Scout Outcome

scout-outcome

This skill answers one question honestly: what did the last set of moves actually do? Not what they were supposed to do, not what they probably did. What a real, dated, sourced number says happened.

1. What actually goes wrong

The obvious failure is inventing a number. It is not the one that does the most damage. The one that does the damage is grading a run where everything "worked," because a retro like that has stopped doing its job. Every real set of moves produces some noise, some flat lines, and at least one thing not worth doing again. A retro that finds everything worked is not a retro, it's a celebration wearing a retro's shape, and it teaches nothing about what to stop.

2. Where results live

Never estimate a number and never recall one from memory of a prior session. Read it fresh, from where it was actually captured:

• F:\Analytics\snapshots: GA4 exports and other captured analytics. The source of record for traffic, conversion, and signup figures.
• The campaign.md that ran: what was actually attempted, and when, so a metric can be tied to a specific move rather than a vague window of time.
• A prior scout-geo audit, once that skill exists: discoverability numbers over time.

The full list of what to check and where, including how to handle a metric with no clean before/after snapshot, is in references/measurement-checklist.md. Read it before grading anything.

Every figure gets a source: and an on: date. Not "roughly last week," the actual date the file was captured or the number was pulled. A figure without both is not gradable; leave it out rather than approximate it.

3. Grade into three buckets

• Moved: a metric that changed, tied to the move that plausibly caused it, sourced and dated.
• Flat: a move that was made and a metric that was checked, and nothing changed. This is not a failure to record quietly; it's the finding.
• Prune: something to stop doing, justified by a null or negative result recorded above it. Not a vibe, not "this doesn't feel worth it." A specific Flat or a specific absence of a Moved line, cited.

A null result is recorded as a null, never spun. "Flat, but early days" or "flat, but building awareness" are not gradings, they're excuses standing in the place a finding should be. If the honest read is that nothing moved, write that nothing moved.

4. The signature discipline

An outcome run only passes if it names at least one thing to prune. If a full review of the last stretch of moves can't find a single thing worth stopping, the review wasn't rigorous enough. Something in every real batch of moves underperforms; finding it is the point of running this skill at all. Go back through Flat before concluding there's nothing to prune.

5. Write .monkeys/outcome.md

assets/outcome_lint.py checks this file's form: three required sections, each bullet's required fields, and that every on: value is a real YYYY-MM-DD date.

# Outcome — what the last moves actually did

## Moved
- <the metric that moved> — by: <amount> — after: <the move> — source: <where measured> — on: <YYYY-MM-DD>

## Flat
- <a move that produced no measurable change> — measured: <what was checked> — source: <where> — on: <YYYY-MM-DD>

## Prune
- <what to stop doing> — because: <the null or negative result that justifies it>

Run python assets/outcome_lint.py .monkeys/outcome.md before handing off. It checks that every bullet carries the fields a human needs to verify it; it cannot and does not judge whether a move actually caused a metric to change, or whether a prune decision is the right call. Form only, never truth, the same discipline scout-intake's lexicon_lint.py holds.

6. Floors

Never invent a figure. Not a plausible-looking round number, not a "roughly," not a recollection of what a dashboard probably showed. If it wasn't read from a source this run, it doesn't go in.

This skill never publishes. It stages a file for a human, and for RAID and FORTRESS to read. It doesn't post a retro, doesn't announce a metric, doesn't send anything anywhere.

The handle, and the permalink, never enter the pack. Where a moved metric traces back to a specific post or account, capture what it proved (the reach, the date, the number), then discard the identifying link the same way scout-intake discards one after verifying with it.

7. What this doesn't decide

It grades what happened. It does not decide the next play (RAID's job), does not clear a claim for publication (FORTRESS's job), and does not re-run the campaign that produced these numbers. Hand outcome.md to whoever plans the next round.

A worked run, including a null result that leads to a real prune, is in ../../examples/scout-outcome-worked-example.md.