Fast-and-Frugal Trees¶
Classify a case through ordered binary cues, with one immediate exit at each nonfinal cue and a final cue that resolves either outcome.
Core Idea¶
A fast-and-frugal tree (FFT) is a compact rule for choosing between two categories. It tests binary cues in a chosen order. At each nonfinal cue, one answer ends the search with a category; the other continues to the next cue. The last cue decides either way. In the standard form with \(m\) binary cues, there are \(m+1\) exits: one at each of the first \(m-1\) cues and two at the last. A case may therefore be classified before most cues are inspected.[ref-d1bbfc7be767][ref-c5430663bcb4]
The form does not by itself choose cues, thresholds, order or exit directions, and it does not guarantee high accuracy. Original simulations found these trees competitive with richer methods in studied problems; that bounded finding is not a universal superiority rule.[^ref-d1bbfc7be767]
Scope of Application¶
The structure has been applied to unlike binary decisions. A Bank of England paper draws a historical coronary-care example based on Green and Mehr: ST-segment change, chief-complaint chest pain and an “other factor” test route a case to one of two labels. Green and Mehr's actual prospective study used HDPI probability-chart cards; the later tree diagram is a retrospective reconstruction, not evidence physicians executed that exact FFT.[ref-ad888b78f93f][ref-769f404d6058]
The same Bank of England paper constructs an illustrative bank-vulnerability red/green tree using leverage, market-based capital, wholesale funding and loan-to-deposit cues. It documents a missed Wachovia case. Those historical thresholds are not current supervisory or investment guidance.[^ref-769f404d6058]
Clarity¶
The tree's topology, its construction, and its measured performance are separate questions. The topology requires one early exit and one continuation at each nonfinal cue. Construction selects particular tests and category labels. Performance depends on data, errors and consequences. A conventional fully branched tree may ask the same questions but let both answers lead to further tests; it is not this restricted form.[ref-c5430663bcb4][ref-769f404d6058]
An early exit is noncompensatory: later cues cannot overturn it within that path, because they are not consulted. This is an operating rule, not evidence that later data have no real predictive value.[^ref-d1bbfc7be767]
Manages Complexity¶
One continuing spine with early exits compresses many possible cue combinations into a small, inspectable rule. The decision maker can trace which cue ended a given case and how many questions were actually asked. The cost is deliberate omission of later information for early-exiting cases. In the Bank of England's historical example, the same simplicity makes both an early UBS red flag and a missed Wachovia vulnerability easy to explain.[^ref-769f404d6058]
Abstract Reasoning¶
To test an alleged FFT, declare two outcomes and order the binary cues. At every nonfinal cue check that exactly one answer exits and the other continues; ensure the last cue resolves both answers. Then test the resulting rule against relevant cases and error costs. A short path is a claim about search economy, not a substitute for out-of-sample validation.[ref-c5430663bcb4][ref-769f404d6058]
Knowledge Transfer¶
The clinical reconstruction and bank illustration share case, ordered cues, early exits and a final resolving cue, while their measurements and consequences differ. Transferring the decision-rule shape does not transfer clinical correctness, financial cutoffs or predictive accuracy.[^ref-769f404d6058]
The live Algorithm prime is a proposed strict parent: an FFT is a finite, definite sequence of tests terminating in a category. The live Heuristic and Bounded Rationality primes are related but do not replace this named one-sided tree grammar. Query-complexity Decision Tree Model and Decision Tree Pruning are different domain-specific identities. The frozen “matching heuristic” vocabulary proposal remains unadjudicated and is not an alias in this draft.
[^ref-d1bbfc7be767]: Laura Martignon, Konstantinos V. Katsikopoulos and Jan K. Woike, “Categorization with limited resources: A family of simple heuristics”, Journal of Mathematical Psychology 52 (2008), pp. 352–361, original university repository abstract and indexed original article. [^ref-c5430663bcb4]: Shenghua Luan, Lael J. Schooler and Gerd Gigerenzer, “A signal-detection analysis of fast-and-frugal trees”, Psychological Review 118 (2011), pp. 316–338, p. 320 definition; original-text indexed rendering, publisher full text access limited. [^ref-ad888b78f93f]: L. Green and D. R. Mehr, “What alters physicians' decisions to admit to the coronary care unit?”, Journal of Family Practice 45 (1997), pp. 219–226, original indexed abstract. [^ref-769f404d6058]: David Aikman et al., “Taking uncertainty seriously: simplicity versus complexity in financial regulation”, Bank of England Financial Stability Paper No. 28 (2014), pp. 7–8 Figure A and pp. 18–21 Figures 3–4.
Relationships to Other Abstractions¶
Current abstraction Fast-and-Frugal Trees Domain-specific
Parents (1) — more general patterns this builds on
-
Fast-and-Frugal Trees is a kind of Algorithm Prime
A fast-and-frugal tree specializes Algorithm to finite ordered cue tests with one-sided early category exits.
Hierarchy paths (2) — routes to 2 parentless roots
- Fast-and-Frugal Trees → Algorithm → Function (Mapping)
Neighborhood in Abstraction Space¶
Fast-and-Frugal Trees sits in a sparse region of the domain-specific corpus (61st percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Memory Encoding & Retrieval Effects (20 abstractions)
Nearest neighbors
- N-Back Task — 0.86
- Gray Code — 0.85
- Kneser–Ney Smoothing — 0.85
- Basic Block — 0.84
- Latent Learning — 0.84
Computed from structural-signature embeddings · 2026-10-08