Close the loop on outcomes

What you'll get

An answer to the question most advice quietly avoids: did it actually work? The outcome ledger ties a recommendation to the single metric it was meant to move, then comes back after a set window and holds the promise up against the realized number. Run that for a few months and you have a win-rate you can audit, in place of a feeling about whether the advice was any good.

The steps

1. Record the bet when you accept a recommendation.

/marg:outcomes record: pricing team recommended moving the Pro tier to 79. Metric: MRR. Baseline: 42k today. Expect up. Review in 60 days.

An entry needs six things: the recommendation, the one target metric, its baseline value and date, the expected direction, how it will be measured, and the review date. Tracking is opt-in, so nothing is recorded unless you ask, though Marg will offer when a recommendation is high-stakes enough to be worth following.

2. Reconcile when the window closes.

/marg:outcomes reconcile due

For every entry past its review date, Marg reads the realized metric from the tool it was registered against, and a live reading grades HIGH. If that tool is unavailable, you supply the number and it grades accordingly. The entry then resolves one of three ways: won if the metric moved materially in the expected direction, lost if it moved against you or barely moved at all, and inconclusive when the signal was confounded or the window was simply too short to tell.

3. Respect the inconclusive verdict. Inconclusive is a real outcome, not a failure to be massaged into a win. A quarter where three changes shipped at once produces confounded metrics, and an honest ledger says exactly that rather than crediting the change you happen to like. No number is ever invented to close an entry, full stop, because the moment the ledger fudges once, it is worthless for every decision after.

4. Read the rollup.

/marg:outcomes

With no arguments you get the whole picture: open entries, overdue reviews, the won, lost, and inconclusive counts, the win-rate, and how many of the realized numbers came from live data rather than your own report.

What comes back

Each entry keeps an append-only history, so reconciling adds a new row instead of overwriting the prediction. The original call stays visible alongside the result, which is what keeps everyone honest, Marg included, because a forecast you can no longer see is a forecast you can quietly rewrite.

Variations

• Founder-supplied metrics: no connected tools required. You can run the entire loop on numbers you type yourself, and the ledger simply grades them as self-reported rather than measured.
• Track your own decisions, not just Marg's: the ledger does not care where a recommendation came from. Record the bets from your Monday meeting the same way, and the win-rate it produces is every bit as revealing.

If something goes wrong

• The review date arrives mid-chaos: reconcile it as inconclusive with a note rather than skipping it. A skipped entry teaches nothing, while a dated inconclusive explains itself when you read the year back.
• The metric moved, but for an obvious other reason: say so as you reconcile. The verdict can be inconclusive with the confounder named, which is worth far more than a win you cannot actually attribute.
• You forgot to record at the time: record it late with the original baseline if you still have it. A late entry on an honest baseline still closes the loop.