AGGIES LabBLOCKMINE

Methods and validation · bm-methods-2.1-draft

How BLOCKMINE turns sources into qualified insights

1. Research goals

Identify opportunities (open problems, unmet healthcare needs) and threats (disclosed vulnerabilities, fragile dependencies) from public evidence, with every claim traceable to a record and a date, and with observations, derivations, and hypotheses kept visibly apart.

2. Source coverage

Operational connectors at this release: CISA KEV (full catalog), FIRST EPSS (400 most recent KEV CVEs), ClinicalTrials.gov API v2 (10 conditions × 40 registrations, plus totals), OpenAlex (8 sector phrase queries × 25 works), openFDA drug shortages (full list). Proposed: NVD, OSV, NIH RePORTER, USGS MCS. Coverage is bounded by design and stated on every page; see Data Sources.

3. Evidence taxonomy

TierLabelRule
E1ObservedDirect statement or numeric field from a specific source; published with provenance.
E2DerivedReproducible transformation of E1 records; published with formula and input snapshot.
E3HypothesisPlausible interpretation; marked unverified with contextual citations.
E4DemonstrationSynthetic example. Defined for completeness; as of 2026-11-08 no demonstration or synthetic record is published anywhere on the site.
E0Insufficient evidenceNot promoted to verified findings.

4. Transformation pipeline

Official APIs → connector adapters (fetch, normalize, checkpoint, audit) → committed snapshots with retrieval time and sha256 → derived aggregates with stated denominators → candidate insights → review → publication. At this release the LLM extraction and hypothesis stages are not run; all published non-demonstration insights are E2 derivations computed by code from snapshots.

5. Scoring definitions (bm-scoring-0.1)

All factors are bounded in [0,1]; a missing factor is reported as unknown and is never imputed as zero. Scores are withheld when essential factors are absent.

BusinessOpportunity

FactorWeightDefinition
problem_evidence0.25Quantity, relevance, and independence of sources showing a real problem.
unmet_need_strength0.2Cited indications that current responses are inadequate.
plausible_demand0.15Publicly documented demand proxies, not fabricated market size.
solution_feasibility0.15Practical ability to prototype or implement within known constraints.
differentiation_evidence0.15Grounded distinction from known alternatives, including uncertainty about undiscovered competitors.
validation_readiness0.1Availability of ethical, feasible experiments to test the hypothesis.

HealthcareResearchOpportunity

FactorWeightDefinition
documented_need0.25Source-backed burden or need.
care_gap_evidence0.2Documented service or access gaps.
research_gap_signal0.15Sparse or stale research activity relative to need (coverage-limited).
feasible_evaluation0.15An ethical, feasible evaluation design exists.
potential_benefit_evidence0.15Cited evidence that an intervention category could help.
equity_and_access_relevance0.1Relevance to underserved populations, with suppression rules respected.

PublicSignalIndex (cyber, public mode)

100 × (0.40 × epss + 0.35 × kev + 0.25 × normalized_cvss). Returned as Not scored when EPSS or CVSS is missing. Because the NVD connector is not yet active, every record is currently Not scored; EPSS and KEV inputs are shown. This index does not estimate loss probability or organizational exposure.

SupplyFragility

FactorWeightDefinition
concentration0.25Observed supplier or country concentration.
documented_disruption0.2Supported disruption incidents in the period.
lead_time_stress0.2Observed lead-time stress (often unavailable).
limited_substitutability0.15Documented scarcity of substitutes.
geographic_exposure0.1Geographic exposure at safe granularity.
digital_dependency_risk0.1Declared software component risk.

HHI

HHI = Σ sᵢ² on [0,1] shares, reported on 0–10,000. Suppressed when the denominator cannot be reconstructed or shares do not sum to one. The current site reports only a reporting-share HHI over shortage postings, explicitly not a market-share HHI.

Weights change ranks: a controlled example

Two items with factors a = 0.9 and b = 0.4. With weights (0.5, 0.5) the score is 65; with weights (0.2, 0.8) it is 50. Weights are research assumptions and are versioned so historical scores remain reproducible.

6. Confidence rubric

  • High: direct authoritative records, strong scope match, independent support, no unresolved contradiction, not stale.
  • Moderate: adequate evidence with coverage or causal limitations.
  • Low: limited, indirect, stale, or ambiguous support, or an unresolved contradiction.
  • Insufficient: cannot responsibly infer a finding.

Confidence is separate from magnitude, severity, or priority.

7. Data quality and time model

Four times are kept apart: event, published, retrieved, assessed (plus source-updated where available). A recent ingestion of an old record does not make the record recent. Zero results mean no evidence meeting filters in the selected sources, not absence of the phenomenon.

8. Known bias and limits

  • Publication and registration counts reflect activity and reporting practice, not prevalence or impact.
  • Phrase retrieval depends on vocabulary, language, and venue coverage.
  • KEV inclusion reflects CISA's evidence; EPSS is a probability of exploitation activity, not asset exposure.
  • openFDA entries are package-level and unvalidated per openFDA; absence of a posting is not adequacy.
  • Well-funded topics have more papers; transparent organizations disclose more weaknesses.

9. Publication review

Any named-entity weakness claim, medical claim, or supplier assertion requires human review before publication. This release publishes no named-entity allegations: the only named entities are those appearing verbatim in official catalogs (KEV vendor/product, FDA manufacturer fields).

10. Reproducibility

Run npm run ingest to refresh snapshots (registry records request URLs, retrieval time, record counts, sha256). Derived insights are recomputed from snapshots at build; exports carry a machine-readable manifest with insight id, snapshot times, and methodology version.

11. Responsible use and security policy

  • No live scanning, exploit execution, credential testing, or asset discovery; the site is read-only public intelligence.
  • No claims that a named organization is vulnerable, compromised, or non-compliant.
  • Healthcare content is research intelligence only; no diagnosis, treatment, or triage; no patient-identifiable data.
  • No inferred private supplier relationships; dashed edges for hypotheses.
  • Source text is untrusted data; no model output becomes a factual record without verification and review.
  • Report errors via the contact page; corrections preserve an audit trail.

12. Method changelog

  • 2026-11-08 (later) · bm-methods-2.1-draft: Blockchain Mining Research program refocused from proof-of-work mining to crawling publicly accessible data on public blockchains for opportunity and threat signals. The original page remains preserved and reproduced. Still no blockchain-derived data is ingested or published; the five registry connectors are unchanged.
  • 2026-11-08 · bm-methods-2.1-draft: removed all demonstration (synthetic) insight fixtures and illustrative graphics; the evidence graph is now drawn from the published insights' citations. assessedAt now equals the ingestion run time so it is never earlier than retrieval. The KEV × EPSS insight is computed over the full 400-CVE join. Site copy corrected to name the actual public sources (no blockchain data is ingested).
  • 2026-11-07 · bm-methods-2.0-draft: initial merged release (v2 specification). Five operational connectors, derived-only insights, demonstration fixtures labeled, PublicSignalIndex defined but Not scored pending CVSS.