Glossary
Canonical terms with business examples. See also Transparency and What we do.
- DEFF
- Discernment Efficiency Frontier Function. Single-option formula from lib/deff.ts: (P * V - OC) / (1 + lambda * sum(weight_i * barrier_i)).
- Business example. A founder comparing “hire employee #4 full-time vs use a contractor for 6 months.” DEFF ranks hire FT at 0.68 and contractor at 0.55 after OC, ramp risk, and barriers—so the board sees why FT wins on paper, not just gut feel.
- Brier score
- Mean squared error of probabilistic predictions vs outcomes. Formula from lib/brier.ts: (1/n) * sum_i (p_i - o_i)^2. Lower is better. D = 1 - Brier (higher is better).
- Business example. Procurement said “90% chance this vendor hits on-time delivery” on 20 POs; only 12 did. Brier is high (poor). After coaching, they say 70% and hit closer to 70%—Brier drops and the team trusts their forecasts more.
- Calibration
- When you say 70%, the event happens ~70% of the time. Tracked per tag and overall.
- Business example. A PM tags decisions #launch-date. Their dashboard shows #launch-date is overconfident by ~15%. Next release plan they widen confidence and schedule buffers—missed dates fall without slowing every call to a committee.
- Hubbard interval
- 90% confidence interval on a numerical estimate; scored by whether the true value falls inside.
- Business example. CFO estimates Year-1 revenue from a new market: P10 $800k · P50 $2.1M · P90 $4.3M. After the year, actual is $1.9M (inside the 90% band). The interval was honest; a fake “exactly $2.5M” would have looked precise but been wrong.
- Opportunity cost (OC)
- Value of the next-best alternative foregone while waiting or choosing poorly.
- Business example. Delaying a ship by 2 weeks “for a clean refactor” costs ~$4,800 in lost user-weeks for a paid feature. OC makes that delay a line item next to engineering pride—not free time.
- Barrier
- Friction or risk factor reducing the effective DEFF of an option (weight × score).
- Business example. “Migration effort to switch payment processors” is a barrier on “move to Square.” Weight 0.4, score 0.8 (painful). DEFF demotes Square vs staying on Stripe even if list price looks better.
- Premortem
- Imagine the decision failed; list why. Surfaces risks before commitment.
- Business example. Before signing a 2-year office lease, the team pretends it failed: “Revenue missed; we are stuck in rent.” Probability ~0.31. That drives a shorter lease or remote-through-M12 path instead of a silent hope.
- Reference class
- Similar past decisions used as a base rate for probabilities.
- Business example. Before promoting engineer A, the system surfaces past #promotion decisions: only 60% of “promote now” calls still looked right at 90 days. The team anchors on 60%, not the last heroic sprint.
- QUICK / STANDARD / DEEP
- Friction tiers: ~15s gut log, ~5min structured decision, ~30min full analysis.
- Business example. QUICK: “Ship this subject line?” STANDARD: “Stripe vs Square for checkout.” DEEP: “Acquire company A for $4M vs build.” Same product, right amount of ceremony for the stakes.
- x402
- HTTP 402 micropayment protocol for agents paying USDC per API call.
- Business example. An automated buyer agent hits POST /api/v1/deff/calculate, receives 402, pays a few cents in USDC on Base, then gets the DEFF JSON—no monthly seat for that bot.
- MCP
- Model Context Protocol — tool interface for AI agents (our server at /api/mcp).
- Business example. In Cursor or Claude Desktop, the agent calls calculate_deff or log_decision over MCP with an API key—same engine as the web app, without scraping the UI.