Decision & Alignment Anti-Patterns

The anti-patterns in this section are specific to how decisions get made, documented, and communicated. They are distinct from general communication anti-patterns in that they fail at the decision layer — the process, the artifact, or the alignment itself. Most of them are organizational diseases that individuals cannot fix alone, but learning to recognize them is the first step.

Process Anti-Patterns

The Endless Discussion

Meetings, threads, and docs that revisit the same decision without ever closing. Often driven by a lack of named decider and absence of a deadline.

Signals:
  - The same topic appears on 3+ meeting agendas
  - Discussion has no clear "decision needed by" date
  - People argue the same points in the same order each time

Fix:
  Apply DACI or RAPID. Name the decider. Set a deadline. After the
  deadline, the decision is made; the discussion is closed.

Decision by Exhaustion

The loudest or most persistent voice wins because everyone else gives up arguing. Worse: the final "decision" is not a consensus but an exhaustion outcome.

Fix: Use structured input collection (written, anonymized where useful). Announce the decision path before the discussion: "We will hear 3 options, take written input for 48 hours, then X decides."

Consensus Masquerading as Decision

"Does everyone agree?" — silence is taken as yes. In reality, people who disagree either said nothing or agreed performatively. Two weeks later, the decision has no support.

Fix: Make disagreement explicit and safe. "Any concerns before we commit?" Actively invite skeptics. Use "disagree and commit" language — disagreement is recorded, commitment is expected.

The Jenga Tower

A series of decisions each of which is rational locally but which collectively create an unstable system. No single person sees the accumulated fragility.

Fix: Periodically review the decision log in aggregate. Look for compounding constraints. Empower someone to call for a "decision audit" when the system feels off.

Meeting as Decision Making

Using a meeting as the primary decision mechanism for complex topics. Decisions made verbally are forgotten, misremembered, or re-litigated.

Fix: Use written pre-reads (RFC, 6-pager) for any decision of consequence. Use the meeting for Q&A and alignment, not for the decision itself. Record the decision in writing before the meeting ends.

Documentation Anti-Patterns

The Decision Graveyard

ADRs, RFCs, or decision docs that are written but never referenced, linked, or updated. They exist as artifacts of ritual rather than as a living record.

Signals:
  - New joiners ask "why do we do X?" and nobody links to the ADR
  - The ADR directory is not indexed or searchable
  - ADRs are never marked deprecated or superseded

Fix:
  - Maintain an ADR index (README with status of each)
  - Link to ADRs from the code they describe (comment headers)
  - Revisit ADRs annually; mark superseded ones explicitly

The Rationale-Free Record

A decision record that states what was decided but not why. Useless 2 years later when someone considers changing it.

Bad ADR:
  "Decision: Use Postgres.
   Consequences: We use Postgres now."

Good ADR:
  "Decision: Use Postgres.
   Context: We needed partial indexes, JSON operations, and logical
   replication. MySQL's feature set was 18 months behind our needs...
   Consequences: Gained X, lost Y, mitigated by Z..."

The Fait Accompli RFC

An RFC published after the decision has already been made. Reviewers realize they are not being consulted but informed, and disengage.

Fix: If the decision is made, skip the RFC and write an ADR. Do not fake open discussion.

The Monster Doc

A single document that tries to be RFC, ADR, project plan, spec, runbook, and status update all at once. Serves no reader well.

Fix: One doc, one purpose. Link between them. Keep the RFC frozen at decision time; maintain implementation docs separately.

Undocumented Overrides

A formal decision exists (ADR, RFC) but day-to-day practice has drifted. No updated doc, no visible acknowledgment that the old decision is now wrong.

Fix: When reality diverges from the record, update the record immediately. The cost of a 20-minute update is far less than the cost of future confusion.

Responsibility Anti-Patterns

The Diffuse Accountable

Multiple people share the "A" (Accountable) in a RACI matrix — or, equivalently, two DRIs are named for the same thing. When something goes wrong, each points to the other.

Fix: One A per task. One DRI per thing. If that feels uncomfortable, the discomfort is a signal that nobody actually wanted to own it; figure out why.

The Ghost Decider

A RAPID or DACI is drawn up with a Decider named, but that person is not actually engaged, makes no decisions, and lets the work accumulate without resolution.

Fix: Escalate the decision to the deciding person's manager. Ghost deciders are an org dysfunction, not a project problem; they need executive intervention.

The Approval Cascade

A decision that requires 7 approvals before it can move. Each approval adds days. Nobody feels empowered to make the call.

Fix: Distinguish between "Agree" (veto power) and "Input" (informs, does not block). Most "approvers" in cascades should be Input. Real vetos should be rare and obvious.

The Unnamed Stakeholder

A decision is made, then blocked weeks later because a critical stakeholder was not consulted. Often Security, Legal, Compliance, or Finance.

Fix: Stakeholder mapping at the start of every decision. Use RAPID to explicitly name the "A" (Agree) roles. Engage them early as co-authors, not gatekeepers.

Alignment Anti-Patterns

The Loudest Room

Decisions made based on who shows up to the meeting, regardless of expertise, accountability, or impact. Typically favors senior or extroverted stakeholders.

Fix: Written pre-reads equalize input. Pre-collect input from people who cannot attend. Weight input by the RAPID role, not by who is in the room.

The HIPPO (Highest-Paid Person's Opinion)

The senior person's view wins by default, even when data or domain expertise contradicts them. Everyone else self-censors.

Fix: Anonymous input collection for proposals. Written-first culture where the senior person sees the written argument before the room. Explicit "disagree and commit" norms.

The Silent Veto

A stakeholder disagrees in private, agrees in the meeting, then quietly obstructs execution afterward. Alignment was performative, not real.

Fix: Surface disagreement in writing before committing. "Any concerns that would prevent you from actively supporting this?" If yes, resolve before the meeting ends.

The Aligned-on-Paper Team

Everyone signed the doc. Six weeks later, three teams are building incompatible things. "Alignment" happened at the doc level but not at the model level.

Fix: After the decision, require each affected team to state in their own words what they are now doing. Surface disconnects within a week, not a quarter.

Decision Re-litigation

A decision is made, then the same people raise the same objections every time it is applied. Each re-litigation saps energy and signals that the original decision lacks legitimacy.

Fix:

1. Acknowledge the objection with a link to the decision log.
2. Ask "what is new information since the decision was made?"
3. If nothing is new, the decision stands; move on.
4. If new information, schedule a proper re-decision, not ad-hoc
   re-litigation.

- Type 1 (irreversible, high-stakes): full RFC, executive review
- Type 2 (reversible, medium-stakes): ADR, peer review
- Type 3 (routine, low-stakes): DRI makes the call, logs if needed

Bezos's well-known "Type 1 / Type 2 decisions" distinction is the source here — match the process to the reversibility.

Diagnostic Questions

After a decision has been "made":

1. Can I name the single Decider/Accountable?
2. Is the decision written down in a discoverable place?
3. Could a new hire read the record and understand why?
4. Would the opponents of the decision say they were heard?
5. Is there a clear next step with an owner?

"No" to any of these means the decision is not actually made — it is likely to re-emerge.