Decision Log

One row per significant decision. Eleven fields. Five minutes to log, ten minutes a week total, one calibration review per quarter. Decision quality scored separately from outcome quality (Annie Duke's rule: don't judge the bet by the result).

Part 1: The Schema

Field	What goes in it
ID	Sequential. D-001, D-002, ...
Date	When the decision was made, not when it was logged.
Decision	One line. What you chose to do (or chose not to do).
Door	One-way (hard to reverse) or two-way (cheap to reverse). One-way doors also get a premortem (Part 5).
Evidence at the time	Only what you knew then. No retroactive polish. "Evidence was thin" is a legal entry.
Confidence	A number, 50-99%. "Pretty sure" cannot be calibrated. 75% can.
Would change my mind	A pre-committed tripwire, written before the outcome. If X happens, I reverse.
Review date	When you will score this row. Two-way doors: 4-8 weeks. One-way doors: one or two quarters.
Outcome	Filled at review. What actually happened, one or two lines.
Quality verdict	Filled at review. One of four: good decision / good outcome, good decision / bad outcome, bad decision / good outcome, bad decision / bad outcome. Judge the decision on the evidence at the time, NOT on the result.
Lesson	One line. What you would reuse or never repeat.

What counts as significant: commits more than a week of team time, hard to reverse, the room disagreed, or you said no to something someone senior wanted. One to three rows a week is the right volume. Do not log routine prioritization.

Part 2: Worked Example Rows

D-014

• Date: [2026-02-10]
• Decision: Sunset the legacy reporting module rather than rebuild it.
• Door: One-way.
• Evidence at the time: Usage down three straight quarters. Two support engineers carrying it. No new customer activated it in six months.
• Confidence: 80%
• Would change my mind: Any top-ten account naming it as a renewal condition before the sunset date.
• Review date: [2026-05-10]
• Outcome: Two mid-tier accounts complained. None churned. Support load dropped as projected.
• Quality verdict: Good decision, good outcome.
• Lesson: The "renewal condition" tripwire was the right kill switch and never fired. Reuse it for every sunset.

D-019

• Date: [2026-03-02]
• Decision: Ship the onboarding redesign before the pricing page test, betting onboarding was the bigger conversion lever.
• Door: Two-way.
• Evidence at the time: Funnel data showed the biggest drop at step two of onboarding. No direct evidence on pricing-page impact. Gut call between two plausible levers.
• Confidence: 60%
• Would change my mind: If the first two weeks post-ship showed no movement in step-two completion.
• Review date: [2026-04-02]
• Outcome: Step-two completion improved, but overall conversion barely moved. The pricing page turned out to be the bigger lever when tested later.
• Quality verdict: Good decision, bad outcome. The evidence pointed at onboarding. The evidence was incomplete, but it was the best available, and the bet was cheap to reverse.
• Lesson: When two levers look close, run the cheaper test first regardless of which looks bigger. Pricing test was one week. Onboarding was five.

D-023

• Date: [2026-03-20]
• Decision: Declined an exec request to fast-track a one-off integration for a single prospect.
• Door: Two-way (could revisit if the deal advanced).
• Evidence at the time: Prospect had not signed a letter of intent. Integration estimated at three weeks. No second customer had asked for it.
• Confidence: 70%
• Would change my mind: A signed LOI, or a second named account asking for the same integration.
• Review date: [2026-05-20]
• Outcome: Deal closed anyway without the integration. No second request materialized.
• Quality verdict: Good decision, good outcome.
• Lesson: "Signed commitment or a second request" is a reusable bar for one-off build asks. Saying no to a senior request is exactly the kind of decision worth logging, because nobody remembers the skipped bets that were right.

Part 3: The Weekly 10-Minute Ritual

Friday afternoon, recurring calendar block.

1. (3 min) Scan the week. Calendar, decisions made in meetings, anything you declined. Pick the one to three that meet the significance bar.
2. (5 min) Log them. Fill the first eight fields only. Outcome, verdict, and lesson stay empty until the review date. Write the evidence field with only what you knew at decision time.
3. (2 min) Check for rows whose review date arrived this week. Fill in outcome, verdict, lesson. Be willing to write "good decision, bad outcome." If every verdict matches its outcome, you are resulting, not reviewing.

If a week has no significant decisions, write nothing. An empty week is data too.

Part 4: The Quarterly Calibration Review

45 minutes, last Friday of the quarter. This is where the judgment training happens.

1. (10 min) Fill every overdue row. Outcome, verdict, lesson for anything past its review date.
2. (15 min) Bucket your confidence numbers. Group resolved rows into bands: 50-65%, 66-80%, 81-95%. For each band, compute the actual hit rate (decisions that worked out / total in band).
3. (10 min) Find the gap. If your 80% band hit at 80%, you are calibrated. If it hit at 55%, you are overconfident by 25 points in that band. Then slice by category (sunsets, sequencing bets, exec pushback, pricing): calibration error is almost never uniform. Find the domain where you are most overconfident.
4. (5 min) Read the lesson column top to bottom. Patterns that appear twice become rules. Write them at the top of the log under "House rules."
5. (5 min) Write one sentence: "Next quarter I will trust my gut less on [domain] and more on [domain]." That sentence is the output of the whole quarter.

Calibration scoring table (copy per quarter):

Confidence band	Decisions resolved	Worked out	Hit rate	Calibration gap
50-65%	[n]	[n]	[%]	[+/- points]
66-80%	[n]	[n]	[%]	[+/- points]
81-95%	[n]	[n]	[%]	[+/- points]

Part 5: Premortem Mini-Template (one-way doors only)

Run before committing to any one-way door. Ten minutes, alone or with the team. Method: Gary Klein's prospective hindsight.

# Premortem: [Decision one-liner]
Date: [date] | Decision ID: [D-0XX]

## The failure story
It is twelve months from now. This decision failed completely.
Write the story of how, in 3-5 sentences. Past tense. Be specific.

[Write the story here. "We sunset the module and our two largest
accounts treated it as a breach of trust during renewal..." Not
"it might not work out."]

## Three most plausible causes
1. [Cause, one line]
2. [Cause, one line]
3. [Cause, one line]

## Disposition (pick one per cause)
- [ ] Changes the decision (do not proceed as planned)
- [ ] Adds a tripwire (copy into the "would change my mind" field)
- [ ] Accepted risk (named, owned, no action)

## Result
Decision stands: [yes / no / modified how]
Tripwires added to log row: [list them]

If the failure story is easy to write and the causes are plausible, that is signal, not pessimism. The premortem exists because forward-looking planning is structurally blind to its own optimism.

---

That's the system. One row per decision, ten minutes a week, one calibration review per quarter. In a year you have a scored record of your own judgment and the data to know where to trust it.