Methodology

Harvard Club Index stores claims, not bare values. A claim is one source saying one thing about a club, for example “the SEAS student-organization page lists this club's website as X” or “the club's join page says applications close October 3.” Each claim keeps the page it came from, when we saw it, the exact quote where there is one, and how it was extracted. What you see on a club page is chosen from those claims by a deterministic resolver. Nothing is typed in from memory, and nothing is guessed.

Badges

• VerifiedStated by an official Harvard or club source, recently observed.
• ReportedStated by a secondary source, or an official source we haven't re-checked recently.
• CommunityReported by students; not yet confirmed by an official source.
• ComputedCalculated from sourced numbers.
• EstimateInferred or derived by us — not directly stated by a source.
• UnknownNo acceptable evidence yet.

When sources disagree we keep every version. The resolver shows one and lists the others under “Other versions.” Deleting evidence is never how a conflict gets fixed.

Source trust

Sources are ranked by who published them. A club's own website or social account counts as official only once ownership is established: a Harvard page links to it, or an already-established club channel does. An Instagram account that merely has a similar name stays unverified.

Confidence is computed as base (by tier) × extraction method × directness, never assigned by hand. Parsing a page scores 0.95; a language model's answer scores 0.85 and is accepted only with a verbatim quote from the page. Anything inferred rather than stated is halved. Independent sources that agree raise the displayed confidence.

Deadlines and semesters

Recruiting information belongs to a specific semester, for example Fall 2026, not to the club forever. Relative dates like “next Friday” are resolved only against the date we fetched the page, and a date with no time is never silently turned into 11:59 PM. Whether a cycle is open is computed from its dates each time you load the page.

Competitiveness

We never estimate an acceptance rate. A rate appears only when a club published both an applicant count and an admit count for a named semester, and we show it with the Computed badge and its inputs. A label is added only with at least 10 applicants: highly selective, selective, moderately selective, admits most applicants. When a club says its process is “very competitive,” we quote the club; we don't turn that into a number. Having a comp or an application says nothing about how selective it is.

Active or dormant?

Activity status is an estimate, shown with the reasons behind it. It uses signals like a current Harvard listing, a dated recruiting cycle, or a recently updated website. Missing signals lower confidence; they are not proof a club is gone.

Duplicates

The same club often appears in several directories under slightly different names. We merge records automatically only on a strong shared identifier, like the same official website or directory ID. Similar names go to a person for review. Every merge is logged and can be undone.

Fields we track

“Shown as old after” is when a Verified value is downgraded to Reported if we haven't re-seen it. Fields marked “no inference” must be stated by a source.

Versions

Every rule is versioned so results can be rebuilt exactly: resolver resolver-v1, confidence confidence-v1, entity matcher matcher-v3, selectivity selectivity-v1, activity activity-v1.

Questions or corrections? See About.