Home/Modes & context/Modes & context index

Open reproduction tasks for preset and context behaviour

You want to contribute a measurement and do not know which ones are missing.

VerifiedVersion and environmentChecked 2026-11-14

Open reproduction	Needs	Produces
PTC paired test: 50-file batch versus 2-file job	A working PTC environment plus 50 identical files	Three numbers from three runs per group
Same follow-up before and after compaction	A long session that triggers compaction	Counts of correct, does-not-know, and fabricated
Four presets on a long task	All four presets bootable	Same task across four presets: tool surface and round trips
Effect of maxParallelSubCalls under PTC	Independent, parallelisable operations	Time and round trips at 1, 10, and higher
Practical ceiling of minimal in a long session	The minimal preset	After how many turns context starts dropping

State host version, model ID, and preset name.
Paste the raw record table, not just the conclusion.
Include failures and what does not generalise — more valuable than the conclusion.

Every 'to be tested' item here has a method and no numbers. Contribute in the same format so others can reproduce it.

What counts as working

Check every line. All of them pass or the job is not done.

Symptom, cause, what to do — written in the order it actually happens.

Symptom	Cause	What to do
Preconditions unavailable	Missing environment, e.g. no code runtime for PTC	Record the gap itself; that is also useful information

Verified means measured here. Compiled means every line carries a source.