Filters
The P&L books and every table below follow these choices; All is the default and means the whole experiment. The URL carries the filter, so a view can be linked to as it stands.
Collector service
Read from the engine checkpoint, the writer lease and the systemd unit. A checkpoint that stops moving is the first sign the collector died; a unit that is active while the checkpoint is frozen means it is wedged rather than gone.
stalled checkpoint has not moved for 253.2 h (stale after 180 s); unit is active but its checkpoint is not advancing
The collector is not writing normally. Read the service logs and the runbook before restarting it: a wedged collector still holds the lease and the journal, and a restart is safe but not free.
Current P&L
realised plus the mark-to-market of positions still open, summed over every branch: paired-cohort branches are alternative worlds funded separately, so read it as what the lab is worth, not as one strategy's return. Realised is net of fees and funding and already contains the profit swept out of the bots, so the withdrawn figure in the run panel is not extra.
Realised is +0.0000 from branches still running and +0.0000 from branches that have ended; an ended branch is left out of the open figure because its equity snapshots stop when it ends.
Run
One config hash is one experiment. Changing any strategy parameter opens a new version; this one is never mutated in place.
paired-cohort branches are alternative worlds; their PnL is not summed.
Ledger per currency
- USDT 0.0000000000 balanced
Branches
- none
Cohorts
- none
Data health
trustworthy is false whenever the spool has a backlog, which live mode hits routinely. Read the matrix as provisional on those samples.
Rows recorded while untrusted are marked data_degraded.
Method matrix
Every selector, entry and exit arm that has produced a branch. Losing and bankrupt arms stay in the denominator, so a high mean PnL on three cohorts is still three cohorts.
| Selector | Entry | Exit | Cohorts | Branches | Mean PnL | Median PnL | Worst | Mean DD | Liq | Horizon end | Not finished | TP cycles | Hold h | Fees | Funding | Samples |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| No outcomes recorded yet. | ||||||||||||||||
Not finished means the branch was still open at the cutoff; its PnL is an interim mark, not a result.
Paired exits
Each exit is differenced against the no-exit baseline inside the same cohort, then bootstrapped over cohorts. Five branches of one cohort are one market event, not five samples.
| Exit | Cohorts | Bootstrap CI 95% | Mean diff | Median diff | Wins | Losses | Ties |
|---|---|---|---|---|---|---|---|
| Needs at least one closed cohort. | |||||||
Bars share one scale. A bar that crosses the centre line is consistent with no difference; a bar wholly to the right favours that exit.
Held slots
Each selector holds up to its max_per_side pairs per direction, as set in the experiment config. A draining slot has left the ranking but keeps its place until its cohort closes. PnL, deals and SO are those of the slot's open cohort.
| Selector | Instrument | Side | Rank | State | Entered | PnL | Completed deals | SO filled |
|---|---|---|---|---|---|---|---|---|
| No slots held. | ||||||||
Recent cohorts
One entry moment, five exit policies. Open a cohort to compare the branches that share it.
| Cohort | Selector | Entry | Instrument | Side | State | Started | Horizon end | Capital/branch | PnL | Completed deals | SO filled |
|---|---|---|---|---|---|---|---|---|---|---|---|
| No cohorts yet. | |||||||||||