THE MACHINERY OF DECISION ARCHITECTURE
Practical Edition
Drills, Templates, Audits, and Weekly Protocols for Operators Who Already Understand the Mechanism
This is the companion to THE MACHINERY OF DECISION ARCHITECTURE.
The pointing piece described how decisions actually get made. The structural properties of the environment that determine, before any choosing happens, what gets chosen. Bounded rationality. Satisficing. Defaults. Reversibility. Convexity. Cognitive depletion. Recognition-primed action.
This piece does not re-argue any of that. If the mechanism is unclear, read the pointing piece first.
What follows are the protocols. The drills the operator runs on the architecture itself. The audit prompts that surface the defaults nobody set on purpose. The templates that turn theory into Tuesday morning behavior.
Use only what fits the operation. Skip what doesn’t. The mechanism is universal. The implementation is specific.
THE WEEKLY OPERATING PROTOCOL
The architecture is not maintained by reading about it. It is maintained by a recurring cycle of audits, decisions, and corrections. The minimum viable cadence is weekly. Below it, drift wins.
THE WEEKLY DECISION ARCHITECTURE LOOP
MONDAY THURSDAY SUNDAY
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ │ │ │ │ │
│ TYPE 1 / 2 │ → │ PRE-MORTEM │ → │ DEFAULT │
│ TRIAGE │ │ (Type 1's │ │ AUDIT │
│ │ │ from Mon) │ │ │
│ Sort the │ │ │ │ What ran on │
│ week's │ │ Failure- │ │ rails this │
│ decisions. │ │ hunt. 30% │ │ week that │
│ Route by │ │ more modes │ │ shouldn't │
│ reversi- │ │ surfaced │ │ have? │
│ bility. │ │ on average. │ │ │
│ │ │ │ │ │
└──────────────┘ └──────────────┘ └──────────────┘
│
▼
┌──────────────┐
│ │
│ ARCHITECTURE│
│ PATCH │
│ │
│ Adjust one │
│ default per │
│ week. Never │
│ more. │
│ │
└──────────────┘
The full week takes under three hours total. Forty minutes Monday. Sixty minutes Thursday. Ninety minutes Sunday. The compounding return is not the time spent. It is the architecture decay prevented.
DRILL ONE: THE TYPE 1 / TYPE 2 TRIAGE
What it is
A ten-minute Monday-morning sort of every meaningful decision the operator expects to make in the coming week. Each decision lands in one of two columns. The column determines the process.
Why it matters
The default failure mode of growing organizations is applying Type 1 process to Type 2 decisions. Every reversible experiment gets routed through committees. Speed dies invisibly while everyone feels prudent.
The triage forces the question: walk through this door, can you walk back? If yes, it is Type 2 and it does not deserve Type 1 weight. If no, it is Type 1 and it deserves Type 1 weight.
Most decisions are Type 2. Operators who run the triage discover this in the first session. The discovery is the value.
The template
WEEK OF: ___________________
DECISIONS THIS WEEK
─────────────────────────────────────────────────────────────
DECISION REVERSIBLE? ROUTING
─────────────────────────────────────────────────────────────
1. ____________________ [ ] Yes [ ] No [ ] Ship now
[ ] Pre-mortem
2. ____________________ [ ] Yes [ ] No [ ] Ship now
[ ] Pre-mortem
3. ____________________ [ ] Yes [ ] No [ ] Ship now
[ ] Pre-mortem
(continue for every decision the operator will touch)
─────────────────────────────────────────────────────────────
TYPE 2 (REVERSIBLE) → DELEGATE OR DECIDE FAST. 70% INFORMATION.
TYPE 1 (IRREVERSIBLE) → SLOW. PRE-MORTEM. CONSULT.
The classification questions
Use these to settle ambiguous cases:
- If this decision turns out badly in two weeks, what is the cost to undo it?
- If the cost is one conversation and a written reversal, it is Type 2.
- If the cost is severance, legal, lost relationships, sunk capital that cannot be recovered, or public reputation damage, it is Type 1.
- If the cost is in between, treat it as Type 2 and pre-mortem it. Better to over-process a Type 2 than under-process a Type 1.
The trap
Most operators classify too many decisions as Type 1. Status, ego, and loss aversion all push toward overweighting irreversibility. The corrective bias: assume Type 2 until proven Type 1. The proof must be specific and concrete. Vague gravity (“this matters a lot”) is not proof of irreversibility.
DRILL TWO: THE 70% RULE
What it is
A working rule for Type 2 decisions. When the team says “we need more data,” the operator asks: would the data take us from 70% confidence to 90% confidence, and what is the cost of the delay?
The script
For every Type 2 decision that triggers a request for more analysis:
THE 70% AUDIT (Type 2 Decisions Only)
1. CURRENT CONFIDENCE
If we decided right now, what is our probability of being correct?
____________%
2. INFORMATION GAP
What specific data would the additional analysis surface?
_____________________________________________________________
3. CONFIDENCE DELTA
If that data confirmed our current direction, how much would
confidence rise? From ____% to ____%
4. DELAY COST
What is the cost of waiting for that data?
- Days to acquire: ______
- Opportunities expiring during wait: _________________________
- Optionality lost: ___________________________________________
5. ROUTING DECISION
[ ] Ship now (delay cost > expected confidence gain)
[ ] Wait (delay cost < expected confidence gain AND
decision is genuinely irreversible)
What changes
The conversation shifts from “do we have enough data?” to “is the marginal data worth the marginal time?” In most Type 2 cases, the answer is no. The architecture has reframed the question, and the new framing produces faster correct decisions.
The trap
Do not apply this to Type 1 decisions. The 70% rule is reversibility-conditional. For one-way doors, more data is often the correct answer regardless of delay. The rule does not abolish caution. It re-routes caution to where caution earns its cost.
DRILL THREE: THE DEFAULT AUDIT
What it is
A weekly inventory of every recurring outcome the organization produces without anyone making an active decision. The defaults the operator never set on purpose.
How to run it
Sunday afternoon. Sixty to ninety minutes alone. Walk through these prompts in writing. Do not delegate this. The operator who delegates the default audit is asking the people who run on the defaults to evaluate the defaults.
The audit prompts
DEFAULT AUDIT | WEEK OF ____________________
CATEGORY ONE: Customer-facing defaults
1. When a customer complains and no one has been assigned the
ticket, what happens?
_____________________________________________________________
2. When a refund is requested under $X, what is the default
response and who has authority to override?
_____________________________________________________________
3. When a regular customer goes silent for 30 days, what
automatically happens?
_____________________________________________________________
CATEGORY TWO: Operational defaults
4. When a vendor invoice arrives and no one disputes it, what
happens? Who would know if it was wrong?
_____________________________________________________________
5. When a shift is short by one person, what is the default
coverage path? Who covers? Who pays?
_____________________________________________________________
6. When a piece of equipment fails outside business hours, what
is the default response?
_____________________________________________________________
CATEGORY THREE: People defaults
7. When an employee is consistently 5-10 minutes late, what
happens? At what point does the default change?
_____________________________________________________________
8. When a high performer asks for a raise, what is the default
response? When a low performer asks?
_____________________________________________________________
9. When a meeting has no agenda, what happens?
_____________________________________________________________
CATEGORY FOUR: Decision defaults
10. When a Type 2 decision is brought to the operator that
should have been made by the GM, what is the default
response? (The wrong default: solve it. The right default:
return it.)
____________________________________________________________
11. When two managers disagree on a Type 2 decision, what is
the default tiebreaker?
____________________________________________________________
─────────────────────────────────────────────────────────────
DEFAULTS THAT WILL CHANGE THIS WEEK: __________________________
(Limit to one. Maybe two. The whole point is that defaults
have momentum. Changing several at once produces no real change
in any of them.)
The frequency cap
One default change per week. Maximum. The temptation will be to find six defaults that are wrong and change them all on Monday. This produces nothing because the staff cannot absorb six default changes simultaneously. The architecture-level patch is high leverage precisely because it is rare and specific.
The trap
Auditing for defaults to remove without auditing for defaults to add. The most powerful default change is often putting a default where there was none. If “no one knows what to do when X happens,” the default is currently chaos plus fastest-talker. Setting a written default is the architectural fix.
DRILL FOUR: THE PRE-MORTEM SCRIPT
What it is
A sixty-minute group exercise run before any Type 1 decision. Not a risk register. Not a brainstorm of what might go wrong. A specific cognitive maneuver: imagine the decision was made, time has passed, the project failed catastrophically. Now write the autopsy.
The script
PRE-MORTEM SESSION | DECISION: ___________________________
PARTICIPANTS: 5-7 people. Mix of senior and junior. Include
the person who proposed the decision. Include at least one
person who will be affected operationally.
DURATION: 60 minutes total.
─────────────────────────────────────────────────────────────
OPENING (3 minutes)
Read aloud: "It is six months from today. The decision we
are about to make has failed. Catastrophically. Everyone
knows it failed. The project is being unwound, the strategy
is being reversed, the hire is being terminated, the deal
is being undone. We are sitting in this room writing the
story of why."
INDIVIDUAL WRITING (15 minutes)
Each participant writes silently. No discussion. The prompt
is "Why did this fail?" They write specific failure modes,
causal chains, decision points where the failure compounded.
Aim for ten failure modes per person, not three.
Anonymous submission. Write on cards. Pass to facilitator.
ROUND-ROBIN READING (15 minutes)
Facilitator reads each card aloud. No commentary. No "yes
but." No defending the decision. The room hears every
failure mode without anyone needing to own it.
CLUSTERING (10 minutes)
Group failure modes by category. Common patterns will emerge:
- Wrong assumptions about market or counterparty
- Underestimated time or capital required
- Failure to anticipate a specific stakeholder reaction
- Operational failure in execution
- Architectural failure (wrong people, wrong incentives)
DELIBERATION (12 minutes)
For the top 3 failure clusters:
- Is this preventable? How?
- What signal would warn us this is happening?
- What is the trip-wire we will set right now?
CLOSING (5 minutes)
Document:
- Top 3 failure modes
- Mitigation for each
- Trip-wires (specific signals, specific dates)
- GO / NO-GO / GO WITH MITIGATION
What it produces
Two outputs. One is the list of failure modes and trip-wires. That is the obvious deliverable. The second is a quieter and more important effect: the room has now imagined the decision failing. Loyalty to the original plan has been weakened just enough that real concerns can surface in subsequent meetings without feeling like betrayal.
The trap
Running pre-mortems on Type 2 decisions. Each pre-mortem costs sixty minutes of senior time across five to seven people. That is five to seven labor-hours per session. For a Type 2 decision that can be reversed in a week, the pre-mortem cost exceeds the failure cost. The pre-mortem is reserved for one-way doors.
DRILL FIVE: THE CALENDAR ARCHITECTURE
What it is
A redesign of the operator’s week so that the most consequential decisions land at the highest-energy slots. Not because energy can be willed. Because the architecture of the schedule pre-determines the cognitive resources available.
The diagnostic
CALENDAR DIAGNOSTIC | Week of _____________
Step 1: List every meeting and decision touchpoint in chrono
order for the past 5 days.
Day Time Item Decision quality req'd
─── ───── ────────────────────────── ──────────────────────
___ _____ __________________________ ______________________
___ _____ __________________________ ______________________
(...)
Step 2: For each item rated HIGH on quality required, mark
where it landed in that day's sequence.
- Position 1-2 of the day: ____ items
- Position 3-5 of the day: ____ items
- Position 6+: ____ items ← red flag
Step 3: Architecture failures.
- High-stakes items at position 6+ this week: ____
- High-stakes items at position 1-2 this week: ____
- Ratio (1-2 / 6+): ____
Step 4: Patch.
Move the next ___ high-stakes recurring items to position
1-2 of their respective days. Reschedule the low-stakes
meetings that currently occupy those slots to position 6+.
The principle
Cognitive depletion is metabolic. The fifteenth decision of the day is not the same decision the operator would have made first thing in the morning. The architecture either honors this or ignores it. Honoring it is free. Ignoring it costs decisions monotonically without anyone seeing the bill.
The shortcut
If the diagnostic feels like too much, run a one-question version: where did the most important decision of last week land in that day’s sequence? If the answer is anywhere after position three, the calendar is broken. Fix that one slot. Run the question again next week.
DRILL SIX: THE DECISION RIGHTS MAP
What it is
A written document specifying who has decision rights for which decisions, ordered by reversibility and stakes. Most organizations have implicit decision rights, drift, and resentment. Explicit rights generate speed.
The template
DECISION RIGHTS MAP | Updated ____________
DECISION STAKES OWNER REVIEW?
────────────────────────────── ───────── ────────── ────────
Hire / fire FOH staff <$50k GM None
Hire / fire BOH staff <$50k GM None
Schedule changes (1 week) <$5k Shift Lead None
Schedule changes (4+ weeks) <$25k GM DM weekly
Menu pricing (existing items) <$20k/mo GM DM monthly
Menu pricing (new items) <$50k DM Owner Q1
Vendor switch (recurring <$5k) <$60k/yr GM DM monthly
Vendor switch (recurring >$5k) >$60k/yr DM Owner Q1
Marketing spend per campaign <$2k GM None
Marketing spend per campaign <$10k DM Owner Q1
Capital expenditure <$5k GM DM monthly
Capital expenditure >$5k DM Owner
Lease decisions Any Owner only Legal review
Equity issuance Any Owner only Board
(Customize categories. The categories matter less than the
specificity. Each row removes a small ambiguity that was
costing decision speed every week it existed.)
The push-down principle
Authority should sit at the lowest competent level. Every level above that adds delay without adding judgment. The test: if the GM has been making this kind of decision correctly for six months, the formal authority should follow the demonstrated competence. If the formal authority has not followed, the architecture is wasting the GM’s judgment.
The escalation path
Every cell in the map needs an escalation path. “What if the GM is wrong?” is not a reason to remove the GM’s authority. It is a reason to specify when escalation triggers. Trip-wires, not gates. The gate prevents action. The trip-wire allows action and surfaces signal that the action may need correction.
The trap
Decision rights maps that are written and never enforced. The map is not the architecture. The behavior the map produces is the architecture. If the map says “GM decides hires under $50k” but the GM still asks the DM about every hire, the architecture is the asking pattern, not the document. The fix is upstream: stop answering. Return the question. The architecture corrects through the redirected behavior.
DRILL SEVEN: THE COMMITTEE SIZE CHECK
What it is
A simple cap. Seven people maximum in any decision-making body. Five for most decisions. Three for fast-cycle Type 2 decisions.
The math
Committee productivity peaks around five to seven and declines monotonically afterward. By twelve, most committees produce worse decisions than the median individual member would have produced alone. The mechanism is social loafing plus information cascade plus diffusion of responsibility plus speak-up cost rising with audience size.
The protocol
COMMITTEE SIZE CHECK
Before scheduling any decision meeting:
1. Who is required to make this decision?
____________________________________________________________
2. Who has unique information not available elsewhere?
____________________________________________________________
3. Who needs to feel ownership of the decision after the fact?
____________________________________________________________
4. Total unique attendees: ____
5. If > 7, which attendee adds the least marginal value and
can be briefed on the outcome instead?
____________________________________________________________
(Repeat until <= 7.)
The trap
Inviting people for political reasons. Every “we should include them too” adds a body to the room without adding judgment. The political cost of exclusion is concrete and felt. The cognitive cost of inclusion is abstract and ignored. The architecture honors the political cost by default. The operator who reverses this honors the cognitive cost too.
When exclusion is politically expensive, the alternative is the briefing. The excluded party gets the decision and the rationale within 24 hours. The participation cost drops from sixty minutes of meeting time to ten minutes of reading. The political cost is also addressed.
DRILL EIGHT: THE CONVEXITY FILTER
What it is
For every Type 2 decision in front of the operator, ask: is this convex or concave? Convex decisions have asymmetric upside. Concave decisions have asymmetric downside. The ratio of convex to concave decisions in the portfolio determines whether errors are cheap or catastrophic over time.
The convexity questions
CONVEXITY FILTER
DECISION: _________________________________________________
1. WORST PLAUSIBLE OUTCOME
What is the worst that can happen if this fails completely?
Cost in dollars, time, relationships, reputation:
____________________________________________________________
2. BEST PLAUSIBLE OUTCOME
What is the best that can happen if this works?
Upside in dollars, optionality, future opportunities:
____________________________________________________________
3. ASYMMETRY
Best / worst ratio: ___ x
Asymmetric upside (>3x): [ ] yes [ ] no
Asymmetric downside (<0.3x): [ ] yes [ ] no
4. CLASSIFICATION
[ ] Convex (asymmetric upside, capped downside)
[ ] Concave (capped upside, large downside)
[ ] Symmetric (close to fair)
5. ROUTING
- Convex Type 2: take many. Failure is data.
- Concave Type 2: take selectively. Failure is loss.
- Symmetric: optimize for speed (other variables dominate).
The portfolio test
Every quarter, list the major decisions made in the past quarter. Classify each as convex, concave, or symmetric. The healthy portfolio has more convex than concave. If the operator is taking mostly concave bets, the architecture is misaligned with how survival actually works in noisy environments.
The trap
Confusing convexity with optimism. A convex bet is not “the best case is great.” It is “the worst case is bounded and survivable, and the best case is much larger than the worst case.” A new product launch with $5k inventory and a viable upside of $100k in monthly recurring revenue is convex. Optimism about the upside has nothing to do with the structural property.
DRILL NINE: THE RECOGNITION CALIBRATION
What it is
A check on when the operator can trust pattern matching versus when the operator must force analysis. Recognition-primed decisions work in domains where the operator has thousands of prior patterns. They produce confident error in unfamiliar domains.
The calibration prompts
RECOGNITION CALIBRATION | DECISION: ____________________
1. DOMAIN FAMILIARITY
How many similar decisions have I personally made before?
[ ] 100+ [ ] 30-100 [ ] 10-30 [ ] 3-10 [ ] <3
2. FEEDBACK QUALITY
In the prior decisions of this kind, did I receive clear,
timely feedback on whether the decision was correct?
[ ] Yes, within weeks
[ ] Yes, within months
[ ] Sometimes, mixed
[ ] No, feedback delayed or noisy
3. CONTEXT SHIFT
How similar is the current context to past contexts?
[ ] Very similar
[ ] Similar with one or two notable differences
[ ] Substantially different
[ ] Almost entirely new
4. ROUTING
- 100+ decisions, clear feedback, very similar context:
RECOGNITION (trust the pattern, decide fast)
- Mixed signals on 1-3 above:
HYBRID (recognition first, then 30 minutes of explicit
analysis to check for blind spots)
- <30 decisions OR delayed feedback OR substantially
different context:
ANALYSIS (force explicit option generation, do not trust
gut, consult outside view)
The expert trap
The most dangerous recognition error is the senior operator entering an adjacent domain. The pattern library fires confidently. The patterns are wrong. The confidence is correct in form, mistaken in content. The fix is structural: in any new domain, default to analysis for the first six months regardless of seniority elsewhere. The architecture overrides the operator’s intuition during the calibration period.
The trap
Forcing analysis when recognition would have been correct, on the theory that more deliberation is always safer. Deliberation has costs. Decision load. Time. Sometimes worse outcomes when the analytic frame strips out the implicit knowledge the pattern library encoded. Match the mode to the domain.
DRILL TEN: THE DISAGREE-AND-COMMIT TRIGGER
What it is
A specific verbal protocol for closing a debate that has surfaced legitimate disagreement but cannot be resolved by further discussion in reasonable time.
The script
When a Type 2 decision has been debated for more than 30 minutes (or two meetings) and the disagreement is honest and the deciding party is clear:
DISAGREE-AND-COMMIT PROTOCOL
DECIDER (to dissenting party):
"I hear your concern. I understand it. I'm not dismissing
the substance of your point. I'm choosing to go in a
different direction because [one or two-line rationale].
I'm asking you to disagree and commit. You can record your
dissent. The dissent is on the record. But once we leave
this room, the team needs you executing the decision as if
it were yours, not arguing it in hallway conversations.
If I'm wrong, the trip-wire is [specific signal, specific
timeline]. When that signal fires, we revisit.
Are you in?"
DISSENTER:
"Disagree and commit. Dissent recorded. Trip-wire set."
[Decision closes. Execution begins.]
What it does
It separates two things that are usually conflated: the right to be heard and the right to delay execution. The dissenter retains the first. The architecture removes the second once the deciding party has decided.
What it requires
The trip-wire must be specific. “If sales drop” is not specific. “If unit sales drop more than 15% week-over-week for two consecutive weeks” is specific. The trip-wire is the architecture’s promise that the dissenter’s concern is not being dismissed permanently. It is being deferred conditionally. The condition is observable. When it fires, the decision reopens. When it does not fire, the dissenter has lived inside a successful execution.
The trap
Using disagree-and-commit on Type 1 decisions. For one-way doors, the architecture should accommodate longer debate, more analysis, and consensus-driven approaches. Disagree-and-commit is a Type 2 protocol. It exists to preserve speed in reversible decisions. Applying it to irreversible decisions trades a small dissenter relationship cost against a potentially catastrophic decision cost. The trade is wrong.
DRILL ELEVEN: THE TRIP-WIRE LOG
What it is
A persistent record of every trip-wire set across all open decisions. Updated weekly. Surfaces the conditional commitments the architecture has made to itself.
The format
TRIP-WIRE LOG | Last reviewed: _______________
DECISION TRIP-WIRE STATUS
──────────────────────── ─────────────────────────── ──────
New POS rollout Order errors >2% for 2 wks GREEN
Hiring freeze (kitchen) Avg ticket time >12 min YELLOW
Marketing spend +30% Ticket count rises <5% GREEN
Vendor switch (proteins) Quality complaints >3/wk GREEN
Avondale lease renewal Rent rise >12% YoY (n/a)
GREEN = signal not firing, decision holding
YELLOW = signal approaching trigger, watch closely
RED = signal fired, revisit decision this week
How to maintain it
Sunday evening. Five minutes. Walk through each row. Update status. If any row is RED, the corresponding decision reopens at Monday’s triage. The discipline is mechanical. The discipline is the point.
The trap
Setting trip-wires that cannot be observed. “If the project is going badly” is not a trip-wire. It is a hope. Trip-wires must be specific signals on observable metrics with timelines. The architecture cannot fire on vague conditions. If the trip-wire is vague, replace it with one that an outside observer could check without asking the operator’s opinion.
DRILL TWELVE: THE NEW-DOMAIN DEFAULT
What it is
A specific architecture for entering domains where the operator has no pattern library. New geographic markets. New product categories. New regulatory environments. New customer segments.
The protocol
NEW-DOMAIN ENTRY ARCHITECTURE
For the first six months in any genuinely new domain:
1. ANALYSIS DEFAULT
All decisions over $X (set the threshold low) require
explicit option generation, written rationale, and a
second perspective from someone with domain experience.
2. FEEDBACK DENSITY
Build deliberate feedback mechanisms that report results
within weeks, not quarters. Recognition-primed expertise
develops through fast feedback loops. Slow feedback loops
produce confident error.
3. CONVEXITY BIAS
Take many small convex bets. Avoid concentrated bets until
the pattern library has begun forming. Concentration
without recognition is gambling.
4. SUNSET CLAUSE
At the end of six months, audit what was learned. If a
functional pattern library has formed, relax the analysis
default. If not, extend the protocol.
5. EXTERNAL ADVISORS
Pay for domain expertise. The hourly cost of a domain
advisor is trivial relative to the cost of confident
error in the new domain. Most operators resist this on
ego grounds. The architecture overrides the ego.
The trap
Treating six months of restraint as wasted time. The restraint is the calibration. The architecture is investing in pattern formation. The return on that investment is everything that comes after. Operators who skip the protocol enter the new domain at full speed and pay for the calibration in the form of avoidable mistakes that the protocol would have prevented.
DRILL THIRTEEN: THE DECISION JOURNAL
What it is
A simple log entry written at the moment of every Type 1 decision and major Type 2 decision. Reviewed quarterly. The single highest-leverage practice for compounding decision skill.
The format
DECISION JOURNAL ENTRY
DATE: _______________
DECISION: __________________________________________________
CLASSIFICATION: [ ] Type 1 [ ] Type 2
CONTEXT (1-3 sentences):
____________________________________________________________
____________________________________________________________
OPTIONS CONSIDERED:
1. ________________________________________________________
2. ________________________________________________________
3. ________________________________________________________
OPTION CHOSEN: _____________________________________
WHY (state the actual reasoning, not the post-hoc story):
____________________________________________________________
____________________________________________________________
CONFIDENCE LEVEL: ____%
KEY ASSUMPTIONS (the things that have to be true for this to work):
1. ________________________________________________________
2. ________________________________________________________
3. ________________________________________________________
PREDICTED OUTCOME:
____________________________________________________________
WHEN TO REVIEW: _______________ (specific date)
The review cadence
Quarterly. Open every entry from the past quarter. Compare predicted outcome to actual outcome. Compare confidence to actual hit rate. The patterns that emerge are the operator’s calibration map.
Most operators discover they are systematically overconfident in one direction (usually optimistic on hires and pessimistic on market opportunities, but the pattern is individual). The journal makes the bias visible. Visibility is the precondition for correction.
The trap
Writing the journal entry after the outcome is known. The whole value is the prediction. The retrospective entry has no calibration value because the brain rewrites the prior probability to match the actual outcome. Entries written before the outcome are honest. Entries written after are fiction.
THE OPERATOR’S CHECKLIST
Daily
- The most consequential decision today is scheduled before position three in the day’s sequence.
- No more than fifteen decisions are on today’s plate. Beyond that, fatigue accumulates faster than judgment.
Weekly
- Monday: Type 1 / Type 2 triage on the week’s decisions.
- Thursday: Pre-mortem on Type 1 decisions from Monday.
- Sunday: Default audit. One default change identified. Trip-wire log reviewed.
Monthly
- Decision rights map reviewed. Anything to push down a level?
- Committee sizes audited. Any meeting consistently >7 participants?
- Recognition vs analysis balance: any new domain decisions still using recognition mode past the calibration window?
Quarterly
- Decision journal review. Calibration patterns identified.
- Convexity portfolio audit. More convex than concave?
- Architecture audit: what produced the worst decisions this quarter, and was it the people or the structure they decided inside?
THE OPERATOR’S TEMPLATES
The drills above contain templates inline. For convenience, this section consolidates the most-used templates as a single reference.
Type 1 / 2 triage card
DECISION REVERSIBLE? ROUTING
───────────────────── ──────────── ──────────────
[ ] Y [ ] N [ ] Ship now
[ ] Pre-mortem
Pre-mortem agenda
1. Read the failure scenario aloud. (3 min)
2. Individual writing: why did it fail? (15 min)
3. Round-robin reading of cards. (15 min)
4. Cluster failure modes. (10 min)
5. Top 3 failures: mitigations + trip-wires (12 min)
6. GO / NO-GO / GO WITH MITIGATION (5 min)
Default audit prompt
For each recurring outcome:
- Does anyone make this decision actively?
- If no one acted, what would happen by default?
- Is the default the correct outcome?
- If not, what is the architectural fix?
Trip-wire test
A valid trip-wire is:
- A specific observable metric
- With a specific threshold
- Over a specific time window
- That an outside party could verify without asking us
If any of these is missing, it is not a trip-wire. It is a hope.
Disagree-and-commit close
"I'm choosing X. The trip-wire is Y. When Y fires, we revisit.
Until then, I need you executing as if it were your decision.
Disagree and commit?"
THE FAILURE MODES OF THIS PRACTICE
The protocols above will fail in three predictable ways. Naming them in advance is the architecture’s defense against them.
Failure mode one: Ritual without function
The operator runs the Monday triage, the Thursday pre-mortem, the Sunday audit. The forms are filled out. Nothing changes. The drills became theater.
The diagnostic: at the end of each month, list the architectural changes made. If the list is empty, the rituals are running but the architecture is not being patched. The fix is to require at least one architectural patch per month, however small. The point of the audit is the patch.
Failure mode two: Over-instrumentation
The operator builds elaborate decision rights maps, exhaustive trip-wire logs, and detailed journal entries on every decision down to choosing a vendor for paper towels. The architecture has consumed more time than the decisions it was designed to improve.
The diagnostic: the time spent on architecture maintenance exceeds 10% of the operator’s working hours. The fix is ruthless subtraction. Most decisions do not need journals. Most decisions do not need trip-wires. Most committees can shrink. The architecture should disappear into the operation, not become the operation.
Failure mode three: Captured by the decider
The operator becomes the architecture. Every drill, every audit, every patch is filtered through the operator’s preferences. The architecture serves the operator’s comfort rather than the organization’s decisions.
The diagnostic: independent observers report that decisions are still made the same way they were before the architecture was implemented, just with more paperwork. The fix is structural transparency. Decision rights maps published to the team. Decision journals reviewed by a partner or board member. Default audits run by someone other than the operator on a rotating basis.
CLOSING
The architecture is not a body of knowledge. It is a set of habits embedded in the calendar, the meeting scripts, the document templates, and the recurring drills. The mechanism described in the pointing piece is universal. The implementation in this piece is one operator’s instantiation of it.
Use what fits. Discard what does not. The test is whether decisions in the operation are getting better, faster, or both. If the answer is yes, the architecture is working. If the answer is no, the architecture is theater regardless of how thoroughly the templates are filled out.
The operator who understands this stops believing decisions are about judgment. They become about structure. And the structure is something the operator builds, audits, and patches every week, the way a captain checks the rigging before weather.