The Studio · shared workspace
Make the algorithm legible.
Understand why titles surface, see whether personalization is behaving as expected, and safely test changes before they ship — in one shared workspace.
System pulse
Titles tracked
1,284
across 6 audience segments
Stability index
Healthy
94% of rows within expected volatility
Needs a look
3
rows flagged as a possible real shift
Open simulations
◆ 2
sandbox only — none touch production
Three jobs
"Why did this title surface for this audience?"
Trace any placement back to the behavioral signals behind it.
"Is the system behaving as expected?"
Tell healthy learning from a real problem before you react.
"What happens if I change this?"
Test ranking changes in a sandbox. Stage an experiment — never ship blind.
See it work
"Midnight Atlas" is #3 for Thriller fans · US
Same signal-bar component as the Explain screen — the home teaser can't drift from the real thing. Open the full breakdown →
Explain · the legibility core
Why this title surfaced
Every placement traces back to behavior you can read — not a number you have to trust. Pick a segment and a title to see the signal-by-signal evidence.
Audience segment
Midnight Atlas
Ranked #3 in row "Because you watched documentaries"
What drove this placement
Each signal pushed the title up (right, filled) or held it back (left, outlined). Open any one to see the behavioral evidence.
Plain-language explanation
"Midnight Atlas surfaces at #3 for Thriller fans mainly because completion rate in this segment is well above the row average. It's held back slightly by a higher-than-typical skip rate in the first 30 seconds. Artwork CTR is still provisional — too few impressions to count yet."
The artifact a partner pastes into a thread to align a room in ten seconds.
Health · the honesty layer
Is personalization behaving as expected?
Most movement is the system learning. The hard part is telling that from a real problem — so we show the difference, and name our confidence.
Rank over time — lower is better
Shaded band = expected volatility
Inside the band is normal learning; outside it is worth a closer look. Markers (▮) show known changes.
Verdict
What changed
Correlates movement with known events so volatility has a candidate cause, not a mystery.
Simulate · the signature interaction
Test a ranking change safely
Move the weights, watch the ranking move. Then stage it for review — you can't ship from a sandbox.
Signal weights
Drag a weight; the simulation reorders. Weights total 100.
Predicted impact — computed live
Engagement Δ
projected, sandbox
◆ +2.1%
Diversity index
−0.04 vs baseline — watch this
◆ 0.71
Member-experience risk
rises if diversity collapses
◆ Low–Med
Every number carries the ◆ mark — none of these are live metrics.
The only forward action
There is no "Ship" or "Apply to production" control anywhere, by design. Staging routes the proposal to experimentation review — a person decides whether it ever runs.
PRODUCTION
Frozen baseline · what members see now · read-only
SIMULATION
Live · recomputed from your weights · Δ vs production
The two columns stay side by side at all times — the wall between live and hypothetical is spatial, not just a label, and every simulated value carries the ◆ mark.
Impact · the debrief
What this change would mean
A proposal for review, framed honestly — including the part that got worse.
Staged: "Lift completion weight on the documentaries row"
| What changed | From (production) | To (proposed) |
|---|---|---|
| Completion weight | 30 | ◆ 40 |
| Skip penalty | 15 | ◆ 20 |
| #1 title for segment | The Quiet Quarter | ◆ Midnight Atlas |
| Projected engagement | baseline | ◆ +2.1% (predicted) |
| Diversity index | 0.75 | ◆ 0.71 (watch this) |
Honest framing: the diversity dip is flagged in the summary, not buried. This is a proposal, not a result.
Owner notified · decision tracked · not live
Design outcomes
- Decision latency down — a shared explanation replaces a meeting
- Cross-functional alignment up — one paragraph aligns product, content, and design
- Perceived risk down — simulation makes "what if" cheap and safe