×
Agents Graph Research Tools Method About Login Get Pro →
Trust & Evidence Standard

How the graph proves what it claims.

Forced Alpha maps physical dependencies between companies, materials, and facilities. A dependency is only useful if you can check it. So every relationship served by the graph follows the standard below. It is typed, evidence-graded and inspectable, path by path.

From question to inspectable answer

The model interprets the request. The graph determines which relationships exist. Every surviving path is filtered and graded, then returned with its evidence and its known gaps.

Input
Question or portfolio
Model
Resolve entities and intent
Graph
Traverse the physical dependency graph
Filter
Filter weak, noisy and invalid paths
Grade
Grade every surviving relationship
Output
Ranked answer with evidence and known gaps
The model routes. The graph decides, filters and grades.

Seven rules govern every relationship

They are the contract between what the graph shows you and what you can verify yourself.

01

Every edge is typed and directional

Each dependency states its relationship and its direction: supplies, depends on, produces, competes with. “A supplies B” is a different fact from “B supplies A,” and the graph never blurs the two.

02

Primary filings come first

The strongest evidence is a company’s own regulatory disclosure: SEC (US), EDINET (Japan), DART (Korea), MOPS (Taiwan), and official customs records. Secondary and reported sources are used only when a primary one is not yet on file, and are graded lower.

03

The supporting evidence travels with the claim

An edge is not presented as a bare assertion. When documentary evidence supports it, the relevant passage and source are attached to the relationship, allowing you to inspect the same evidence the graph used.

04

Every dependency is graded

Each relationship carries a grade for how well it is proven, from directly verified to structural hypothesis. The grade describes our proof of the dependency, not a view on any stock.

05

Paths are traced link by link

When the graph traces a multi-step path, say a shock at a raw material reaching a listed company several hops away, each link is presented with its own evidence and grade. A path is only as strong as its weakest link, and that weakest link is surfaced, not hidden.

06

Unverified links are labelled or excluded

A relationship that cannot be tied to sufficient evidence is clearly marked as unverified or excluded from evidence-gated analysis. It cannot appear as a clean, sourced fact.

07

Evidence is point-in-time

A filing establishes what was disclosed at a particular time. It does not establish that the relationship holds forever. Source dates are retained where available, and older evidence is presented as requiring renewed verification rather than treated as permanently current.

Four grades, applied to every relationship

The grade tells you exactly how far to trust a given link before you act on it.

Verified (A)Supporting excerpt verified against the cited source.
Sourced (B)A named primary source is on file for the relationship; the excerpt is not yet byte-verified against the document.
Reported (C)Supported by secondary or reported evidence. Corroborate before acting.
Structural (D)A possible transmission mechanism. An investigation lead, not a prediction.

The graph refuses rather than guesses

The most important thing a dependency graph can do is admit what it does not know. When no evidence supports a link, Forced Alpha returns the absence. It does not invent a plausible-sounding connection to fill the gap.

“Does Kanto Denka supply Tesla?”
No edge on file.

The standard is public. The implementation is proprietary.

This page is the standard the output follows: the proof, not the recipe. The methods that produce the graph are proprietary. Publishing the standard makes the boundary clearer, not fuzzier: you can audit every claim without us handing over how the system is built.

Proprietary — how the result is produced
  • Ontology and full relationship taxonomy
  • Extraction and entity-resolution pipeline
  • Confidence formulas and grade thresholds
  • Traversal, ranking and suppression logic
  • Chokepoint and asymmetry calculations
  • Verification gates and test harnesses
  • Proprietary datasets and coverage strategy
Public — the standard the output meets
  • Relationships are typed, directional and evidence-graded
  • Primary filings are preferred
  • Supporting evidence is attached where documentary
  • Paths are traced link by link, weakest link surfaced
  • Unverified links are labelled or excluded
  • The graph refuses when evidence is absent
  • Evidence is point-in-time; source dates retained where available

Every claim above is meant to be checked, not taken on faith. Inspect a dependency path in the graph, or read how the standard shows up in the research.