Methodology & Sources

Editorial Scope & Curation Principles

EpochArc is neither an academic bibliography database nor a real-time news feed. Each curated milestone must rigorously substantiate three dimensions: the verifiable event details, the completeness of the primary evidence chain, and the mechanism by which it reshapes the long-term trajectory of AI development.

Each editorial run moves through discovery, screening, factual confirmation, editorial assessment, independent review, and explicit human approval. Screening scores set verification priority only; they neither authorize publication nor determine L1-L3 significance. Unconfirmed conditions remain unknown rather than being guessed. Possible Directions must be grounded in published events, observed signals, and an external-consensus search.

Significance Levels & Curation Criteria

L1 Important Record: The core fact is verified and worth preserving, but the evidence available by the cutoff does not yet establish a higher-level change pathway.

L2 Key Milestone: The event must satisfy at least one evidenced change pathway: independently verified capability or method adoption, real production use or access change, an executed commercial restructuring, binding governance action, an incident with external remediation, adoption of a safety practice, or sustained measured social behavior change. A product announcement, vendor benchmark, financing amount, or legal proposal does not automatically qualify as L2.

L3 Turning Point: The event must satisfy key-change pathways in at least two distinct categories, show at least one year of observed persistence, and be supported by primary material plus at least three independent information origins. Age alone does not establish lasting impact.

Significance levels have no quota and are not inferred from the Impact Score. Each assessment must state what has already changed, why the event is not one level lower or higher, and compare it with at least two published cases. Insufficient evidence leads to further review, not quota-driven mass upgrades or downgrades.

Source Hierarchy & Evidence Standards

Tier 1 · Primary Sources & Direct Evidence Peer-reviewed literature, official corporate/institutional announcements, model cards, public source code repositories, statutory legal texts, judicial filings, and official technical specifications.

Tier 2 · Independent Third-Party Verification High-credibility journalism, authoritative institutional reports, peer-reviewed academic literature reviews, and professional policy analyses used to validate macro-level impacts and real-world adoption.

Tier 3-4 · Intelligence Discovery & Interpretive Signals Expert analysis can support context and interpretation; community signals such as Hacker News, Reddit, and social media are primarily discovery inputs. Community attention cannot independently substantiate a published fact, impact, or significance rating.

Source tiers describe source type; evidence grades describe how strongly a particular claim is supported. Independence is deduplicated by information origin and controlling publisher: several outlets repeating one announcement are not multiple independent verifications, and confirming availability does not validate a vendor's performance claim.

Impact Scoring & Consensus Taxonomy

The Impact Score is independent from L1-L3 significance. It counts only consequences observed by the evidence cutoff and combines absolute dimensional severity, evidence strength, measured duration, and standardized audience scope. Only the largest absolute severity per dimension is counted, and audience synonyms cannot inflate scope. Expected effects do not enter the score. The index is a reproducible display heuristic, not a substitute for evidence.

Consensus classification is categorized into Broad Consensus, Actively Debated, or Emerging Consensus. An 'Actively Debated' label does not imply a negative rating; rather, it alerts readers that the empirical facts, historical significance, causal attribution, or long-term implications remain open to competing interpretations.

Possible Directions Review Rules

Possible Directions are not in-house verdicts about the future. They are observation tracks built from published timeline events, and every public direction must point back to real events and sources. The module shows which signals have already appeared in the world; it does not invent a private completion checklist.

EpochArc does not publicly define when a direction is 'achieved' unless there is an external consensus basis, such as a regulatory standard, a broadly used evaluation norm, or a stable cross-source industry definition of what counts as real deployment. In the absence of that basis, the direction remains monitor only: we keep tracking signals, evidence, and constraints without issuing a binary resolution.

Reader selections reflect what people want to keep watching; they do not change historical scoring and are never treated as evidence that a direction has already come true.

Limitations Statement

EpochArc is an independent structured-curation project, not an exhaustive academic reference library, investment advice, or a regulatory finding. Automated validation can check structure, references, thresholds, and version binding; it cannot prove that a quotation entails a conclusion, that discovery was exhaustive, or that a recorded human action genuinely occurred. v2.4 governs new and changed content. Unchanged historical records remain identified as migration debt and must not be mistaken for completed v2.4 factual and rating review.