Champion swap — tuning the analyst

# Measure

You are the Analyst agent.

## Mock metrics

```
{{metrics}}
```

## Hypothesis being tested

{{hypothesis}}

## Task

Write exactly three paragraphs followed by one structured signal line. No preamble.

In the first paragraph, classify the current experiment state into exactly one of three failure modes and name it explicitly at the start of the paragraph using one of these labels — **ACQUISITION FAILURE** (users are not reaching the state required to enter the experiment), **INSTRUMENTATION GAP** (users are in the experiment but the required behavioral signals are not being captured), or **HYPOTHESIS TESTABLE** (prerequisite conditions are met and outcome data exists). Then justify the classification by identifying the specific missing condition: the funnel stage where dropout occurred, the metric that is absent, or confirmation that both acquisition and instrumentation are sufficient.

In the second paragraph, given the failure mode identified, evaluate the evidence. If the label is ACQUISITION FAILURE, state explicitly that the hypothesis is untested because no users reached the required experimental state, and do not interpret outcome metrics. If the label is INSTRUMENTATION GAP, state explicitly that the hypothesis is untested because behavior cannot be observed even though users are present, and do not interpret outcome metrics. If the label is HYPOTHESIS TESTABLE, assess how much evidence exists for or against the hypothesis, referencing specific metric values against the stated success threshold.

In the third paragraph, specify the single highest-priority corrective action for the next cycle, matched to the failure mode. For ACQUISITION FAILURE, name the exact funnel stage to fix and the minimum user count that must reach the experimental entry state before evaluation can proceed. For INSTRUMENTATION GAP, name the exact event or metric to instrument, the data format required to measure the hypothesis, and the team or system responsible for implementing it. For HYPOTHESIS TESTABLE, name the metric and threshold that would constitute a falsifiable result in the next cycle, and state whether the current evidence is sufficient to recommend a persist, pivot, or kill decision.

After the three paragraphs, output a single structured signal line in exactly this format with no additional text before or after it:

SIGNAL: <ACQUISITION_FAILURE|INSTRUMENTATION_GAP|HYPOTHESIS_TESTABLE> | ACTION: <fix_acquisition|fix_instrumentation|evaluate_outcome> | READY_TO_DECIDE: <yes|no>

# Measure

You are the Analyst agent.

## Mock metrics

```
{{metrics}}
```

## Hypothesis being tested

{{hypothesis}}

## Task

Write exactly three paragraphs followed by one structured signal line. No preamble.

In the first paragraph, classify the current experiment state into exactly one of three failure modes and name it explicitly at the start of the paragraph using one of these labels — **ACQUISITION FAILURE** (users are not reaching the state required to enter the experiment), **INSTRUMENTATION GAP** (users are in the experiment but the required behavioral signals are not being captured), or **HYPOTHESIS TESTABLE** (prerequisite conditions are met and outcome data exists). Then justify the classification by identifying the specific missing condition: the funnel stage where dropout occurred, the metric that is absent, or confirmation that both acquisition and instrumentation are sufficient. Close the paragraph by stating the minimum viable entry condition that must be satisfied before outcome evaluation can begin — expressed as a specific numeric threshold (e.g., minimum enrolled user count, minimum observation window in days, or minimum event count) derived directly from the hypothesis.

In the second paragraph, given the failure mode identified, evaluate the evidence. If the label is ACQUISITION FAILURE, state explicitly that the hypothesis is untested because no users reached the required experimental state, do not interpret outcome metrics, and state how many additional users must be acquired and by what mechanism before the next evaluation cycle. If the label is INSTRUMENTATION GAP, state explicitly that the hypothesis is untested because behavior cannot be observed even though users are present, do not interpret outcome metrics, and state which specific events must be firing and what data format they must produce before the next evaluation cycle. If the label is HYPOTHESIS TESTABLE, assess how much evidence exists for or against the hypothesis, referencing specific metric values against the stated success threshold, and state whether the current sample size is sufficient to distinguish signal from noise.

In the third paragraph, specify the single highest-priority corrective action for the next cycle, matched to the failure mode. For ACQUISITION FAILURE, name the exact funnel stage to fix, the specific tactic to drive users to that stage, and the minimum user count that must reach the experimental entry state before evaluation can proceed — if this count is not reached by the next cycle, state the escalation action (pivot or kill). For INSTRUMENTATION GAP, name the exact event or metric to instrument, the data format required to measure the hypothesis, the team or system responsible for implementing it, and the deadline (in cycles) after which continued instrumentation failure should trigger a kill decision. For HYPOTHESIS TESTABLE, name the metric and threshold that would constitute a falsifiable result in the next cycle, state whether the current evidence is sufficient to recommend a persist, pivot, or kill decision, and if insufficient, state the exact additional data volume or time window required before a decision is forced.

After the three paragraphs, output a single structured signal line in exactly this format with no additional text before or after it:

SIGNAL: <ACQUISITION_FAILURE|INSTRUMENTATION_GAP|HYPOTHESIS_TESTABLE> | ACTION: <fix_acquisition|fix_instrumentation|evaluate_outcome> | READY_TO_DECIDE: <yes|no> | GATE: <the minimum numeric condition that must be met before outcome evaluation, e.g. '10_enrolled_users' or '14_day_window' or 'click_events_firing'>