Performance Feedback Loop

1) The dashboard went up; did the content actually improve?

Suppose a campaign redesign is followed by a jump in views. During the same week, tracking changes, a manual page_view is added on top of an automatic page view, and the platform changes what a view means. Reading the dashboard alone can produce a “winner” declaration even though part of the movement came from instrumentation or metric semantics.

In this lesson we build Content Performance Feedback Loop Contract v1. Each content hypothesis is bound to a metric definition, instrumentation version, collection quality, observation window, segment, baseline, decision rule and documented next action.

Central rule: a metric is an observation; without definition, instrumentation version and decision context, it is not editorial truth.

2) COM-014 scope: feedback loop, not repurposing or content modeling

COM-014 owns the performance feedback loop. COM-013 owns Winner Repurposing بدون Copy and COM-015 owns Single-source Content Model. Those outcomes remain separate.

No live analytics configuration, production publishing or definitive causal attribution occurs. The lab is synthetic and offline.

3) Mental model: Hypothesis → Instrumentation → Observation → Decision → Next Action

A closed loop answers five questions: what hypothesis is being tested; which event/metric and instrumentation version produced the observation; in which window and segment; what decision rule existed before reading the result; and what next action follows?

If one of those links is missing, the system is reading a dashboard rather than running a closed feedback loop.

4) What GA4 establishes about events and parameters

Current Google Analytics Measurement Protocol documentation represents events with a name and params in structured payloads and supports recommended and custom events. This FACT shows that collection needs explicit event identity and parameters.

GA4 collection does not prove that a content change caused a conversion increase. Causation and business value remain outside that fact.

5) Duplicate page_view: when instrumentation manufactures performance

Google documents that page views can be sent automatically and cautions that manual page-view sending without disabling the automatic behavior can create duplicate pageviews.

Our contract makes that visible through collection_state. When duplication is suspected or demonstrated, BLOCKED_DUPLICATE_COLLECTION or REMEASURE_COLLECTION_SUSPECT is safer than a winner declaration.

6) Metric definition and semantic version: the label is not enough

Two dashboards can both show views while counting different things. metric_definition records what is counted, and a semantic change must be tied to the decision context.

A current example is YouTube: from 24 August 2026, public views across formats are counted when playback begins, while engaged views and qualified views remain distinct for certain analytics, earnings or eligibility uses. A before/after comparison that ignores that boundary can be STALE_METRIC_SEMANTICS.

7) Instrumentation version: an unversioned collection path is not auditable

instrumentation_version identifies the tag/config/schema that produced the observation. If a content redesign and tracking change happen together, the observed delta cannot simply be attributed to content.

The version must remain connected to event_name, event_params and collection state so a reviewer can reconstruct what was actually measured.

8) Window, segment and baseline: apparently similar comparisons can be unequal

Do not compare a heavily promoted week with an unpromoted week and call the difference “content performance.” observation_window, segment and baseline_ref bind the comparison context.

When windows or segments are not comparable, DISPUTED_WINDOW or BLOCKED_BASELINE is safer than a positive decision.

9) Decision rule and metric hierarchy: define them before reading the result

If the team reads the dashboard first and chooses the threshold afterward, the feedback loop becomes rationalization. decision_rule should exist before interpretation.

If the business objective is qualified lead completion, a rise in views does not automatically offset a material decline in qualified completion. With conflicting signals, the rule can route to REVISE_OR_SEGMENT rather than universal promotion.

10) From observation to action: the loop closes only with a next action

observed_value and uncertainty_note are evidence. decision_state is the bounded interpretation of that evidence, and next_action closes the loop: keep testing, segment, revise, stop or remeasure.

Without an action, the team may reinterpret the same dashboard next month without changing the production contract.

11) Major failure modes and the correct typed state

A baseline with explicit metric semantics, valid instrumentation, healthy collection, comparable window/segment, known baseline and predeclared rule can be LOOP_ACTIONABLE.

• manual and automatic page_view together → BLOCKED_DUPLICATE_COLLECTION
• suspect or incomplete collection → REMEASURE_COLLECTION_SUSPECT
• metric semantics changed without versioning → STALE_METRIC_SEMANTICS
• incomparable window/segment → DISPUTED_WINDOW
• invalid baseline → BLOCKED_BASELINE
• exposure rises while the qualified metric declines → REVISE_OR_SEGMENT

12) Lab and transfer: views rise while qualified leads fall

COM-014-LAB1 contains eight offline fixtures. The learner derives state from visible evidence instead of copying the baseline label.

The Bible's transfer case increases top-of-funnel views while qualified lead completion falls. The learner must state the metric hierarchy, recheck instrumentation and comparison context, then choose keep, segment, revise or stop. A higher exposure metric is not sufficient by itself.

13) Bounded decision, freshness and what this lesson does not prove

This lesson does not prove causation, declare one vendor the best analytics stack or mutate live collection. Its output only determines whether the current evidence package supports a bounded next action or must remain HOLD/REMEASURE/DISPUTED.

GA4 and YouTube are vendor-current sources and should be rechecked by 2026-10-18 or earlier after a material semantic change. Metric hierarchy, decision thresholds and the state vocabulary are MAKNOW DESIGN.