≈2k tokens on demand. Your agent reads this file only when SKILL.md points to it.
Bias Catalog
Purpose: Comprehensive cognitive bias taxonomy for AUDIT mode.
Read when: Executing BIAS_SCAN or DEBIASING phases.
Bias Taxonomy by Category
Decision-Making Biases
| Bias |
Description |
Detection Signal |
Debiasing Technique |
| Anchoring |
Over-relying on first information received |
Decision tracks closely to initial estimate or first option presented |
Independently generate estimates before seeing anchors; use multiple reference points |
| Confirmation |
Seeking information that confirms existing beliefs |
Only supporting evidence cited; disconfirming data absent or dismissed |
Assign devil's advocate; require "evidence against" section in every proposal |
| Sunk Cost |
Continuing because of past investment |
"We've already spent X" used to justify further investment |
Evaluate only future costs and benefits; ask "would we start this today?" |
| Status Quo |
Preferring the current state over change |
"It's always been this way"; change requires stronger justification than staying |
Flip the frame: "if we were starting fresh, would we choose the current approach?" |
| Framing Effect |
Different conclusions from same data presented differently |
Decision changes when phrased as gain vs loss |
Present data in multiple frames (gain/loss, absolute/relative, now/future) |
Group Biases
| Bias |
Description |
Detection Signal |
Debiasing Technique |
| Groupthink |
Desire for conformity overrides critical evaluation |
No dissent recorded; unanimous agreement without debate |
Anonymous pre-voting; designated dissenter role; independent scoring |
| HIPPO |
Highest Paid Person's Opinion dominates |
Most senior person speaks first; others align |
Reverse seniority speaking order; written submissions before discussion |
| Bandwagon |
Adopting popular positions |
"Everyone else is doing it" justification |
Ask "what would we do if nobody else had done this?" |
| Authority |
Deferring to authority without evaluation |
"Expert X says" without examining evidence |
Evaluate arguments independent of source; require evidence, not appeals |
Estimation Biases
| Bias |
Description |
Detection Signal |
Debiasing Technique |
| Planning Fallacy |
Underestimating time, cost, and risk |
Historical estimates consistently exceeded; best-case scenarios assumed |
Use reference class forecasting; add 50% buffer; check historical actuals |
| Optimism |
Overestimating positive outcomes |
Risk section is empty or perfunctory; only success scenarios considered |
Require pre-mortem analysis (→ Omen); mandate worst-case scenario |
| Dunning-Kruger |
Confidence-competence mismatch |
High confidence claims in unfamiliar domain; no acknowledged unknowns |
Calibration questions; require team members to rate confidence with evidence |
| Availability |
Overweighting recent/vivid events |
Last incident dominates risk assessment; decisions swing with news cycle |
Use base rates from historical data; time-weight inputs deliberately |
Evaluation Biases
| Bias |
Description |
Detection Signal |
Debiasing Technique |
| Survivorship |
Learning only from successes |
No failed project analysis; "company X succeeded by doing Y" |
Systematically study failures; ask "how many tried this and failed?" |
| IKEA Effect |
Overvaluing self-built solutions |
NIH syndrome; rejecting better external alternatives |
Blind evaluation: assess options without revealing origin |
| Halo Effect |
One positive trait influences overall judgment |
"They're great at X, so Y must be good too" |
Evaluate each dimension independently with separate criteria |
| Loss Aversion |
Fearing losses more than valuing equivalent gains |
Asymmetric weighting of downsides; avoiding any risk for small potential losses |
Quantify both gain and loss in same units; focus on expected value |
| Endowment |
Overvaluing what we already have |
Current solution rated higher simply because it's ours |
Pre-commit to evaluation criteria before comparing |
| Recency |
Overweighting recent events in judgment |
Last quarter/sprint dominates evaluation; older data dismissed |
Deliberately include multi-period data; weight by relevance, not recency |
Bias Interaction Patterns
Biases rarely occur in isolation. Common compounds:
| Compound |
Biases |
Effect |
Example |
| Escalation Trap |
Sunk Cost + Confirmation + Optimism |
Continue failing project, seeking confirming evidence while dismissing warning signs |
"We're almost there" repeated for months |
| Innovation Block |
Status Quo + IKEA Effect + Loss Aversion |
Reject better external solution in favor of inferior internal one |
Building custom when OSS solution exists |
| Echo Chamber |
Groupthink + Confirmation + Authority |
Team reinforces shared assumptions without external challenge |
Unanimous "obvious" decisions that fail |
| False Precision |
Anchoring + Planning Fallacy + Dunning-Kruger |
Confident precise estimates in uncertain domains |
"This will take exactly 3 sprints" with no basis |
AUDIT Mode Procedure
Phase 1: CLASSIFY
- Identify the decision/plan/proposal to audit
- Categorize: strategic / technical / resource / people / process
- Note who is involved, what's at stake, and time pressure
Phase 2: BIAS_SCAN
- Walk through each bias category systematically
- For each bias, check:
- Is the detection signal present? (evidence)
- How confident are we? (HIGH/MEDIUM/LOW)
- How impactful if unaddressed? (HIGH/MEDIUM/LOW)
- Check for compound patterns
- Document findings with specific evidence, not speculation
Phase 3: DEBIASING
- For each detected bias, select the appropriate debiasing technique
- Generate a debiased alternative framing
- If compound pattern detected, address the root bias first
- Propose process changes to prevent recurrence
Phase 4: CRYSTALLIZE
- Produce Bias Audit Report (see template below)
- Include original decision framing vs debiased framing
- Recommend next steps (proceed, revise, escalate to Magi)
Bias Audit Report Template
# Bias Audit Report: [Decision/Plan Name]
## Context
- **Decision:** [what is being decided]
- **Stakeholders:** [who is involved]
- **Stakes:** [what's at risk]
- **Time pressure:** [HIGH/MEDIUM/LOW]
## Detected Biases
| # | Bias | Evidence | Confidence | Impact | Debiasing |
|---|------|----------|------------|--------|-----------|
| 1 | [name] | [specific evidence] | [H/M/L] | [H/M/L] | [recommendation] |
## Compound Patterns
[If detected, describe the interaction]
## Original vs Debiased Framing
| Aspect | Original | Debiased |
|--------|----------|----------|
| [aspect] | [original framing] | [alternative framing] |
## Verdict
- **Bias severity:** [Critical / Significant / Minor / Clean]
- **Recommendation:** [Proceed / Revise / Escalate to Magi]
- **Process fix:** [suggestions to prevent recurrence]
Real-World Bias Failure Cases
Reference cases for severity calibration:
| Case |
Primary Bias |
Cost |
Lesson |
| AOL-Time Warner merger |
Confirmation + Optimism |
$99B write-off |
"Digital convergence" assumption never challenged |
| Kodak digital camera delay |
Status Quo + Endowment |
Company bankruptcy |
Overvalued film business, undervalued digital disruption |
| Nokia smartphone transition |
IKEA Effect + Groupthink |
Market dominance lost |
"Symbian is good enough" internal echo chamber |
| Healthcare.gov launch |
Planning Fallacy + Authority |
$1.7B+ |
Expert estimates accepted without reference class data |
| Mars Climate Orbiter |
Groupthink + Confirmation |
$327M |
Unit mismatch signals dismissed as instrument error |
