| Claude modelclaude-opus-5-5 by default. | | default claude-opus-5-5 |
| Effortlow, medium, high, xhigh or max. | | default high |
| Allow web searchLets the analyst check unlocks, listings and news. | | default True |
| Max searches per analysis | | default 3 |
| Alert threshold (confidence)Long calls below this are logged but not sent. 0 to 1. The regime offsets below move it. | | default 0.55
learner may adjust within 0.45–0.8 |
| Bar offset: BTC downtrendAdded to the threshold when BTC is in a downtrend. The learner may tune it. | | default 0.1
learner may adjust within 0.0–0.25 |
| Bar offset: choppy / unknownAdded when BTC is choppy or the regime is unknown. | | default 0.05
learner may adjust within 0.0–0.15 |
| Bar offset: broad uptrendAdded (usually negative) when BTC is in an uptrend and breadth is broad. | | default -0.05
learner may adjust within -0.15–0.0 |
| Broad participation (% of top 200 up)Breadth at or above this counts as broad. | | default 60.0 |
| Horizon the analyst is told (hours)Free text such as '2 to 12'. Short, sharp runs; the tracker clamps to the maximum under Tracking. | | default 12 to 48 |
| Intraday base rates the analyst is toldMeasured numbers about how fast breakouts peak and fade. Re-measure now and then and paste the new ones here. | | default Measured on this universe over the last week: 15-minute breakouts through the 24-hour high on 2x volume peaked after a median 6 hours (41% within 4 hours, 65% within 12); +3% was reached 43% of the time (median 3.75 hours) and +5% 28% (median 5.75 hours); the median 24-hour return after a breakout was -1%, and breakouts that closed back under the level within two 15-minute bars (17%) went on to -1.9%. Targets must be modest and exits quick. |
| Runner base rates the analyst is toldMeasured numbers about the 60% runs: where they start, how long they take, what the exits captured. | | default Measured over four months on this universe (98 Kraken coins): 36 trough-to-peak runs of +60% or more a month across the universe, a median 20 days from trough to +60%, starting a median 22% under the 20-day high, not from coils. Entries off the lows (a reclaim of the 10-day high from near the 20-day low, a first close back above the 20-day EMA, a volume jump near the low) held with a 10% stop and a 30% trailing stop averaged +19% net per entry over 45 days with 54% winners and one in ten reaching +60%; the 20-day-high breakout entry averaged +10%. Losers were stopped in a median 6 days; the 30%+ winners took a median 36 days. |
| Spike base rates the analyst is toldMeasured numbers about quick breakouts: how often they run, what the early signs were, how fast the peak came. | | default Measured on this universe over 20 days of 4-hour bars (98 Kraken coins): a close through the prior 24-hour high on 2x volume reached +15% within 48 hours 22% of the time and +30% 5% of the time, and the peak came a median 16 hours after the trigger. When the week was already up 10% or more and volume was building the day before, 29 to 34% reached +15% and 6 to 7% reached +30%; a break out of a flat week reached +15% only 8% of the time; a trigger bar already up 8% or more lost money on every exit tried. Quick +60% days happen about three times a month across the whole universe; +30% inside two days about 37 times. |
| Re-analyse same setup after (h)A coil that turns into a breakout is always re-analysed. | | default 18.0 |
| Max Claude calls per dayHard cost cap. | | default 40 |
| Keep back for triggersScheduled scans stop this many short of the cap so a trigger that prints late in the day can still be analysed. | | default 8 |
| Monthly spend ceiling (USD)Estimated Claude spend per calendar month. A Telegram warning at 80%; at 100% no analysis runs until the 1st. 0 means no ceiling. | | default 60.0 |
| Also send 'watch' callsNoisier. | | default False |