# Methodology — Paths

> Look sideways for information, never for a verdict. 3,580 documented paths showing what people brought, were handed, and were surrounded by, with perseverance and luck held apart.

Canonical: https://paths.significanthobbies.com/methodology/

Research protocol

# Methodology & limitations

The public model is three questions. The detailed scores are secondary analyst annotations, not measurements.

Interpretation boundary
**Descriptive evidence, not a causal recipe.**

Reading contract

## What this project can say

- **Can describe:** 
Repeated conditions and capabilities in selected early-breakthrough paths.

- **Cannot establish:** 
Base rates, causality, counterfactuals, or a probability of success.

- **Current evidence:** 
[Inspect the live coverage ledger →](https://paths.significanthobbies.com/coverage/)

## Research question

Why does comparison fail when documented paths contain different conditions, repeated work, sequence, and unchosen openings?

## Unit of analysis

One person and one selected milestone that occurred at or before age 30. The milestone may be commercial, technical, creative, athletic, scientific, or institutional. The dataset does not claim that this was the person's final or most important achievement. The stored field remains `milestone_by_age_26` for dataset compatibility while the corpus is migrated; it now means the selected age-relative milestone.

## Selection

The project is building an uncapped public-evidence census of people with unusually strong documented outcomes for their exact age, across founders, operators, creators, athletes, researchers, and independent engineers. It does not claim access to private or unpublicized outcomes, and it does not downsample verified qualifiers to meet a record target. The live [evidence coverage ledger](https://paths.significanthobbies.com/coverage/) reports the discovered universe, research backlog, published corpus, and known blind spots.

The public [evidence coverage ledger](https://paths.significanthobbies.com/coverage/) reports the current source
depth, trajectory completeness, annotation confidence, indexability, cohort
composition, and independent-audit boundary directly from the published data.

New admissions use three versioned, fail-closed verdicts. The discovery gate
promotes dated, typed, age-relative outcome signals into research. The research
gate requires an applicable age-banded outcome rule and independently useful
evidence. The publication gate then reproduces identity, age, field, source,
and trajectory consistency against the completed record. A researcher-written
`eligible` status is not sufficient by itself. Current age-banded outcome rules
are labelled provisional research screens until exact-age field prevalence is
independently calibrated.

## Public explanatory model

The public product asks the same three questions on every primary surface:

1. **What they brought:** documented capability, drive, health, or early skill located in the person.
2. **What they were handed:** money, family standing, network, permission, tools, or safety already in place.
3. **What surrounded them:** place, era, institutions, mentors, peers, platforms, and timing that made later moves available.

The three sources are never summed. **Perseverance** is shown only through
sourced repeated work, setbacks, recovery, or sustained practice. **Luck** is
shown through documented structural timing, encounters, shocks, and the outcome
variance the archive cannot recover. Neither becomes a causal or moral score.

The detailed 22-field starting-condition and leverage rubric remains available
below as secondary research annotation. It is not the reader's primary mental
model.

This is a descriptive model, not a causal recipe. The dataset can show which
conditions and capabilities repeatedly accompany early breakthroughs. It cannot
establish that a condition made the outcome happen.

## Outcome reach

The selected milestone determines eligibility for this early-breakthrough dataset. `success_tier` is a separate editorial summary of documented career recognition through the data cutoff:

- **T1 — Global icon:** legendary or globally iconic career standing; a durable reference point well beyond the immediate domain.
- **T2 — Field-leading:** a dominant figure at the top of a field through major prizes, championships, commercial impact, or sustained elite recognition.
- **T3 — Domain-recognized:** notable and widely recognized among people who follow the domain.
- **T4 — Specialist-known:** notable, but recognition remains primarily within a niche or among specialists.

For public reading, T1 is described as **extreme public outlier**, T2 as
**field-leading**, and T3/T4 together as **professionally distinctive**. The
latter broadens the archive toward the kind of success often described informally
as top 0.1%. It is not a measured population percentile. Numeric percentiles may
only be published when an authoritative field-specific denominator exists.

The tier is not calculated from advantage or leverage scores, and it should not be read as the significance of only the selected early milestone. It does not
measure human worth, virtue, or future potential. The stored field remains
`success_tier` for compatibility, but career-recognition tier is the
interpretive name: recognition can describe a harmful or notorious career
without endorsing it.

## Outcome distribution context

The product translates each tier into a count, share, and cumulative rank band
inside the published dataset. It also shows how many lower-tier profiles are
present and the lower-tier-per-member ratio for the selected tier. These values
are rebuilt from the current dataset rather than stored as annotations.

The result is a **tier band**, not an exact individual percentile. The tier
does not order Bill Gates against the other T1 profiles. The denominator is also highly selected:
every record already cleared the early-breakthrough inclusion threshold.
Population prevalence, ordinary attempts, near-misses, and people who never
reached a documented milestone are absent. Tier shares and lower-tier ratios
therefore are not success probabilities or estimates of how many people failed.

Career outcomes in many fields can be heavy-tailed or power-law-like: a small
minority can account for a disproportionate share of recognition, reach,
wealth, citations, or attention, and repeated compounding can widen outcome
gaps. This is an interpretive model, not a fitted power law from the tier counts.
The four editorial tiers compress continuous careers into broad bands; the
project does not estimate a Pareto exponent or claim the tier distribution is a
statistical power law.

In the current build, average starting advantage and built or converted leverage rise
from T4 to T1, but the score ranges overlap heavily. The live [Insights analysis](https://paths.significanthobbies.com/insights/)
reports the current correlations and score-cell overlap. These are descriptive associations, not predictive thresholds.

## Two evidence layers

### Biographical layer

Names, milestones, age assignments, early-history summaries, family context, and source URLs are intended to summarize public biographical evidence.

### Annotation layer

Leverage engines and numeric scores are analyst interpretations applied through a common rubric. They are hypotheses about acceleration mechanisms.

## Secondary annotation reliability

On 31 July 2026, a secondary coder independently recoded a deterministic
16-record sample stratified across four cohorts and four selected-milestone age
bands. The coding packet withheld every published tier, dimension score, total,
archetype, confidence label, and annotation status until the secondary artifact
was complete.

Tier agreement was 50.0% exact, 100.0% within one band, with
quadratic-weighted Cohen's kappa of 0.723. Starting-advantage decisions were
64.1% exact and 97.4% within one point across 192 comparisons. Leverage
decisions were 33.8% exact and 91.9% within one point across 160 comparisons.
The secondary leverage totals were 6.875 points higher on average.

This is a diagnostic result, not a reliability certification. It shows that the
recognition ladder preserves ordinal direction better than exact boundaries,
the starting-advantage rubric needs narrower definitions, and the leverage
rubric is not anchored tightly enough for precise numeric interpretation. No
published scores were changed based on this pass. Independent verification of
the source facts remains at zero records.

## Starting-advantage score (0–24)

This score asks: **What access or conditions were documented near the beginning
of the path?** It sums twelve dimensions scored from 0 to 2:

- **0:** no clear documentation in reviewed sources
- **1:** meaningful documented advantage
- **2:** unusually strong, scarce, or directly catalytic advantage

Zero is not proof of absence. Biographies systematically under-report wealth, informal tutoring, introductions, family logistics, permission, and social capital.

For future coding, 1 requires explicit evidence of meaningful presence and 2
requires explicit evidence that the condition was unusually scarce, sustained,
or catalytic by the selected milestone. Later career evidence cannot backfill
an early advantage. Achievement alone does not establish prodigy status,
adversity as a catalyst, rare tools, or a strong peer. When one fact supports
several dimensions, the overlap must be disclosed rather than interpreted as
independent causal inputs.

## Built or converted leverage score (0–25)

This score asks: **What multiplying capacity was documented later in the path?**
It sums ten dimensions covering early serious reps, practice volume, scarce
skill, distribution, network, team, structural timing, concentration, capital
safety, and domain proximity.

For the 0–3 dimensions, 1 means a documented presence, 2 means strong or
repeated evidence, and 3 requires exceptional, path-dominant evidence by the
selected milestone. The binary early-reps field remains 0 or 1, and the
remaining 0–2 fields use the starting-advantage anchors above.

Distribution requires repeatable access to an audience, users, customers, or
selectors, not publicity after a win. Network requires documented relationships
that moved information or opportunity, not membership in an elite organization
alone. Concentration and capital safety require direct evidence rather than an
inference from achievement or institution. Domain proximity means repeated
contact with real problems, users, or operating constraints, not expertise
alone. A structural wave requires a documented timing mechanism that amplified
the path.

The score records the presence of leverage, not its provenance. A lever may be:

- **Self-built:** accumulated directly through practice, skill, focus, or distribution.
- **Advantage-enabled:** easier to develop because of a starting resource.
- **Earned access:** unlocked by earlier work, selection, or proof.
- **External:** supplied by timing, a platform change, or a structural wave.
- **Mixed:** produced by several origins that cannot be cleanly separated.

The product now adds a **person-level provenance inference** for every non-zero
lever. It links each leverage field to relevant documented starting conditions,
then considers limited biographical language indicating repeated self-directed
work or earned access. Structural timing is classified as external. When
several signals coexist, the origin is mixed. When the record cannot distinguish
origins, the output is unresolved.

Each inference displays its evidence signals, rationale, and low or medium
confidence. Missing evidence never becomes “self-built.” This is a transparent,
deterministic reading of the current annotations—not a measurement of merit, a
self-made percentage, or a claim about private effort.

## Condition factors (−1 to 3, read separately)

Alongside the twenty-two scored dimensions, every record carries three condition
factors that answer a different question. The dimensions describe *how much* of
something was present. These describe *where it came from*:

- **`personal_endowment_score` — what they brought.** Documented capability,
drive, or early skill located in the person rather than their surroundings.
- **`inherited_leverage_score` — what they were handed.** Money, family
standing, network, or permission already in place before the work began.
- **`catalytic_ecosystem_score` — what surrounded them.** Place, timing,
institutions, and peer group that made the next step available.

Each runs from −1 to 3:

| Value | Meaning |
| --- | --- |
| **−1** | An active headwind. The record documents a condition working against the path—poverty, displacement, a learning disability, absent infrastructure. |
| **0** | Nothing notable either way. Neither a documented advantage nor a documented obstacle. |
| **1–3** | An increasing tailwind, where 3 is exceptional. |

A −1 is a reading, not a missing value. Every record carrying one has a paired
prose summary describing the disadvantage. Collapsing −1 into 0 would erase the
distinction between a path that began level and a path that began behind, which
is the distinction this project exists to make visible.

Each score has a matching prose field (`endowment_summary`, `inherited_summary`,
`ecosystem_summary`) holding the evidence it rests on, plus a
`scoring_confidence` for the set.

**The three factors are never summed.** A combined total would be one more
number to rank people by, and ranking is the interpretation this project
refuses. Read side by side they explain why two paths diverged. Added together
they would score a person, which the evidence does not support and the argument
does not permit.

These remain analyst readings of what sources record. They are not measurements
of merit, talent, or effort, and a −1 is not a claim about anyone's capacity.

## Person-specific comparison breaker

Every published profile keeps a compatible `/am-i-the-next/<person>/` route,
but it is now a reading page rather than a questionnaire. It places the same
three factors side by side, then shows sourced perseverance, luck, and the
sequence of documented events.

The page identifies where comparison breaks: different conditions, different
timing, different encounters, and unobserved alternatives. It does not produce
a resemblance percentage, probability of success, predicted reach, ceiling,
or claim that the visitor is “the next” person. Comparison is the doorway to
the argument, not its conclusion.

## Luck and outcome variance

Luck is treated as cross-cutting rather than as a residual score:

- **Structural luck:** birthplace, era, family, geography, institutions, and proximity to a frontier.
- **Encounter luck:** meeting a collaborator, mentor, coach, investor, selector, or first customer.
- **Event luck:** an algorithm boost, market shock, competitor failure, injury avoided, or unexpected opening.
- **Outcome variance:** similar visible inputs can still yield different outcomes for reasons the record cannot recover.

The dataset has no failed control group and cannot reconstruct counterfactuals.
A numeric luck score would therefore imply false precision. Luck remains
explicit in the explanatory model, comparison breaker, Evidence page, and person
profiles, but unscored.

## Coin-toss thought experiment

The homepage includes a 64-to-1 thought experiment: 64 plausible starts narrow
to one visible peak after six hypothetical independent 50/50 gates. This is a
teaching device, not an observed attrition rate, career probability, fitted
model, or estimate of how many similar people failed.

Real careers are not fair independent coins. Starting advantage can change
which opportunities are available and their initial odds. Built or converted
leverage can improve later odds. Runway can create more attempts. Timing,
health, encounters, gatekeepers, shocks, and outcome variance can still redirect
an individual path. The illustration exists to make repeated consequential
uncertainty and survivor selection intuitive without converting luck into a
score.

## Programmatic page evidence gate

Person-specific pages are generated for all published records so internal
navigation remains complete. A page is indexable only when the record has at
least two listed sources, non-low leverage evidence confidence, a non-empty
milestone, and a trajectory. Records below that threshold receive `noindex`.
Pages expose the person-specific evidence and never claim affiliation or
endorsement.

## Source standard

Version 0.3 retains the sources used in the original exercise, including encyclopedia pages and stronger supporting sources where available. The release is therefore labelled `not_independently_audited`.

For a stronger v1.0:

- Require two independent sources for the milestone date.
- Prefer primary or official evidence where practical.
- Require two strong sources for sensitive claims about family wealth, class, or parental influence.
- Record retrieval date and the exact evidence span.
- Mark disagreement rather than forcing a score.
- Use `unknown` in a redesigned schema where missingness must be separated from a confirmed zero.

## Survivor bias

The sample begins with successful people. It cannot estimate whether an advantage causes success, how many similarly advantaged people failed, or how frequently people without the coded advantages succeed.

## Early-bloomer bias

The dataset is restricted to people who had a notable milestone **by age 30**. This is a deliberate design choice—the project is about unusually early outcomes—but it creates a structural bias:

- Late bloomers are underrepresented. Stan Lee published his first comic at 38, Vera Wang designed her first dress at 40, Samuel L. Jackson broke through at 43. None of them appear here.
- Person-specific resemblance only compares visible ingredients from early-breakthrough profiles. It cannot represent late-bloomer routes that fall outside the study window.

This bias is accepted for the project's age-relative coverage purpose and must remain visible when interpreting any annotation or comparison.

## Measurement concerns

- Wikipedia and English-language media create geographic, gender, class, and recency bias.
- Fame and page views are not equivalent to achievement.
- Family advantages are less visible than institutional affiliations.
- Later biographies often rewrite messy paths into coherent narratives.
- Several concepts overlap: an elite institution may supply tools, mentors, peers, credibility, and geography simultaneously.
- `direct_customer_domain_exposure_score` was coded broadly in v0.1 and should be re-audited before causal analysis.

## Appropriate use

- Exploratory pattern analysis
- Hypothesis generation
- Qualitative, person-specific path comparison
- Diagnosing buildable leverage gaps
- Building a better research protocol

## Inappropriate use

- Ranking human worth or talent
- Predicting individual success
- Claiming that resemblance makes someone “the next” named person
- Claiming causal effects
- Treating inferred family class as verified fact
- Publishing aggregate percentages without the survivor-bias and missing-data warnings
