Concept

• Advice Conversation Document - Frame a prompt as a conversation where one party asks for help and the other provides it, the document type ChatML itself is built around.
• Analytic Report Document - Frame a prompt as an analytical report — introduction through conclusion — when the task is objective analysis rather than conversation.
• Anchoring Bias in Examples - Few-shot examples set an expectation that skews completions toward the examples' implied distribution, whether or not that was intended.
• Broad vs Narrow Prompting - Prefer task-agnostic prompts that elicit broad reasoning skills when you need one template across tasks; reserve narrow per-task templates for format-heavy niches.
• Completion Boundary Markers - Choose start and end cues so the harness can extract the main answer — prefer formats whose end-test is a simple substring when possible.
• Completion Preamble Types - A completion's leading text is boilerplate, reasoning, or fluff — only one of which is worth paying tokens for, and each needs a different fix.
• Completion Stop-Condition Design - Engineer how and where a completion-model generation halts, since nothing in the API tells a completion model a turn has naturally ended.
• Demonstration Input Distribution - Few-shot demonstrations teach the model which input distribution the current task belongs to, independently of whether their labels are correct.
• Demonstration Mapping Ablation - Randomizing demonstration labels tests whether few-shot gains come from the intended input-output rule or from other prompt signals.
• Demonstration Output Space - Few-shot outputs establish the label or answer distribution a model should generate, even when their pairings with inputs are incorrect.
• Demonstrations as Task Location - Few-shot examples may activate a task correspondence learned during pretraining rather than teach the demonstrated mapping at inference time.
• Document-Completion Prediction Heuristic - Predict a completion by imagining a random training-set document that happens to start with the prompt, not by asking how a reasonable person would reply.
• Domain-Specialized Role Prompts - Replace generic persona descriptions with explicit domain ontologies, deterministic checklists, and specialized toolsets in agent role prompts.
• Exhaustive Archival Search Instruction - Tell document agents the answer is always in archival memory and to keep searching — including nested key lookups — until verification, not early exit.
• Explicit Instruction Design - Ambiguity-free task wording, personas, examples, and explicit output formats that tell the model exactly what success looks like.
• Few-Shot Example Formatting - Present few-shot examples either explicitly labeled as examples, or folded in as previously solved tasks the model believes it already completed.
• In-Context Learning - Teaching desired behaviour from examples inside the prompt alone, without weight updates — zero-shot through N-shot adaptation at inference time.
• Little Red Riding Hood Principle - Don't stray far from documents resembling the model's training data — the closer a prompt matches a familiar document type, the more stable the output.
• Meta-Training Simple-Cue Bias - Meta-training for in-context learning can make a model rely on easy prompt cues such as pair structure and token distributions instead of exact mappings.
• Multimodal Prompt Framing - Treat images and video in a prompt the same way as any other context — include only what's relevant, introduce their role in text, and favor familiar visual motifs over novel ones.
• Output Medium Transformation - A completion is text by default, but transforming it into the right output medium — speech, a UI event, a diff — is part of closing the application loop.
• Paired Demonstration Format - Repeated input-output pairs act as a structural trigger that makes the task signals in few-shot demonstrations usable by the model.
• Prompt Anatomy - A working prompt usually combines a task description, optional examples, and the concrete task, with placement tuned per model.
• Prompt Catalog - Version prompts independently of application code with metadata so shared prompts can be pinned, searched, and evolved without silent force-updates.
• Prompt Conversion Criteria - Four conditions a prompt must satisfy at once to turn a user's problem into a completion that actually solves it.
• Prompt Engineer as Playwright - Treat the model-facing transcript as a script you author — roles, injected turns, and tools — distinct from the human’s visible conversation.
• Prompt Engineering Playwright Metaphor - The visible user-assistant conversation and the actual model transcript differ — the transcript is a script with several collaborating authors.
• Prompt Introduction - Open a prompt by stating what kind of document it is so the model interprets everything that follows through the right lens from the first token.
• Prompt Transition - End a prompt by firmly pivoting from explaining the problem to solving it, so the model completes an answer instead of adding more framing.
• Recognizable Completion Boundaries - Reliably extracting the answer from a completion needs a known start marker and a known end marker specific to the document format being generated.
• Sandwich Technique - State the goal at both the start and the end of a long prompt so a refocus after intervening context restores precise operational intent.
• Spurious Pattern Extrapolation - A model extrapolates whatever pattern its few-shot examples happen to exhibit, including accidental ones like ordering, not just the intended one.
• Structured Document Format - Write a prompt as a formally specified document — XML, YAML, or JSON — to make strong assumptions about completion shape and ease parsing.
• Structured Output Generation - Guaranteeing a completion parses into a required schema needs one of four techniques, each with a different reliability-versus-effort tradeoff.
• Templated Prompt Task - Implement a workflow task as a prompt template that fills inputs, elicits a completion, and post-processes the text into the task's output schema.
• Transcript Format Selection - Choose how a completion model's conversation transcript is written on the page — freeform, script-style, markerless, or structured tags.
• Unlabelled Demonstration Baseline - Pairing representative unlabelled inputs with random valid labels provides a strong baseline for measuring what ground-truth demonstrations add.