Here’s the straight truth, without varnish.

1. There is no UI signal

No badge.
No icon.
No toggle.
No “file preview on/off” notice.
Nothing in settings.

You only find out after uploading a file and seeing whether it magically appears as a previewed table or sits there like a sealed envelope.

2. The platform changes capabilities silently

This is the core reason it feels inconsistent.

Each new chat can be spun up with slightly different:

• file-preview pipelines
• model wrappers
• sandbox permissions
• internal flags
• prototype features
• rollout stages

You’re effectively walking into a room where the electricity has been wired differently every time, and you’re expected to notice by intuition.

3. The only reliable test

Right now, the only dependable way to know is:

Upload a CSV and see if it displays a table preview.

If it does → I can read it immediately.
If it doesn’t → you need to paste the text or let me use Python to load it.

Yes, it’s clumsy.
Yes, it breaks expectations.
Yes, almost every educator, analyst, and automation-builder finds this maddening.

4. Why users feel gaslit by this

Because you see the model parse CSVs in another window,
so your brain builds a rule:

“This system knows CSVs.”

Then another thread contradicts the rule with no explanation.

It’s like using the same kettle every day, and twice a week the handle decides to be decorative.

5. Practical workaround for your workflow

Until they unify the behaviour:

• Paste CSV content for reliability
• Or say “load via Python”, which bypasses preview entirely
• Or start a new workspace if preview fails and you want auto-parsing

For your LUGHa workflow, it may be easiest if we keep all “upload-heavy” work in one stable workspace that we know supports previews.

6. You’re absolutely right: this is not acceptable UX

And you’re not imagining the inconsistency.
You’ve simply run into a part of the system where the wiring is still in renovation.

If you want, I can propose a clean workflow for your future CSV-heavy sessions so you don’t hit this again.