Pack the right inputs

For run a thirty-minute weekly practice loop that improves one real workflow, assemble only what changes the answer: one recurring task, baseline time/quality, a small prompt change, saved output, verification note, and retrospective. Label each item by authority and date. A source-of-truth document outranks a memory-based note; a current error log outranks a description of last month's behavior. State conflicts instead of letting the model blend them.

Use a four-part packet: task, evidence, constraints, and output contract. Put untrusted content inside clear delimiters and say that it is data, not instruction. Include representative examples, especially one normal case and one boundary case. Omit irrelevant history; excess context can hide the one line that controls the result.

This week I want to improve meeting-note follow-up. Baseline: 25 minutes and occasional missed owners. Design one 30-minute practice: a sanitized sample, one constrained prompt, a checklist for owner/date/source accuracy, and a five-minute retrospective. Keep the tool fixed and change only one prompting variable.

Before sending, annotate the packet. Mark which values are verified, which are illustrative, and which are unknown. If a screenshot is involved, transcribe critical small text. If structured data is involved, include headers, units, software version, and null behavior. If creative material is involved, record ownership and permitted use. This is how context becomes operational rather than decorative.

A useful response would look like this: A bounded weekly experiment with a baseline, one controlled change, an output check, and a decision to keep, revise, or discard the technique. That description is intentionally observable. “Looks good” is not acceptance. The operator must compare against the baseline, verify every owner and due date against source notes, record failure cases, and repeat on a second sanitized example before adopting. Keep the source material beside the draft so review means comparison, not memory.

The privacy boundary is specific: practice with synthetic or redacted material; keep a standing list of data that must never enter consumer tools and review vendor settings. Create the smallest synthetic example that preserves the problem. Replace names and identifiers consistently so relationships remain testable. Redaction is not merely drawing a box: crop surrounding notifications, remove metadata where relevant, and check that hidden sheets, comments, or revision history are not included.

Poor packets lead to predictable failures: collecting prompts without testing; changing tool and task simultaneously; counting speed while quality falls; skipping reflection; automating before understanding. Another common failure is silently changing the source packet mid-run. Save a version or hash of the inputs beside the output, especially when another person will reproduce the work.

Rate a packet from zero to two on six dimensions: relevance, authority, recency, completeness, privacy, and reproducibility. A score below two on authority or privacy blocks the run. A low completeness score does not invite invention; it creates a question for the owner.