Research method and limitations

The research question is: which IRTF results were demonstrably used in IETF work, through which mechanisms, and with what observed reach? Research groups develop research agendas; IETF working groups develop engineering specifications. An RFC's publication status is distinct from its stream. Neither an RFC number nor WG adoption alone proves an IRTF contribution.

What is counted

  • Candidate: a hypothesis selected for inspection, not a confirmed event.
  • Signal: the reason a candidate was discovered; several signals may belong to one candidate.
  • Review: a person's evidence-based decision, preserved with corrections.
  • Consolidated decision: agreement of independent roles or adjudication under the study plan. It is not automatic deduplication of all historical versions into scientific events.

Documentary candidates connect documents. Organizational candidates investigate RG/WG processes, including possible adoption of research in a WG mandate. Human candidates connect an observed person/address hint to documents, either through an existing documentary candidate or through an independent exploratory intersection. A base event is optional at discovery and required for T6 confirmation. Keep these units separate: an address is not a verified person, a document pair is not a confirmed contribution, and multiple T/O labels do not multiply an event. T6 describes a documented mediation role in an independently confirmed T1–T5 event.

What the application supports

Preserved source collection, incremental analysis, typed candidate discovery, source inspection, independent reviews and descriptive charts/exports are implemented. The discovery rules describe actual coverage. The evidence taxonomy defines what can be confirmed.

The application does not automatically verify technical/conceptual/empirical use, resolve global person identity, deduplicate all transfer events, generate a probability sample, calculate reviewer agreement or provide causal/network analysis. Those claims require additional work and a declared research design.

What each processing step does

Collect & Preserve retrieves source files and responses, preserves the original material and records provenance, integrity information, requests and checkpoints. An API can supply a citation relationship already structured; collecting a text does not itself parse its references.

Normalize & Discover organizes observations into document identities and typed relationships, parses references in preserved texts, and applies discovery rules. In the implementation, some graph normalization is performed in discovery.py. Separate web discovery functions generate organizational and human leads. A citation can become a documentary hypothesis; charter attribution plus a mandate can enrich an RG/WG hypothesis; a shared address can create an exploratory human connection without a documentary base. Each result retains supporting source references and unresolved checks.

The cutoff is an upper historical bound, not the beginning of the search. A completed query certifies pagination for that query, not a complete institutional history. Charter revision observations, revision publication and approval are distinct dates; none alone establishes when research was adopted.

Validation and interpretation

Audit the eligible origin universe independently of the discovered queue. Reviewing only retrieved candidates cannot measure missed cases. For a probability audit, declare strata, sample size, seed, selection probabilities and precision targets; report purposively selected difficult cases separately.

Use independent reviewers and measure agreement before adjudication. Percentage agreement and Cohen's kappa can be reported with label prevalence and missing cases; the application does not calculate them. Record disagreements, corrections and any rule changes. Software tests verify implementation, not scientific conclusions.

Report unique candidates, overlapping signals, review states, acquisition gaps and confirmed findings separately. A transfer rate needs a validated denominator of eligible consolidated origins. Citation count is not confirmed technical use or research quality. RG/WG attribution needs evidence at the relevant date; unknown and multiple affiliations remain explicit.

Dates retain their precision. Collection time is not event time. Publication-year charts and destination-date proxies are descriptive, not validated transfer latencies. Origins with no observed event at the cutoff are right-censored; do not invent event dates or interpret unequal follow-up as a causal difference.

Mutable metadata, missing history, ambiguous people, incomplete archives and undocumented influence limit coverage. A zero candidate count is not evidence of no transfer. NO_EVIDENCE requires documented search coverage and is not an automatic classification. Association is not causation.

Next: prepare a study plan → collect → discover → review.