`doctor`¶

gptnt doctor checks benchmark integrity and player configuration. It can also check a run plan, runtime infrastructure, the KTANE mod, or one provider request per configured model.

Usage¶

Command syntax

gptnt doctor [MANIFEST] [--check-mod-load] [--live] [--config-only]
             [--force]

Input	Required	Effect
`MANIFEST`	No	Restricts player checks to the manifest roster and adds roster, generated-plan, and resume checks. Without it, doctor checks every discovered player configuration.

Options¶

Option	Effect
`--config-only`	Checks benchmark integrity, players, image-token settings, and an optional run plan. Skips Redis, KTANE, display, local-service, and machine checks.
`--check-mod-load`	Starts bare KTANE and polls the mod `/health` endpoint. The check can take up to 45 seconds and runs only when binary, mod, and display prerequisites pass.
`--live`	Sends one provider request per configured model after composition and instantiation. This can incur provider charges.
`--force`	Continues despite failed policy and preflight checks. Output produced after a forced integrity check omits release provenance and cannot be submitted.

Select a mode¶

Need	Command
Check all discovered configuration without local services	`gptnt doctor --config-only`
Check one manifest without local services	`gptnt doctor runs/<name>.yaml --config-only`
Check one manifest and the complete machine	`gptnt doctor runs/<name>.yaml`
Confirm that the installed mod responds	`gptnt doctor runs/<name>.yaml --check-mod-load`
Confirm provider endpoints with one request per model	`gptnt doctor runs/<name>.yaml --live`

Report sections¶

The report renders only sections that apply to the selected mode.

Section	What it checks
Benchmark	Release tag, release commit, and protected-content comparison
Players	Configuration composition, object construction, and optional provider request
Image tokens	Whether the player has a usable image-token value
Run plan	Roster coverage, generated specification count, and completion state
Infrastructure	Redis, KTANE binary, mod files, Linux display, experiment-manager port, optional telemetry, and optional mod load
Machine	Machine and disk conditions used by the benchmark

Benchmark section

The values depend on the installed release. A matching checkout has this form:

Expected output

Benchmark
Reference          v2.0.0
Release commit     abc1234
Protected content  matches

Success and failure¶

Successful check

The command exits without a failing row. It does not print a separate final success sentence.

Doctor found problems

A fatal row makes the command exit with an error. Use the row's hint before repeating the command. --force renders the full report and continues after failed checks; a later command can still fail when a required input or service is unavailable.

Without --force, benchmark-integrity and run-roster failures stop generation and execution before they write specifications or start processes. Use installation and doctor troubleshooting for configuration or integrity failures. Use the game or Redis troubleshooting page for a row from that subsystem.

doctor¶