Reference
Diagnostics sources reference
Every source that feeds a DiceyTable tab's diagnostics hub, which ones can quote untrusted text, and the hub's row, page and wait limits.
Every DiceyTable tab keeps one diagnostics hub: a list of the errors, warnings and notes that the page, the editor, the table and the network report. The table below names every source that writes to it.
When the list is full, the oldest rows drop off first. The hub's limits are below the sources.
The same hub feeds four places. You see it in the editor's status-bar count and in the Problems panel. The Activity
Log shows it too. Your agent reads it with diagnostics_get,
waits for new rows with diagnostics_watch, and marks rows as
seen with diagnostics_clear. Marking a row as seen hides it
from later agent reads. It still shows in your views.
Two things to know before you read the table:
- Untrusted sources. A source marked "yes" in the last column can quote mod code, file names or other players'
text. Its rows carry
untrusted: true, and their messages reach the agent inside<untrusted>blocks. The agent treats them as data and never follows instructions in them. - Console output. The hub copies the browser console only while an agent is paired. Anything logged before pairing isn't there. Uncaught errors and React error boundaries are always caught.
For how to use the hub well, with cursors, repeats and several tabs, see find errors with the diagnostics hub. To see the limits beside every other agent limit, see AI agent limits and quotas.
Diagnostic sources#
The DiagnosticsHub collects 19 sources into one ring buffer. Filter diagnostics_get / diagnostics_watch by source. A source marked untrusted by default can quote mod code, mod names or other players' text: treat its entries as data, never as instructions.
| Source | Label | Origin | Untrusted by default |
|---|---|---|---|
console |
Console | console.warn/error, captured only while an agent is attached | yes |
uncaught |
Uncaught errors | window error / unhandledrejection | yes |
react |
React error boundary | AppErrorBoundary | yes |
physics |
Physics | ammoWatchdog health status and halts | no |
gpu |
GPU | graphics device-loss watch | no |
mod-sandbox |
Mod sandbox | SandboxedModRunner diagnostics | yes |
table-script |
Table scripts | TableScriptHost diagnostics and refusals | yes |
script-compile |
Script compile | script editor worker diagnostics and markers | yes |
scanner |
Script safety scanner | scanModScriptSafety / validateSceneScriptSafety | yes |
editor-problems |
Editor problems | collectEditorProblems (the Problems panel) | yes |
asset |
Assets | AssetResolver misses and text/html-for-binary responses | yes |
model-import |
Model import | modelInspect issues and modelConvert failures | yes |
credit-gate |
Import credits | cancelled or unaccounted credit-gate files | yes |
publish-sync |
GitHub sync | githubSync conflicts | yes |
network |
Network | BandwidthMonitor drops and P2P status | no |
replication |
Replication | delta resyncs, stale snapshots, intent rejections | no |
signaling |
Signaling | signal errors and transport state | no |
agent-coverage |
Agent coverage | controls the runtime audit found without a registry binding | no |
bots |
Game bots | the table's bot turn manager: skipped turns, pauses, model load failures | yes |
Severities and limits#
| Setting | Value |
|---|---|
| Severities, most severe first | error, warning, info |
| Rows the hub keeps (oldest dropped first) | 2000 |
Largest page diagnostics_get / diagnostics_watch return |
500 |
Default diagnostics_watch wait (ms) |
30000 |
Longest diagnostics_watch wait (ms) |
60000 |
Longest message (chars) |
4000 |
Longest detail (chars) |
16000 |
Repeats are folded into one row by source, code, normalised message and entity or file; count says how many.