Adaptive Magazine — the rate-to-learn loop

Type: algorithm Origin repo(s): people-analyst (the People Analytics Toolbox) — the insight-player, player, signals, and elicitation libraries; the immersive media-player engine was vendored from vela Extraction readiness: live — capture → durable store → personalize → re-rank is closed end-to-end on the feed; every insight card is rate-able for usefulness by default Depends on: the Insight Card contract, the durable signal store, the personalization layer, and the sequencing engine Last reviewed: 2026-06-08

What it is

A content system that gets better for each person the more they use it. Every piece of content — an insight card, an article, an immersive sequence — is rate-able for usefulness, and that rating (plus quieter signals like view, skip, expand, and complete) flows into a durable store that re-ranks what comes next. Ask it "what is the next best thing for this person?" and it answers from their accumulated reactions, not a static editorial order.

The shape is match + path + learn. Match the content to the person's persona, goals, and diagnostics. Sequence a path toward their goal. Learn from their reaction to the last thing and feed it back into the ranking. It is built on one engine with two render families — an immersive media player and an enterprise insight-card renderer — over a shared card contract and a shared signal contract.

Who it's for

Two readers at once. The product builder across the portfolio who wants a content system that personalizes per person without standing up a bespoke recommendation stack — they vendor the card, player, and signal contracts and get capture → store → personalize → re-rank for free. And, downstream, the end reader — the analyst getting role-targeted insight cards, the executive getting findings worth their time — who simply experiences a feed that gets better the more they use it because every card is rate-able and every signal counts. The concrete outcome for the builder is a closed loop on the feed today: a person rates something, it lands in a durable store, their affinity profile updates, and the next ranking reflects it. The honesty rail is explicit about the seam — the full persona/goals/classification match step and cross-property identity are documented next units, not shipped features — so the claim is "the learn loop is closed," not "the whole match-path-learn vision is finished."

How the loop closes

• Capture — the universal usefulness rating means no card escapes the loop, and implicit signals (view, skip, expand, complete) are captured alongside the explicit rating.
• Store — signals land in a durable store behind a stable interface, so the history persists across sessions.
• Personalize — an affinity profile is built from the signal history and used to score and re-rank candidate content.
• Path — a sequencer orders the queue by relevance, freshness, and focus-match, with a diversity bonus and a redundancy penalty, so the path advances toward the goal instead of repeating itself.

The loop is closed end-to-end today: a person rates something, the rating is stored durably, the affinity profile updates, and the next ranking reflects it.

Honesty rail — what is built and what is next

What ships: the rating capture, the durable store, the personalization re-rank, and the sequencing path. What is not yet a first-class input is the full persona / goals / classification → content match step — affinity-based match exists, but a single unifying selector that composes match, path, and learn into one "next best piece" answer is the documented next unit, not a shipped feature. Cross-property identity and routing signals back into the source analytics are also still ahead.

Why it is shaped this way

• One contract, vendored not forked. Consuming properties vendor point-in-time copies of the card, player, signal, and elicitation contracts; the toolbox is the source, and the only way to evolve them is to bump the toolbox.
• Content-agnostic. The loop does not care whether the content is a research insight, a comp finding, or an article — the same signal-and-rank machinery applies.
• One engine, two players. The immersive media path (vendored from vela) and the insight-card path share the same sequencing and signal layers, so they never drift.

Related capabilities

• PA Instruments — the elicitation building blocks that capture persona and diagnostics.
• Adaptive learning queue / pool ladder — the sibling exploration-and-promotion shape behind the sequencing intuition.
• Survey respondent interface — the token-gated capture surface signals can originate from.