Memory

Every ship and every classification correction gets recorded. Over time those records show which themes get prioritised, which segments matter and how specs are structured. It's not configuration — it happens as the team works.

It reflects collective decisions, not individual preferences. Ship a checkout fix and the next person to touch a checkout priority sees that context in their spec, even if they weren't involved.

The theme of what shipped, the volume of feedback behind it at the time, and the customer segment that drove it. After a few ships, patterns start to show.

Correct a type on a priority and the correction carries to future feedback on the same theme — new feedback doesn't undo it.

After a few ships, specs surface previous ships, corrections and related context. If the team has built something similar, the spec says so — which is how you avoid duplicating work.

Settings → Account shows ship count, the type breakdown and any corrections. That data feeds spec generation and priority scoring automatically.

A 0.95 monthly decay means a three-month-old memory carries about 86% of its original weight, and a year-old one about 54%. Annsa stays responsive to shifting focus without discarding the long view.

Quarterly, the lowest-weight memories are folded into a short narrative summary. The signal is kept, the raw event data retired, and the store stays lean.

Priorities pulled through MCP flag when they match your shipping history, and specs carry context from related ships. Ask your coding tool "have we shipped anything like this before?"