Also known as: Cluster Hierarchy V2, Hierarchical clustering
Three-level segment clustering — Category → Subcategory → Cluster — built from production trace segments with no manual tagging.
Moda's cluster hierarchy groups trace segments into a three-level taxonomy: broad categories, mid-level subcategories, and tight clusters. New segments are assigned into the existing hierarchy as they arrive, so the taxonomy stays current between full clustering runs, and every node carries a plain-language label so the map reads as user goals rather than cluster IDs.
RelatedIntent discovery, Segment, Trace analytics
Also known as: Context compaction, Memory compaction, Context summarization
The harness step that summarizes older turns, completed workflow phases, and bulky tool outputs into a compact digest, so the context window carries decisions and constraints instead of raw transcript.
Compaction in an agent harness is the step that rewrites accumulated context — older turns, completed workflow phases, verbose tool outputs — into a shorter digest that preserves decisions, constraints, and open questions while dropping the raw transcript. Agents fill their own windows by construction, and model performance degrades as input grows (context rot), so long-running agents compact instead of appending forever. The compaction prompt is a harness artifact like any other: tuned well, it keeps a forty-turn session sharp; tuned badly, it silently summarizes away the constraint from turn three and produces context loss while every call still returns 200. Moda treats compaction as a first-class harness component — when it detects context loss across production traces, it attributes the failure to the component responsible, compaction prompt included, and verifies the candidate fix against replayed traffic before it ships.
RelatedAgent harness, Context loss, Context rot, Behavioral failure, Moda
Read the guide: Context rot →
A behavioral failure where the agent forgets earlier turns, contradicts itself, or asks the user to re-supply known information.
Context loss happens when an AI agent loses track of information from earlier in the trace — forgetting user preferences, repeating questions already answered, or contradicting prior statements. It is common in long sessions, multi-step workflows, and traces that exceed the model's effective attention window. Context loss is a behavioral failure: every individual call may succeed, but the trace as a whole degrades.
RelatedBehavioral failure, Frustration root cause, Context rot
Also known as: LLM context rot, AI context rot
The degradation of an LLM's performance as its input context grows longer: recall, reasoning, and instruction-following decline unevenly as tokens accumulate, even well below the advertised window limit.
Context rot is the model-level mechanism named by Chroma's 2025 technical report — which measured it across 18 models — and adopted by Anthropic's engineering guidance: as the number of tokens in the context window increases, the model's ability to accurately recall information from that context decreases. It happens because transformer attention relates every token to every other token, so a growing window stretches the attention budget thin. In a production agent it surfaces as context loss — forgotten turns, dropped constraints, re-asked questions — and no model choice removes it, since every model tested shows the effect. Every mitigation that works is a change to the agent harness: compaction, trimmed tool outputs, explicit memory, load-bearing instructions positioned at the window edges, and regression evals on mid-context misses, each verified against production traces. Moda detects the symptom as context loss across every production trace, attributes each detection to the harness component responsible, and verifies each candidate fix against replayed traffic before it ships.
RelatedCompaction, Context loss, Agent harness, Behavioral failure
Read the guide: Context rot →
Also known as: Continuous learning, Learning loop, Harness-layer learning
Harness-layer improvement from production signal: updating prompts, tools, workflows, retrieval, memory, and evals so the learnings live outside the model weights and apply across whichever model the harness mounts.
The agent harness is where continual learning (sometimes called continuous learning) actually happens in production: revised system prompts, retightened tool schemas, restructured workflows, expanded retrieval indices, new skills, updated eval sets, and per-user memory. The term describes closing the loop between what an agent does in production and how the agent is built. Model fine-tunes are an option but rarely the first step — they are heavier, harder to inspect, and bound to a specific model. Keeping learnings outside the model weights makes them portable across models, inspectable, reversible, per-user adaptable, and continuously updatable without retraining. Moda is not a continual learning system; it is a harness engineering platform that does the work this loop needs — turning production traces into verified improvements to the harness.
RelatedAgent harness, Self-improving agent, Intent discovery, Behavioral failure, Moda
Read the guide: Continual learning →