How Evidence Works | Evidence Dependency — When Many Sources Share the Same Upstream Evidence

Ten articles say the same thing.

That sounds like ten confirmations.

But all ten may point back to one dataset, one witness or one mistaken summary.

Evidence dependency is the degree to which apparently separate pieces of evidence share upstream observations, methods, assumptions, instruments, witnesses or transformations—and therefore share some of the same possible errors.

This is the first pillar beneath How Evidence Works. The master owns corroboration broadly. This page owns the deeper question: how many genuinely separate routes to the world do we actually have?

Quick Read

Evidence should not be counted by surface appearances alone. Different websites can copy one report. Different studies can analyse the same dataset. Different observers can rely on the same instrument. Different models can inherit the same training corpus or assumptions. These items still add information, but not as much as truly independent routes would. The evidential job is therefore to map dependency: what observation sits upstream, which transformations are shared, which failure modes are common, and where a genuinely new route enters. Strong triangulation is most valuable when methods fail differently.

claim → evidence items → trace upstream routes → identify shared nodes → discount duplicated signal → seek differently failing route → update belief

Publication Count Is Not Independence Count

A claim appears in:

• a newspaper;
• a blog;
• a social post;
• a video;
• a newsletter.

All five cite the same press release.

There are five publications but one primary evidential route.

Shared Datasets Create Partial Dependence

Two research papers can use different analyses while drawing from the same underlying sample.

They may still provide distinct analytical insights.

But they do not provide two independent samples of the world.

Shared Instruments Can Correlate Error

Three observers use the same miscalibrated sensor.

Agreement among them may reflect shared instrument bias rather than true convergence.

How Measurement Works owns instrument quality broadly. Evidence Dependency asks how common measurement pathways change the weight of apparent corroboration.

Shared Assumptions Can Hide Inside Different Methods

Two models use different algorithms but assume the same boundary conditions.

If that shared assumption is wrong, both may fail together.

Method diversity is not enough if the crucial failure mode is shared.

Dependency Can Be Social

Five witnesses discuss the event before giving statements.

Their later agreement is less independent than five isolated observations.

Communication can transmit both truth and error.

Dependency Can Be Temporal

A later source cites an earlier source that itself relied on an even earlier estimate.

What looks like repeated confirmation across years may be a long citation chain with no new observation.

Dependency Can Be Hidden by Aggregation

A meta-analysis combines many studies.

Some studies may share participants, datasets, labs, protocols or measurement systems.

Evidence synthesis should consider overlap rather than treating every entry as fully separate by default.

Independent Routes Matter Because Errors Need Different Ways to Survive

If documentary records, physical traces and independently collected measurements all support the same claim, one shared error mechanism becomes less plausible.

The value of triangulation rises when the routes have different failure modes.

Not All Dependence Makes Evidence Worthless

Dependency reduces incremental weight; it does not necessarily erase it.

Two analyses of the same dataset can reveal different robustness checks.

Two witnesses who discussed an event may still preserve useful details.

The right response is calibration, not automatic rejection.

A Practical Dependency Map

CLAIM: Programme improved scores
SOURCE A: School report
SOURCE B: News article citing School report
SOURCE C: Blog citing News article
SOURCE D: Independent parent survey
SOURCE E: External assessment dataset

UPSTREAM ROUTES:
A/B/C → same school report
D → independent perception route
E → independent assessment route

EVIDENTIAL COUNT: 3 meaningful routes, not 5

A 20-Lens Evidence Dependency Audit

1. What exact claim is being supported?
2. How many visible evidence items exist?
3. How many primary observations sit upstream?
4. Which items copy one another?
5. Which share the same dataset?
6. Which share the same witness?
7. Which share the same instrument?
8. Which share the same method?
9. Which share the same assumptions?
10. Which share the same institution?
11. Could incentives create correlated reporting?
12. Did observers communicate before recording?
13. Are apparently new sources merely later summaries?
14. What transformations are common?
15. Which failure modes are shared?
16. Which routes fail differently?
17. What truly independent route exists?
18. How much incremental weight does each item add?
19. Would one upstream error collapse several items at once?
20. After dependency is mapped, how many genuine routes to the world remain?

For Primary Readers

If five friends repeat the same story they all heard from one person, that is not the same as five friends seeing the event themselves.

For Secondary Readers

Take five sources supporting one claim and trace them backward. Count how many genuinely independent observations remain after copied or shared routes are merged.

For Advanced Readers

Model evidential dependence as correlation among observations induced by shared latent upstream nodes. Naively multiplying confidence across dependent items double-counts signal; robust updating discounts correlated routes and values orthogonal acquisition mechanisms.

Final Thought: Count Routes, Not Repetitions

Ten echoes are not ten independent encounters with reality. Evidence becomes stronger when different routes reach the same conclusion without inheriting the same possible mistake.