Reference-Class Forecasting Workbook¶
Forecasting procedure — instantiates Reference-Class Planning Calibration
A step-by-step worksheet that defines the forecast object, selects a comparable class, and pulls the estimate toward that class's actual outcome distribution by a documented adjustment.
Between a pile of historical cases and a corrected estimate sits a procedure, and Reference-Class Forecasting Workbook is it: the guided, step-by-step worksheet that takes one specific forecast and runs it through the outside view. It has the team state exactly what is being forecast, choose the comparable class, read that class's actual outcome distribution, and then apply a documented adjustment that moves the inside-view estimate toward the class — recording, in writing, how far it moved and why. Its defining move is the adjustment rule: an outside view that does not change time, cost, scope, or risk is merely decorative evidence, so the workbook's whole point is to force the estimate to actually shift and to make that shift auditable. It is the per-forecast reasoning engine of the archetype — it selects a class and computes the correction for this project; it does not maintain the library of cases those classes are cut from, nor produce probabilistic ranges as its primary output.
Example¶
A city transit authority is estimating the cost of extending a light-rail line and, as with most rail projects, the engineers' bottom-up estimate is confident and low. The team works a reference-class forecasting workbook.[n1] Step one, define the forecast object: capital cost to revenue-service, in constant dollars, excluding rolling-stock procurement (which is on a separate contract). Step two, select the class: completed urban rail extensions of comparable length and tunneling fraction — a boundary drawn to include the messy ones, not just the showcase projects. Step three, read the class's actual cost-overrun distribution against original estimates. Step four, the adjustment: the workbook applies the class's typical uplift to the engineers' figure, moving the estimate up by a documented percentage and recording the reasoning so a reviewer can see the inside-view number, the class it was compared against, and the exact size of the correction. The output is not "the engineers were wrong" but a calibrated capital figure with a visible audit trail from optimistic estimate to comparable-case adjustment. That trail is what survives into the funding decision.
How it works¶
- Define the forecast object first. State precisely what is being estimated and what is excluded, because the class cannot be selected — and overruns cannot be prevented from hiding as scope changes — without it.
- Select and bound the class. Choose comparable completed cases and draw the boundary explicitly: broad enough to capture real friction, narrow enough to stay relevant, transparent about exclusions.
- Read the class distribution, then adjust. Compare the inside-view estimate to the class's actual outcomes and apply a documented rule that moves the estimate toward the class — a percentage uplift, a shift to the class median, or a blend.
- Record the correction. Log the inside estimate, the class, and the size and rationale of the adjustment, so the outside view is auditable and cannot be silently discarded.
Tuning parameters¶
- Class breadth — how widely the comparable set is drawn. Broad classes have more cases and capture more friction but dilute relevance; narrow classes are more similar but risk too few cases and cherry-picking.
- Adjustment strength — how far the estimate is pulled toward the class. Full regression to the class median maximizes debiasing but can override genuine, evidenced local differences; a partial pull respects local knowledge but leaves optimism partly intact.
- Blend of inside and outside — the weight given to the team's own estimate versus the class. More outside weight corrects bias harder; more inside weight preserves tailored expertise. This is the archetype's central tension, made into a dial.
- Documentation depth — how fully the adjustment's rationale is recorded. Thorough documentation makes the forecast auditable and defensible but adds friction to every estimate.
When it helps, and when it misleads¶
Its strength is that it operationalizes the outside view into a repeatable procedure and forces the estimate to move — the exact discipline the archetype demands, since an unchanged forecast means the reference class was decorative. The written adjustment trail also satisfies the invariant that a commitment preserve a trace from initial estimate to class to correction.[n1]
It misleads most through class selection, because the same procedure that corrects optimism can launder it: a subtly gerrymandered class — comparables chosen to resemble the hoped-for outcome — yields a small "principled" adjustment that ratifies the original estimate under the authority of method. It can also over-correct, flattening a genuinely novel project into a class it does not belong to. The guarding discipline is to expose the class boundary to a separate membership challenge (a premortem or a skeptic) rather than trusting the workbook's own author to have drawn it honestly, and to keep the adjustment rule fixed before the class is finalized so the correction is not tuned to a preferred answer.
How it implements the components¶
outside_view_adjustment_rule— its core engine: a documented rule that moves the inside-view estimate toward the class's actual distribution, so the outside view changes the number rather than merely informing it.reference_class_boundary— it draws the explicit boundary of comparable cases for this specific forecast, deciding which cases count.forecast_object_definition— it opens by stating exactly what is being forecast and excluded, the precondition for selecting any class.
It does not maintain the standing library of comparable cases and their outcomes (base_rate_distribution) — that repository is the Historical Project Outcome Database, which the workbook queries. It also does not, on its own, adversarially test whether its class was gerrymandered; that skeptical class_membership_challenge is the Premortem as Auxiliary Probe, which returns findings to widen the class here.
Related¶
- Instantiates: Reference-Class Planning Calibration — the per-forecast procedure that turns comparable-case evidence into a corrected estimate.
- Consumes: Historical Project Outcome Database — supplies the cases the class is cut from; Independent Estimate Round supplies the un-anchored inside estimate it corrects.
- Sibling mechanisms: Historical Project Outcome Database · Premortem as Auxiliary Probe · Independent Estimate Round · Three-Point Estimate with Base Rates · Schedule and Cost Risk Register · Contingency Reserve Formula · Launch or Commitment Readiness Gate · Forecast Backtesting Review
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Reference Class Forecasting Workbook operates by selects comparable completed cases and computes an outside-view forecast with explicit scope. That concrete deployed or enacted form is Analysis, Modeling & Optimization under the frozen taxonomy.
Nearest alternative: Interface, Display & Cue — Although Interface, Display & Cue can support this mechanism, the frozen evidence makes its operative form the act that selects comparable completed cases and computes an outside-view forecast with explicit scope; the alternative is therefore secondary rather than defining.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Behavioral Economics
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: The workbook operationalizes behavioral economics' outside-view correction for planning fallacy.
Related originating lineages:
- Statistics & Experimental Design — Comparable-class selection and adjustment rely materially on statistical sampling and estimation.
Review resolution: Both blind reviewers agree that behavioral_economics is the primary origin. Explicit reconciliation of alternate origin disagreement adopts reviewer_a's classification because the workbook operationalizes behavioral economics' outside-view correction for planning fallacy. The resulting lineage records alternates=statistics_experimental_design, origin_mode=cross_disciplinary_synthesis, and domain_reach=multi_domain; these describe formative provenance separately from later applicability.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Reference-class forecasting, developed by Bent Flyvbjerg from Kahneman and Tversky's outside view, forecasts a project by (1) identifying a class of comparable past projects, (2) establishing that class's distribution of actual outcomes, and (3) adjusting the project's own estimate toward that distribution — the procedure this workbook operationalizes. ↩a ↩b