Evidence Base · how grounding works

Evidence grounds the tradeoffs — it does not predict your future

Forked Futures does not use evidence to forecast a personal outcome. It uses a small, curated base of public sources and decision frameworks to ground each route's assumptions and make its tradeoffs clearer. Every piece of evidence is labeled by where it came from, and occupation-level data is only attached when the decision is actually occupation-shaped.

15 curated cardsProvenance-labeledMatch score, not a probabilityMock-first · keyless

These are evidence-grounded future scripts — plausible trajectories, not deterministic predictions. Forked Futures drafts the scripts; you choose what to test.

1 · Evidence pipeline

From your answers to a grounded route review

A directed flow. Your own answers drive it; evidence enters to ground the route assumptions, never to score you.

Step 1
Your answers

one messy situation + a short causal question chain

Step 2
Journey state

inferred values, constraints, fears — soft signals, never verdicts

Step 3
Route archetypes

10 distinct strategies scored against your signal

Step 4
Curated evidence cards

public sources + frameworks, gated to the decision

Step 5
Evidence-fit score

a transparent match score — not a probability

Step 6
Route review

assumptions, tradeoffs, and a low-cost first test

2 · Provenance labels

Every piece of evidence is labeled by where it came from

So you can weigh it accordingly. An inferred assumption is never dressed up as a sourced citation.

User-provided signal

Something you actually told us in the journey — your values, constraints, or fears.

Curated / reference-backed

A public source or decision framework, used at its true coverage level — never as a per-person forecast.

AI-inferred assumption

A reasonable assumption the system inferred from your answers. Flagged as such, and never shown as a citation.

3 · Source cards

The curated evidence base

Each card states what its source can and cannot support, at its true coverage level. No invented statistics; nothing here is an individual prediction.

Public data sources

career-shaped only

Occupation, education, labor-market and outcome data — attached only when the decision is occupation-shaped.

Occupational Outlook Handbook
U.S. Bureau of Labor Statistics
Labor-market dataoccupation-level
Can support
  • +Typical day-to-day duties of an occupation
  • +The direction (not the magnitude for you) of projected demand for a role
  • +The education and training a role typically expects
Cannot support
  • Whether you specifically will get hired or how much you will earn
  • Outcomes for a single person or a single employer
Limitation Occupation-level aggregates with a multi-year projection and reporting lag
O*NET OnLine
U.S. Department of Labor
Occupation dataoccupation-level
Can support
  • +The specific skills, tasks, and work activities an occupation involves
  • +Which of your strengths a role actually draws on day to day
  • +Adjacent occupations that share a skill profile
Cannot support
  • Whether a specific role will suit you personally
  • Your individual performance or satisfaction in it
Limitation Describes occupations in the aggregate, updated on a survey cycle
College Scorecard
U.S. Department of Education
Education datacohort-level
Can support
  • +Field- and program-level earnings and cost patterns across cohorts
  • +How outcomes vary by program rather than by school name alone
  • +The typical debt a program's graduates have carried
Cannot support
  • What you specifically will earn after a program
  • Causation — earnings reflect who enrolls, not only the program
Limitation Cohort aggregates, federally-aided students only in parts of the data
First-Destination Survey
National Association of Colleges and Employers (NACE)
Career-outcome datacohort-level
Can support
  • +Where recent graduate cohorts land roughly six months out (employed / continuing study / seeking)
  • +How first-destination patterns differ by field of study
  • +A base rate for 'still figuring it out shortly after graduating'
Cannot support
  • Your individual destination or timeline
  • Long-run career outcomes (it is a first-destination snapshot)
Limitation Self-reported, six-month snapshot by participating institutions
Baccalaureate & Beyond (B&B)
U.S. National Center for Education Statistics
Career-outcome datacohort-level
Can support
  • +Longer-run employment and further-study patterns of bachelor's cohorts over years
  • +How early choices relate to later enrollment and debt across a cohort
  • +A multi-year view that a six-month snapshot cannot give
Cannot support
  • Your individual long-run path
  • A causal effect of any one decision
Limitation Cohort longitudinal study with multi-year lag
American Community Survey — PUMS
U.S. Census Bureau
Labor-market datapopulation-level
Can support
  • +Population-level distributions of income, field-to-occupation flows, and demographics
  • +How spread out outcomes are within a field (the variance, not just an average)
  • +A reality check on how often a field leads to a given occupation
Cannot support
  • Any individual's income or path
  • Forward-looking projections (it is a current cross-section)
Limitation Population microdata sample; estimates carry margins of error

Decision frameworks

Framework-level reasoning lenses that apply to any decision.

Pre-mortem (prospective hindsight)
Gary Klein, Harvard Business Review (2007)
Decision frameworkframework-level
Can support
  • +Surfacing failure modes by imagining the route has already failed
  • +Turning a vague worry into named, checkable risks
  • +Designing early-warning signs before committing
Cannot support
  • The probability that the route will fail
  • Any individual outcome
Limitation A reasoning technique, not a measurement
One-way vs two-way doors (reversibility)
Decision-science / reversibility framing
Decision frameworkframework-level
Can support
  • +Classifying a route as reversible (two-way) or hard-to-undo (one-way)
  • +Matching deliberation to how reversible a step really is
  • +Justifying a fast, cheap try when the door swings both ways
Cannot support
  • Whether the route is right for you
  • A forecast of the outcome
Limitation A framing lens, not evidence about results
Outside view / base-rate thinking
Decision-science framing (Kahneman & Tversky tradition)
Decision frameworkframework-level
Can support
  • +Anchoring on how a reference class of similar situations usually goes
  • +Counterweighting an over-optimistic inside view of your own plan
  • +Framing 'how long / how hard' more realistically
Cannot support
  • A precise number for your case
  • Removing the genuine uncertainty in your specific situation
Limitation Only as good as the reference class you can honestly name
Regret minimization
Decision-science framing
Decision frameworkframework-level
Can support
  • +Projecting forward to which path you'd most regret not trying
  • +Separating fear-of-loss from fear-of-missing-out
  • +Weighing action vs inaction regret explicitly
Cannot support
  • Which choice is correct for you
  • How you will actually feel later
Limitation A values-clarifying lens, not a predictor of feelings
Smallest decisive experiment (value of information)
Decision-science framing
Decision frameworkframework-level
Can support
  • +Designing the cheapest test that could actually change the decision
  • +Pre-committing a pass/fail signal before running it
  • +Spending effort where it most reduces uncertainty
Cannot support
  • The result of the test before you run it
  • Certainty — only a cheaper way to learn
Limitation Quality depends on choosing a genuinely falsifiable signal
Deliberate practice / skill compounding
Skill-acquisition research framing (Ericsson tradition)
Decision frameworkframework-level
Can support
  • +Why focused, feedback-rich effort compounds a distinctive edge
  • +Structuring a long build around the hardest sub-skill
  • +Explaining why depth is hard to copy
Cannot support
  • How fast you specifically will improve
  • That effort guarantees any outcome
Limitation A mechanism, not a measured rate for your case
Barbell (floor + upside) risk framing
Risk-management framing (Taleb tradition)
Decision frameworkframework-level
Can support
  • +Pairing a protected floor with a bounded, asymmetric upside bet
  • +Capping downside while keeping exposure to a large upside
  • +Justifying a fallback floor underneath a riskier move
Cannot support
  • The size of the upside in your case
  • That the structure removes risk — it bounds it
Limitation A structuring lens, not a return estimate

Startup / customer-discovery frameworks

For routes that involve building or validating something new.

Customer Development
Steve Blank
Startup frameworkframework-level
Can support
  • +Testing whether a real demand exists before building heavily
  • +Getting out of the building to talk to target users early
  • +Separating a vision from evidence that anyone wants it
Cannot support
  • Whether your specific venture will succeed
  • Market size or revenue figures
Limitation A process framework; results depend on honest execution
The Mom Test
Rob Fitzpatrick
Startup frameworkframework-level
Can support
  • +Asking about users' past behavior, not their opinions of your idea
  • +Avoiding false validation from polite encouragement
  • +Designing interview questions that surface real demand signals
Cannot support
  • Whether the idea will work
  • A quantitative demand estimate
Limitation An interviewing discipline, not a measurement of the market
4 · Evidence-fit score

A match / support score — not a probability

The evidence-fit score expresses how strongly a route matches your answers and the reference support behind it. It is not a probability of success and not a forecast. It is computed transparently in code from six components:

User signal match+ raises

how much of your stated direction the route reflects

Constraint fit+ raises

how well it fits the constraints you named (higher when reversible)

Route archetype fit+ raises

how strongly the strategy matches your dominant values

Curated evidence support+ raises

how many curated reference cards apply to the route

Uncertainty penalty− lowers

lowers the score when the route is inherently uncertain

Inference penalty− lowers

lowers it for every assumption the system had to infer

The result is clamped to a sane band and shown with a qualitative label (Strong / Moderate / Loose), so the number never reads as certainty. It is a way to compare routes — not a claim about your outcome.

5 · Responsible-AI limitations

What this evidence layer is — and is not

Stated plainly, because honesty about limits is the point.

What it will not do

  • Not a prediction of what will happen to you.
  • Not a success probability or any kind of odds.
  • Not a personal outcome forecast — sources are occupation-, cohort-, population-, or framework-level only.
  • No private data required; the app never identifies real people.
  • Mock-first in production — the full experience runs locally with no API key.

Optional providers, honestly stated

  • The production demo defaults to this curated local evidence — it does not depend on live web search.
  • An optional live research / search provider can be connected when keys are present; the dispatcher upgrades transparently and is provider-ready.
  • If no provider is configured, the fallback stays deterministic — the same curated cards and local mock, so the demo never breaks.

No required environment variables. Nothing here sends your answers to a third party in the keyless demo.

See the evidence layer in action.
Walk the decision tree, then open any route's full review to see its cards.