Every candidate action in the Markin Opportunity Feed now carries the signal, segment and prior experiment it descends from. One click to audit.
Every candidate action in Markin now ships with its full provenance attached. Which signal surfaced it, which segment it targets, which hypothesis it descends from, which prior experiment shaped the prior, and which team member last touched it. One click to audit, one click to ship.
This is a small change in the interface and a large change in how growth teams operate against the Opportunity Feed. The rest of this note explains what we shipped, why, and how to use it.
Every card in the Opportunity Feed now exposes a Provenance panel. It shows the causal chain that produced the candidate action, from the raw signal at the bottom to the shipped variant at the top. Each node is clickable. Each node preserves the version of the model, prompt or rule that produced it.
Two failure modes kept appearing in customer reviews. The first: a growth lead ships a variant, it wins, and six weeks later nobody can reconstruct why the system proposed it in the first place. The winner cannot be reused because its lineage is lost.
The second: a variant regresses and there is no clean rollback. Somebody has to reason backwards through the stack to figure out which prompt, model or rule needs to be reverted. That reasoning takes hours. Every one of those hours is trust leaking out of the system.
Provenance closes both. Winners carry their lineage forward into the next hypothesis. Losers can be rolled back at the exact node that produced the regression, with one click.
Three places where the panel earns its rent immediately.
Open the Provenance panel before you approve. If the hypothesis reads as vague or the priors point to a stale experiment, send it back rather than shipping it. The queue is the right layer to enforce hypothesis quality.
A weekly scan of the top ten winners with their provenance expanded is the fastest way to see which parts of the underlying model are compounding. If the same prior keeps showing up under winners, it is worth reinforcing. If the same author keeps appearing, promote their hypothesis style as a template.
When a variant regresses on a guarded metric, the Provenance panel now offers a Revert control at every node. Revert the prompt, revert the model version, revert the rule, or revert the whole action. The system regenerates downstream candidates automatically from the reverted state.
Two extensions are in progress. First, exportable audit packs: a single archive per experiment for finance and legal review, generated on demand. Second, provenance-aware similarity: when a new signal is scored, the system will surface the closest historical opportunity by lineage, not just by segment overlap.
Both ship this quarter. Everything above is live today for all Markin workspaces. For the context on why provenance matters in a continuous operating model, see From campaign calendars to continuous decisioning.
Frequently asked
See it in the product
The same loops this note describes run 24/7 against your customer base. Watch the workspace decide, experiment and execute 1:1.
