Reading a run while it is still running
Run Desk covers one hour of the job: the hour a campaign is open and every decision costs something. What the telemetry in front of you actually says, which readings deserve a response, which deserve nothing at all, and the conditions that end the run whether you feel ready or not.
Every threshold on this site is a line you set in advance. The desk publishes no measured rates, no targets and no results.
What this desk covers, and what it deliberately leaves alone
Most writing about trading automation stops at the moment you press start. This desk begins there. The live window has its own skills, its own failure modes and its own temptations, and none of them are the same as the ones you dealt with while configuring.
In scope
The window between the first submitted transaction and the last settled one.
- What a live reading is made of, and where each part of it comes from
- Which mid-run adjustments are worth their cost and which are fidgeting
- Stop conditions with an owner, a window and an action attached
- The close-out sequence and the four ledgers that have to agree
Out of scope
Work that belongs before the run, or to somebody else's discipline entirely.
- Choosing or testing a tool before you trust it with live value
- Key custody, funding procedure and wallet lifecycle mechanics
- Send-path engineering: priority fees, bundles, landing rate design
- Anything resembling a prediction about price, demand or outcome
House rules
Constraints that decide what appears on a page here and what stays off it.
- No measured rates, because the desk runs no campaigns to measure
- Every threshold is presented as a choice with two directions of cost
- Arithmetic is shown in full and labelled illustrative, never sourced
- Irreversible steps are named as irreversible before the step, not after
The live loop, in the order it happens
Five stages, and the discipline is doing them in order rather than jumping to the one that feels urgent. Skipping a stage does not remove it; it moves it to the point where it costs more to perform.
Open
Establish the first reading before anything can be blamed on drift.
stage 01Read
Take the same measurements from the same sources every time you look.
stage 02Decide
Choose one of five responses, and know that observing is one of them.
stage 03Stop
Execute the rule you already wrote instead of arguing with the screen.
stage 04Close
Quiesce, settle, reconcile four ledgers, then write the run report.
stage 05Four pieces that carry the method
Read in this order the first time. After that, the one you need is usually the one you are already arguing with yourself about.
Monitoring a volume bot run
A watch loop with a cadence: what the first check is for, what the tenth is for, where each answer comes from, and the run journal that makes the next hour readable.
Live runOpenThe signals worth watching
Six families of run health signal, what each one refuses to tell you, and how long a sampling window has to run before a reading means anything at all.
SignalsOpenWhen to intervene and when not to
An escalation ladder from observe to abort, the two-reading rule that keeps you off the controls, and the reversibility test every mid-run change has to pass.
Live runOpenStop conditions written in advance
A stop rule needs a signal, a window, a threshold, an action and an owner. Write all five before the run opens, because you will not write them well during it.
SignalsOpenTwo sections and a question list
One section is about the hour itself. The other is about what a reading is allowed to trigger. The question list catches everything that arrives sideways while a run is open.
Live run
The hour the run is open: watching it, deciding about it, and closing it out.
- Monitoring a volume bot run
- When to intervene and when not to
- Closing a run and reconciling
Signals
What a reading is made of, and the rules that decide what a reading may trigger.
- The signals worth watching
- Stop conditions written in advance
- Sampling windows and threshold cost
Questions
Twelve short answers to the things operators ask at the worst possible moment.
- Handing a live run to somebody else
- Telling a finished run from a stalled one
- When the console and the chain disagree
Signal, then decision, then nothing
The table beside this is the compressed version of the framework. Each row names a reading, the response class it feeds, and the mistake that follows from treating it as something it is not. The full reasoning, including how long a window has to run before a reading counts, sits in the signals section.
Note how often the correct response is to take a second reading. Most bad mid-run decisions are made on a single data point that would have looked ordinary five minutes later.
Open the signals section| Reading | Feeds | Common misread |
|---|---|---|
| Submission failures rising | Second reading, then adjust | Treating one bad minute as a trend |
| A wallet under its floor | Adjust, or pause that wallet | Topping up without asking why it drained |
| Route changed venue | Observe, verify against plan | Assuming the plan still describes the run |
| Pace ahead of schedule | Observe | Slowing a run that is simply going well |
| Nothing has changed for a while | Liveness check | Reading a frozen panel as a calm one |
| A stop condition met | Stop, no debate | Renegotiating the rule while it fires |
What a hosted console changes about the live window
Everything on this site assumes you can see the run. That assumption is doing a lot of work. Operators running their own scripts often discover during the live hour that the view they need was never built, and the hour is the wrong time to build it.
- Run state, attempts and outcomes appear in one view instead of a log file you grep under time pressure.
- Pause and stop are buttons with defined behaviour rather than a signal you send and then hope about.
- Per-wallet balances and activity are visible together, which is where uneven drain shows up first.
- What it does not change: the thresholds are still yours, and so is the decision to act on one.
Before you trust any view of a run, decide what you would do if it disagreed with the chain. That question is worked through in the question list.
Look at a live run view
The console referenced across this site is a third-party product that runs and reports on multi-wallet activity across Solana venues. Open it if you want to see what a built live view contains before deciding which parts you need to assemble yourself.
See a live run consoleSeparate site, opens in a new tab. Run Desk does not operate it, does not audit its reporting and earns nothing from your use of it.
How this desk works
A page about the live window is only useful if it is correct while you are tired, mid-run, and reading it on a phone. That constraint decides the form as much as the content.
Mechanism before advice
Each page explains what produces a reading before it says what to do about one. If you understand where a number comes from, you can adapt the response when your setup differs from the one described, which it will.
Thresholds are yours, with two costs
The desk never states a correct failure rate, pace or balance floor. It describes what setting a line too tight costs you and what setting it too loose costs you, and leaves the number where it belongs, with the person carrying the run.
Written by a desk, not a persona
Pages are published by The Run Desk with no invented author, no credentials and no case studies. Scope, sourcing and correction handling are set out in about the desk.