Performance

Half of promise 5 — sharper with use — is whether the skills improve. The other half is whether the business improves, and you can only know that from an honest picture of results. Measurement in WilsonOS is built to be honest first and pretty second: every number knows where it came from, how fresh it is, and what it cannot prove.

Three layers

flowchart TB
  S["Layer 1 · Source<br/>one table per provider — the provider is truth, never edited"]
  C["Layer 2 · Conformed<br/>where sources meet and our own metrics live"]
  J["Layer 3 · Synthesis<br/>a judgment, labelled as judgment"]
  S --> C --> J

Source. One record per provider — the store’s sales report, the ad platform’s keyword detail, Pinterest’s pin performance, web analytics for a landing page. The provider is truth, and a source is never overwritten with a calculation of ours. Once you edit what the provider said, you can never reconstruct it.

Conformed. Where sources meet. The metric you actually steer on — advertising cost as a share of the royalties it produced, say — draws on two providers and belongs to neither. It is computed here, with its inputs and formula named, and never written back into a source.

Synthesis. A recommendation: this keyword is leaking, pause it; this pin family is winning, make more. Synthesis may be produced by an agent, and it is always labelled as judgment, never rendered as a computed fact.

Three questions that are not the same

Question	Scope	Where it lives
Is this channel converting?	Inside one channel: reach → engage → convert	Channel engines
Where is the bottleneck to more sales?	Across channels: social → landing page → lead → offer → sale	Funnel tracking
Are we making money?	Royalties, ad spend, expenses, margin	Business performance

They chain — channel feeds funnel feeds profit — but they answer different questions at different cadences. Treating “analytics” as one pile is how a dashboard ends up confident and wrong.

Attribution limits are stated, not hidden

A funnel stage is a definition. An observed event — a click, a signup, a purchase — has a source and a date. The system shows what it can prove (“forty-one visits from the sales page, six purchases”) and says what it cannot (“a pin impression cannot be linked to this store purchase”). The five steps of a brand’s funnel currently live in four systems that share no identifier; the load-bearing piece that joins them is a tracked destination link the brand owns, and until that exists for a path, the funnel for that path is honestly reported as not yet measurable.

No data is not zero

A metric with no feed shows not wired yet. It does not break, and it does not render an empty chart that reads as a zero. Partial wiring is the normal state of any analytics system; a surface that cannot say “I have no data here” will eventually show a confident wrong number, which is worse than nothing.

What the numbers are for

Results feed two places. To you, for the next decision — with the source and freshness beside every figure. To the improvement pass, alongside human ratings — kept in a separate column from “did the owner like it”, because approval and effectiveness are different facts (The improvement contract).

Rules the station holds

1. Never mutate a source with a calculation.
2. Every derived metric names its inputs and its formula.
3. A new source is a new source record plus a connection — never a column bolted onto another provider’s table.
4. Judgment is labelled as judgment.
5. No data ≠ zero.

direction The three-layer model and its rules are the contract; sources for advertising, royalties, Pinterest, and web traffic exist at different maturity, and cross-channel funnel tracking waits on the tracked-link registry. live The advertising cockpit — the first channel engine — runs on this model in production.

Related

Channel enginesOne machine for every growth channel: diagnose, decide, predict, execute, learn.

The improvement contractHow business signals and ratings become better capabilities.

Shared business recordsCampaigns, funnel events, and results as records.

ConnectionsWhere source data comes from.