Modes & context

Which of the four modes to use, when to compact the context, and how to phrase an instruction that does not need a redo.

Technical group6 nodesUntested

The same task in a different mode changes both the token count and the result. This group covers the differences and the phrasing; the numbers live in the experiment group until they are actually measured.

Working with the modes Untested

What each mode does and which tasks fit it; whether it saves tokens is left to the lab nodes.

Required：Host version · Mode · Task type · How switched · Observed result

First topics 2

PTC vs standard Untested

A controlled standard-vs-PTC comparison on token, time and turns, with raw logs.

Required：Question · Control groups · Environment & versions · Raw data · Where it does not generalize

First topics 2

Context & compaction Untested

What fills the context, when it gets compacted, and how to investigate when the model starts forgetting.

Required：Host version · Context breakdown · Compaction trigger · Symptom · Diagnostic steps

First topics 2

Prompt structure & placeholders Untested

How recipes write full prompts, use 【】 placeholders, and make output verifiable.

Required：Use case · Version & environment · Full prompt · Placeholder notes · Acceptance checklist

First topics 2

Glossary Untested

Plain definitions for easily confused terms (modes, compaction, version gate, profile), each with a source and a confusion note.

Required：Term · Definition · Source · Commonly confused with · Verification status

First topics 2

Modes & context index Untested

Node notes, mode-related term links and the open reproduction list.

Required：Node count · Last updated · Open claims · Status labels

First topics 2

How missing content gets filled

If you have a run record, an error text or a working recipe, post it in the matching board. If you are unsure about the format, the submission note lists the fields.

Claim one DSH directory