_05 / bid_1 / ask_1 (book depth in USDT within 0.5% and 1%), spread_bps, vol24 and chg24 (24h quote volume and price change from the last scan), vol_60s / n_60s / vol_5m / n_5m (traded USDT and number of trades in the last 60 s and 5 min), stack (the stack's walls: low, high, USDT) and wall_age_s (seconds since the wall was first seen). Stored in the database, so it survives restarts.
Example
A long placed on a 310,000 USDT bid wall: wall_share_1pct 0.14, bid_1 1.2M / ask_1 1.0M, spread 1.8 bps, vol_60s 85,000 in 140 trades, wall age 95 s.
When to change it
Leave it on.
Higher, lower, on or off
A few hundred bytes per trade.
Risk notes
None: log only.
Conviction log: request budget kept for trading
conviction_log_min_budget_pct
- Default
- 50 %
- Range
- 0 to 100 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Log only: a candle request of the conviction log (trendlines, market structure, swing points) is only sent while at least this share of the request budget is left; otherwise that cycle uses the candles already cached (one close older) or skips the feature, and the record says so. It never waits.
What it does
Log only. The conviction log's own candle requests (swing points of the score, trendlines, best-fit and Tori-style lines, market structure) are sent only while at least this share of the request budget is left (requests per minute; the budget itself is rate_limit_budget_pct of Binance's limit) and Binance is not backing off. Otherwise that cycle uses the candles already cached (closed candles, one close older; the record lists them under budget.stale_tf_s) or, when nothing is cached yet, skips the feature (budget.skipped). It never waits, so trading requests always come first. The scoring itself runs in the background and never holds up order-book processing or fill detection.
Example
At 10:00 UTC the budget is 35% left after the hourly scan: a wall scored then uses the 4h trendlines from the candles fetched at 06:00 and records budget.stale_tf_s = [14400]; the next wall after the budget recovers fetches the new candle.
When to change it
Leave it at 50%.
Higher, lower, on or off
Fewer logging requests when trading needs the budget.
Risk notes
None: log only.
Conviction log: hypothetical stack positions
conviction_pos_log
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Log only, no orders: for every scored wall in a stack of 2+ walls, a hypothetical wall-hug entry at each position (front, biggest, middle, second to last, last) with one common stop behind the last wall, followed 60 minutes to a fill and 4 hours after it; plus the leverage the bot would pick and a risk-targeted one.
What it does
Log only: no order is ever placed, nothing changes in trading and the conviction score does not read it. For every scored wall in a stack of 2 or more walls, the conviction log computes a hypothetical wall-hug entry at each stack position: front (the nearest wall), biggest, middle, second to last and last (the farthest), each entry_ticks in front of that wall's core orders, with one common stop stop_beyond_wall_pct beyond the farthest wall's far edge. Each is followed: filled when price trades through the entry by at least one tick within 60 minutes of scoring, then for 4 hours the first of the stop or +0.5 / +1 / +2 % from the entry, the best and worst move, R at the end and the stack's walls still standing at the fill. Also recorded: the entry-to-stop distance, the leverage the bot would pick (the live planner and its checks) with the margin share lost at the stop, and the same for a risk-targeted leverage (50% / the distance, capped by the coin's maximum and the liquidation / fee checks). Planned in a worker thread from cached data; the leverage brackets of all symbols are refreshed at start and every 4 hours with one read-only request (weight 1; it never sets or changes leverage), and a record says whether its brackets were cached or assumed. Each fill, stop or target is saved at once as one short line in a journal, and the full position records at most every 60 seconds; both are restored after a restart.
Example
A 3-wall bid stack at 99.6 / 99.3 / 98.9: entries 99.62, 99.32, 98.92 (2 ticks), common stop 98.4 (0.5% beyond 98.9); the front entry fills after 4 minutes, reaches +0.5% after 20 minutes, then the stop: first +0.5.
When to change it
Leave it on; the 9 Dec analysis compares the five positions (exploratory, with the 8787 stack-entry paper test).
Higher, lower, on or off
No trading effect; a few hundred bytes per scored stack and a little planner CPU in a worker thread.
Risk notes
None: log only.
Conviction log: market structure
conviction_ms_log
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Log only, never part of the score: liquidity sweeps (15m, 1h), equal highs / lows (1h, 4h), fair value gaps (15m, 1h, 4h), order blocks (1h, 4h), the latest break of structure (15m, 1h, 4h) and the session, from completed candles, built once per candle close per coin.
What it does
Log only, NOT part of the conviction score and never a trading input. For every scored wall the conviction log also records market-structure facts from completed candles (built once per candle close per coin and timeframe, cached): a liquidity sweep in the last 8 candles on 15m and 1h (price traded beyond equal highs / lows or a prior swing on the wall's side: depth, closed back inside, candles ago, distance to the wall), the nearest equal highs / lows on the wall's side on 1h and 4h (distance, touches, swept yet), an unfilled fair value gap containing the wall or within 0.3% on 15m, 1h and 4h (direction, size, age, share filled), an order block (the last opposite candle before an impulsive move of 1.5 x ATR or more that broke structure) containing the wall or within 0.3% on 1h and 4h (direction, age, tested before), the latest break of structure on 15m, 1h and 4h (direction, BOS / MSS, candles ago, agrees with the trade) and the session (Asia, London, New York, late; within 60 minutes of the London 07:00 or New York 13:30 UTC open).
Example
A bid wall at 99.5 scored at 13:50 UTC: 'ms: sweep 15m 2c; eq 1h 0.4% (3); fvg 1h in; ob none; bos 1h bullish agree; new_york (ny_open)'.
When to change it
Leave it on; the 9 Dec analysis compares walls with and without each feature (exploratory).
Higher, lower, on or off
No trading effect; two more candle requests per coin per 15 minutes / hour while it is scored.
Risk notes
None: log only.
Conviction log: stack-zone outcome
conviction_zone_outcome
- Default
- On
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Log only: for every scored wall in a stack of 2 or more walls, also follow the whole stack zone (all its walls combined) with the same 60 / 30 minute windows: touched, held / broke, how it ended (held, all pulled, only the front wall pulled, partly pulled, absorbed, faded), the walls still standing when price arrived and the zone's size at the touch vs at scoring. The single-wall outcome is unchanged.
What it does
Log only, NOT part of the conviction score; the single-wall outcome stays exactly as it is. For every scored wall whose stack (confirmed same-side walls with their centre within stack_max_span_pct of its centre) has 2 or more walls, the whole stack zone (from the lowest to the highest of those walls) is followed too, with the same windows: touched = price reached the zone's near edge within 60 minutes; broke = after the touch a price more than 0.1% beyond the zone's far edge within 30 minutes; held = touched and not broke. Also recorded: how the zone ended while followed (held = no member wall ended; all_pulled; front_only_pulled = only the wall nearest to the price was pulled; partly_pulled; absorbed = a member was eaten; faded), how many of its walls were still standing when price arrived, and its total size at scoring and at the touch. Saved and restored across restarts like the wall outcome (after a restart the member walls' ends are no longer matched).
Example
A 3-wall bid stack: the front wall is pulled 5 minutes after scoring, price touches the zone 12 minutes later with 2 walls standing, size 2.1M at scoring / 1.4M at the touch, no break: zone held, end front_only_pulled; the single-wall record says pulled, not touched.
When to change it
Leave it on; the 9 Dec analysis compares the single-wall and stack-zone outcomes (exploratory).
Higher, lower, on or off
No trading effect; a few more numbers per scored stack.
Risk notes
None: log only.
Conviction: follow after the touch
conviction_track_min
- Default
- 30 min
- Range
- 5 to 240 min
- Type
- Whole number
- Unit
- min
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: How long after a scored wall's first touch its outcome (held / broke) is followed.
What it does
How long after a scored wall's first touch the outcome is followed: broke = a price more than 0.1% beyond its far edge in that time; held = touched and not broke.
Example
30: a wall touched at 12:00 is judged at 12:30.
When to change it
Keep it fixed during an analysis period.
Higher, lower, on or off
Longer: more walls count as broke.
Risk notes
None: log only.
Conviction: keep records
conviction_keep_days
- Default
- 120
- Range
- 7 to 730
- Type
- Whole number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Days of score records kept in the instance's data folder (conviction/).
What it does
Days of conviction records kept (data/conviction/conv_YYYYMMDD.jsonl).
Example
120: about four months.
When to change it
Lower it if disk is short.
Higher, lower, on or off
Older files are deleted once a day.
Risk notes
None.
Conviction weight: 4h level
conviction_w_sr4h
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the 4h level factor in the conviction score (log only).
What it does
Weight of the 4h level factor in the conviction score total (total = sum of weight x points). Log only.
Example
1.0: the 4h level points count once; 2.0 doubles them; 0 leaves the factor out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period; a change starts a new one.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: stack
conviction_w_stack
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the stack factor in the conviction score (log only).
What it does
Weight of the stack factor in the conviction score total (total = sum of weight x points). Log only.
Example
1.0: the stack points count once; 2.0 doubles them; 0 leaves the factor out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period; a change starts a new one.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: refill
conviction_w_refill
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the refill factor in the conviction score (log only).
What it does
Weight of the refill factor in the conviction score total (total = sum of weight x points). Log only.
Example
1.0: the refill points count once; 2.0 doubles them; 0 leaves the factor out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period; a change starts a new one.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: wall age
conviction_w_age
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the wall age factor in the conviction score (log only).
What it does
Weight of the wall age factor in the conviction score total (total = sum of weight x points). Log only.
Example
1.0: the wall age points count once; 2.0 doubles them; 0 leaves the factor out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period; a change starts a new one.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: depth share
conviction_w_depth
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the depth share factor in the conviction score (log only).
What it does
Weight of the depth share factor in the conviction score total (total = sum of weight x points). Log only.
Example
1.0: the depth share points count once; 2.0 doubles them; 0 leaves the factor out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period; a change starts a new one.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: imbalance
conviction_w_imbalance
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the imbalance factor in the conviction score (log only).
What it does
Weight of the imbalance factor in the conviction score total (total = sum of weight x points). Log only.
Example
1.0: the imbalance points count once; 2.0 doubles them; 0 leaves the factor out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period; a change starts a new one.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: group
conviction_w_group
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the group factor in the conviction score (log only).
What it does
Weight of the group factor in the conviction score total (total = sum of weight x points). Log only.
Example
1.0: the group points count once; 2.0 doubles them; 0 leaves the factor out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period; a change starts a new one.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: stack position
conviction_w_position
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the stack position factor in the conviction score (log only).
What it does
Weight of the stack position factor in the conviction score total (total = sum of weight x points). Log only.
Example
1.0: the stack position points count once; 2.0 doubles them; 0 leaves the factor out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period; a change starts a new one.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: defended zone
conviction_w_defended
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the defended-zone factor (touch -> refill cycles) in the conviction score (log only).
What it does
Weight of the defended zone (touch -> refill cycles: 0, 1, 2, 3+ = 0-3 points) factor in the conviction score total. Log only.
Example
1.0: counted once; 0 leaves it out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: zone refill
conviction_w_zone_refill
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the zone refill-ratio factor in the conviction score (log only).
What it does
Weight of the zone refill ratio after the most recent hit (no hit 1, >= 0.8 2, >= 0.5 1, else 0) factor in the conviction score total. Log only.
Example
1.0: counted once; 0 leaves it out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Conviction weight: swing points
conviction_w_swing
- Default
- 1
- Range
- 0 to 10
- Type
- Number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Weight of the swing-point factor (distance + cluster + type) in the conviction score (log only).
What it does
Weight of the swing-point factor (distance + cluster + type, up to 8 points) in the conviction score total. Log only.
Example
1.0: counted once; 0 leaves it out of the total (its points are still stored).
When to change it
Keep the weights fixed during an analysis period.
Higher, lower, on or off
Changes only the logged total.
Risk notes
None: log only.
Unlock-short paper test
unlock_short_enabled
- Default
- Off
- Range
- On or off
- Type
- On/off
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: The pre-registered unlock-short rule as a forward test on paper: short a coin at 00:00 UTC seven days before a cliff unlock of at least 1% of its circulating supply, with a BTC long of the same size, both closed when the unlock day ends. Own paper balance and pairs; never touches the wall bot's slots, caps or wallet. Paper mode on the Research instance only (refused elsewhere in code).
What it does
Runs the pre-registered unlock-short rule as a forward test on PAPER: once a day (22:30 UTC) it reads DefiLlama's public unlock data and checks it (fresh, format unchanged, 12 reference unlocks unchanged); at 00:00 UTC it shorts each coin whose cliff unlock (at least 1% of circulating supply) is 7 days away, with a BTCUSDT long of the same notional, and closes both legs at 00:00 UTC after the unlock day. Results come from the daily closes (as in the backtest), the prices read at 00:00 and a stop variant, all with fees, slippage and real funding. Own balance and pairs: the wall bot's slots, caps and wallet are never used. Refused in code on the Trading instance and outside paper mode.
Example
An unlock of 2.4% of circulating on 2026-12-15: the pair opens 2026-12-08 00:00 UTC (short 50 USDT of the coin, long 50 USDT BTC) and closes 2026-12-16 00:00 UTC.
When to change it
Switch it on only for the forward test on the Research instance (8787), after its pre-registration is written.
Higher, lower, on or off
On: pairs are opened and closed on paper as described. Off: nothing new opens; pairs already open still close and settle.
Risk notes
None for real money: paper only, no order is sent anywhere.
Unlock-short: paper balance
unlock_short_paper_balance
- Default
- 500 USDT
- Range
- 50 to 1e+06 USDT
- Type
- Number
- Unit
- USDT
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: The test's own starting paper balance (separate from the wall bot's).
What it does
The unlock-short test's own starting paper balance. Each pair's executable result is added to it; the wall bot's paper balance is separate.
Example
At 500 with 50 USDT per leg and 2x: each pair ties up 50 USDT of margin, so 10 pairs use the whole 500.
When to change it
Set it before the test starts; leave it during the test.
Higher, lower, on or off
Higher: more room for pairs. Lower: pairs are skipped once the margin does not fit.
Risk notes
None (paper).
Unlock-short: size per leg
unlock_short_notional_usdt
- Default
- 50 USDT
- Range
- 5 to 100000 USDT
- Type
- Number
- Unit
- USDT
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Notional of the short and of the BTC hedge in each pair (2x: margin is half of it per leg).
What it does
Notional of each leg of a pair: the short on the unlocking coin and the BTCUSDT long. Results are also reported per 100 notional.
Example
At 50: short 50 USDT of the coin and long 50 USDT of BTC; at 2x each leg uses 25 USDT of margin.
When to change it
Leave it at 50 for the forward test.
Higher, lower, on or off
Scales the USDT results; per-100 results do not change.
Risk notes
None (paper).
Unlock-short: max open pairs
unlock_short_max_pairs
- Default
- 10
- Range
- 1 to 50
- Type
- Whole number
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: At most this many pairs open at once; later unlocks are skipped (and counted) while all are in use.
What it does
The most pairs open at once. On a day with more new unlocks than free slots, coins are taken in alphabetical order and the rest are skipped and counted.
Example
At 10: in the backtest about 84% of the unlocks found a free slot.
When to change it
Leave it at 10 during the test.
Higher, lower, on or off
Higher: fewer skipped unlocks, more margin in use. Lower: the opposite.
Risk notes
None (paper).
Unlock-short: stop variant
unlock_short_shadow_stop_pct
- Default
- 30 %
- Range
- 5 to 45 %
- Type
- Number
- Unit
- %
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: The stop variant calculated alongside each pair: both legs closed when the short's mark price is this far above its entry. Reported, not judged; the pre-registered rule has no stop.
What it does
A stop variant calculated on the same pairs: when the short's mark price rises this far above its entry (checked on hourly mark-price highs), both legs of the variant close. Reported next to the rule's result; it never changes the paper pairs themselves.
Example
At 30: a short entered at 1.000 is closed in the variant when the mark reaches 1.300 (plus slippage); in the backtest this cut the worst pair from -80 to -46 per 100.
When to change it
Leave it at 30 during the test.
Higher, lower, on or off
Lower: the variant stops out more often. Higher: closer to no stop.
Risk notes
None (calculated only).
Unlock-short: cross-check source
unlock_short_crosscheck
- Default
off- Choices
off,coinmarketcal,coindar- Type
- Choice
- Live confirmation
- No. Changes apply in every mode without the typed confirmation.
In the dashboard: Optional daily cross-check of upcoming unlocks with an event calendar (free key saved under API keys). Never opens a pair; a 'postponed / cancelled' report closes it at the next 00:00 UTC.
What it does
Optional daily cross-check of the unlocks in the next 14 days (and of open pairs) against an event calendar, with a free key you save under API keys. off: none. coinmarketcal: CoinMarketCal's API. coindar: Coindar's API. A matching unlock or vesting event within 3 days counts as confirmed; a postponed or cancelled one closes the open pair at the next 00:00 UTC (listed as 'moved'). It never opens a pair.
Example
coinmarketcal: an unlock on 2026-12-15 is confirmed by an event dated 2026-12-14 to 2026-12-18.
When to change it
Optional; only if you have a free key.
Higher, lower, on or off
Adds a confirmed / not found / postponed column to the upcoming unlocks.
Risk notes
Low: an error only marks the cross-check unavailable.