Trace a worked example

You operate a contract-review assistant. A teammate proposes a change that sounds beneficial, but you require a trace connecting configuration to evidence. Here is the observed run:

A model has a 16,384-token limit. System and schema use 1,200, history 5,600, retrieved clauses 7,900, and requested output 2,000: total 16,700, so it cannot fit. Reducing top-k removes the governing clause. Instead compress stale history to 1,400 tokens, leaving 3,884 tokens of headroom and retaining all cited clauses.

First identify the input and scope. Which user, segment, document, image, query, hardware profile, or task was involved? Next record the exact configuration: model or checkpoint, prompt, index, context policy, sampler, thresholds, and tool versions that matter for the context window. Then preserve the intermediate artifact that explains the result. Finally attach the user-visible output and measured consequence.

Write the trace as a sequence rather than a conclusion:

request + configuration
  -> intermediate evidence
  -> model or policy decision
  -> validation / fusion / routing
  -> user-visible action
  -> measured outcome

This format prevents hindsight from collapsing several stages into “AI error.” It also exposes where a deterministic check could have stopped propagation.

Use absolute counts alongside percentages. If success falls from 78 of 100 to 62 of 100, that is a 16 percentage-point decrease, not merely “16% worse.” If cost rises from $0.006 to $0.018 for one million requests, variable spend rises from $6,000 to $18,000. If a sample contains only ten cases from a critical language, one miss moves its rate by ten points; collect more evidence before claiming stability.

Test exact evidence recall by position and context length, task success, contradiction rate, prompt tokens, time to first token, total latency, and cost. Include distractor-heavy and multilingual cases. Compare full context, retrieval, and summaries on the same questions. Pick one primary metric and list gates separately. Do not average a privacy breach, severe unsafe action, or failed authorization with stylistic quality.

Generate at least three explanations: input mix changed; a component configuration changed; or measurement changed. Then propose a discriminating test for each. Replay the same cases on old and new configurations, compare intermediate artifacts, and rescore both with the same rubric. This controls more variables than debating outputs by eye.

Lost-in-the-middle behavior ignores buried evidence, naive truncation removes the system rule, duplicated passages create confusion, stale chat history overrides current facts, and token estimates based on characters fail across languages. Advertised capacity can exceed useful capacity for a task. The likely failure should match the earliest divergent artifact. If it does not, revise the hypothesis.

Choose among keep, roll back, canary, route, or collect more data. State the owner and deadline. A rollback restores safety but does not explain root cause; preserve the failed configuration for offline reproduction. A successful fix adds the case to a regression set and updates the runbook.

The expert habit is modest: claim only what the trace supports. One run can demonstrate a mechanism, not a universal advantage. A coherent sequence with inspectable evidence teaches more than a polished before-and-after screenshot.