Cross-Screen Memory Context Resurfacing Automation: The Same Lesson, Wherever the Decision Happens

A pattern detected somewhere in the product is only useful at the moment someone is actually making the decision it's relevant to — and that moment is almost never spent looking at the analytics screen where the pattern was originally found. An insight that requires a deliberate detour to a separate dashboard competes with everything else on someone's plate and, in practice, mostly loses.

Detecting a pattern happens once, in one engine; making it useful means getting the right two memories, out of everything the engine has ever found, in front of the right screen at the right moment — with wording that fits that screen's context rather than a generic label, and a visual weight that actually reflects how much the system trusts that specific memory right now. Doing that by hand, screen by screen, means someone has to remember to go check the Memory Engine before every decision review, mind-map session, and executive question — exactly the kind of manual step this is meant to remove.

A single function pulls the current memory list, matches it against the screen currently asking — each screen is mapped to a short list of related screens it's allowed to draw from — and returns at most the top two matching memories, sorted by lifecycle stage first and confidence second. Each screen has its own fixed opening phrase for the note it renders — "Com base em memórias semelhantes..." for Mind Map, "Essa decisão já funcionou em contextos semelhantes..." for the Decision Engine, "A memória operacional reforça que..." for the Executive Copilot — so the same underlying memory reads naturally wherever it appears, without needing separate copy written for each place by hand.

The visual weight of that note is never chosen by a person either: a memory already at the "reutilizável" stage renders in green, a memory with high confidence but not yet at that top stage renders in the brand color, and everything else renders in amber — a direct, coded mapping from the memory's own computed status and confidence level, not a manual judgment call made screen by screen. And the moment a note is actually rendered on screen, that exposure is logged — which memory, which module, when — so which insights are genuinely reaching people, and how often, is itself a real, queryable number rather than an assumption.

Confirmed by reading the actual render calls rather than the type definitions alone: three screens genuinely wire this in today — the Mind Map page, the Executive Copilot, and the Decision Engine panel, which renders it twice, once in its compact "best decision now" view and once in its full decision list. The underlying code also defines "War Room" and "Strategic OS" as valid targets a memory could theoretically be matched against, but no actual render call for either currently exists in the product — so this page describes the three screens confirmed wired in, not the two that only exist as a possibility in the type definitions.

A recurring approval-delay memory for a specific client has already reached "consolidado." The next time someone opens that client's mind map, the same memory appears with Mind Map's own phrasing, tagged in the brand color for its confidence level. Later that week, the Decision Engine is asked for its best next move for that same client, and the identical memory resurfaces there too — this time introduced as "this decision has already worked in similar contexts," colored the same way, because the underlying confidence and stage haven't changed — only the room it's being read in has.

A pattern the system has already earned trust in stops being something someone has to remember to go look up — it shows up inside the exact moment a related decision, plan, or question is actually happening, worded for that context. And because every appearance is logged, whether an insight is actually being seen, anywhere, becomes a real number instead of a guess about whether anyone opens the Memory Engine's own dashboard.

A memory that hasn't yet built enough occurrences, contexts, or confidence to clear a reasonable bar simply won't be selected for resurfacing — the lookup is deliberately limited to the top two, sorted by strength, not a firehose of everything ever detected. A team hoping to see this inside War Room specifically will not find it there today — that screen is defined as a valid target in the underlying type system, but no actual render call wires it in yet, so this page does not claim it as a working integration.

A detected pattern that only lives in its own analytics screen depends entirely on someone remembering to go check it before every relevant decision — which, in practice, means it mostly gets missed. Centriu Run instead pushes the two most relevant memories directly into the screen where a related decision, plan, or question is already happening, worded for that specific context and colored by how much the system actually trusts it, with every appearance logged so the reach of an insight is measurable rather than assumed.