Why does LongChat Guard say “Unable to assess”?
Start with the correct status
Not calibrated
There is no usable, user-confirmed historical conversation-length failure reference (R). Reading an ordinary long chat cannot safely create one.
Unable to assess
The current chat's measured history, sequence, or environment cannot yet be proven complete. This is different from knowing its remaining capacity.
Automatic reading stopped
The user cancelled, the foreground tab changed, or the browser could not finish the scroll and verification. Incomplete results are rejected rather than saved.
What to do next — an evidence-based checklist
| You see | Try this | Do not assume |
|---|---|---|
| New chat says “Unable to assess” | Continue naturally from its first turn; if coverage stays unknown, request “Read full current chat” from the menu. | A few visible messages prove the complete sequence. |
| Older chat cannot be fully measured | Open that exact chat, keep its browser tab in front, start a full automatic read and let both directions plus verification finish. | Reaching the top alone means all older messages were captured. |
| “Not calibrated” despite a successful current read | Use a different historical chat only if you personally confirmed it reached a conversation-length limit; then calibrate it. | An ordinary long conversation is a proven maximum. |
| Auto-read cancels or the browser warns about debugging | Keep the ChatGPT tab focused, avoid scrolling or switching tabs, then retry or choose Cancel. Review the privacy permission disclosure. | The strong debugger permission reveals official token allowance. |
If ChatGPT has already reached a hard conversation limit
A hard limit can prevent new messages in that same thread, so “ask the old conversation for one last summary” may no longer work. Use any existing user-visible information and source files you can still access, gather the current objective, confirmed decisions, constraints and unfinished work, and begin a fresh thread. The extension does not bypass ChatGPT's limit or automatically transfer a conversation.
For an actionable example, read how to prepare a compact handoff. If you use ChatGPT Projects, OpenAI's Projects documentation explains that project context and memory vary with plan and settings. Do not assume a new project chat can reconstruct every old turn.
Why does automatic reading require debugger access?
Only when you start a full read or calibration, LongChat Guard temporarily uses the browser's powerful debugger permission to send mouse-wheel input to the active ChatGPT tab. The original reader then checks message continuity. Normal passive risk monitoring does not attach a debugger; the extension detaches at the end, on cancellation or error. The browser permission itself is broader than this self-imposed restriction. Read the exact disclosure.
Illustrative scenario (not user telemetry)
Frequently asked questions
Does “Unable to assess” mean ChatGPT is out of tokens?
No. It describes evidence quality for the extension's local measurement, not an official token or capacity reading.
Will recalibrating solve every unknown status?
No. The historical reference and current chat coverage are separate. Do not overwrite a reliable historical reference with an unverified sample.
What if automatic reading repeatedly fails?
Preserve your local data, verify you're in the correct foreground ChatGPT tab, and retry. Report the website/browser version and sanitized steps in GitHub Issues—never paste private conversation text.
Next steps
LongChat Guard is an independent open-source extension, not affiliated with OpenAI. This page is guidance, not a claim of a universal ChatGPT length limit.