# Recruiting coverage and eligibility audit

Window: August 1, 2025 through May 31, 2026, inclusive UTC. Updated September 5–6, 2026.

## Measured retained coverage

The full-description internship audit returned **34,236 records**, of which 32,855 retained nonempty descriptions. Its broader title vocabulary includes summer associate/analyst, co-op and selected multilingual internship words, as well as retained internship classifier tags. A metadata query recovered location/source fields for 1,821 newly retrieved candidates. Four earlier reviewed recall records sit outside this query vocabulary. After six repeated same-company listing URLs are linked, the comparison corpus has **34,234 records**.

The inspected database contained 2,163,597 hot/archive job rows at the initial inventory, with 1,186,272 database insertions inside the window. Those totals include full-time and unrelated jobs. No August–November insertions survived in the inspected tables. December starts on the 12th. The cause of missing earlier history is unknown; this is not a finding about Scoutify's launch date.

| Added to Scoutify, UTC | Internship candidates | Company IDs | Explicit MBA reviewed | Graduate-business reviewed |
| --- | ---: | ---: | ---: | ---: |
| 2025-08 | Unavailable | Unavailable | Unavailable | Unavailable |
| 2025-09 | Unavailable | Unavailable | Unavailable | Unavailable |
| 2025-10 | Unavailable | Unavailable | Unavailable | Unavailable |
| 2025-11 | Unavailable | Unavailable | Unavailable | Unavailable |
| 2025-12 | 1341 | 104 | 52 | 31 |
| 2026-01 | 3550 | 789 | 69 | 135 |
| 2026-02 | 4809 | 1039 | 95 | 206 |
| 2026-03 | 7524 | 1111 | 142 | 359 |
| 2026-04 | 9780 | 1971 | 112 | 405 |
| 2026-05 | 7230 | 1445 | 56 | 294 |

Counts reflect recruiting and collection coverage together. The number of represented companies changes each month. There is no defensible uniform monthly quota or market-wide peak estimate from this retained sample.

## Eligibility expansion

The original explicit-MBA/title/tag screen and later recall audit reviewed 807 candidate records. This expansion added full descriptions for the broader internship universe, searched role-specific graduate-business routes, and recovered fuller historical qualifications.

The local full-description retrieval identified 3,000 previously unreviewed leads: 2 explicit-term matches and 2,998 graduate/business-vicinity matches. These were leads, not automatic acceptances. The two explicit matches proved to be a generic business-master qualification and an employee education benefit, illustrating why keyword counts are unreliable. The graduate routes were divided into 689 US-included business-title leads, 1,129 other/unknown-geography business-title leads (one parent-owned Philips overlap omitted), and 358 prioritized additional role-title leads. Eighty earlier uncertain decisions were also reassessed. Remaining low-priority vicinity matches were not all individually reviewed.

The combined Scoutify audit now contains **2,983 records: 1,956 included, 392 excluded and 635 held**. The included set has 526 explicit MBA records and 1430 graduate-business records, before links with external sources and any cross-alias duplicate consolidation in the merged dataset.

Graduate-business inclusion requires role-specific current student/master's eligibility in a relevant business field and plausible responsibilities. It does not require the literal acronym MBA. Technical-only graduate degrees, incidental master/MBA words, undergraduate-only pathways, current degree-completed-only routes and incompatible local-university restrictions are held or excluded. Some specialized analytical roles remain suitable only for students with relevant quantitative skills. Long or off-cycle placements retain their constraints; working-student, fellowship and training formats are not automatically counted as ordinary internships.

Each review saves a per-ID decision, short evidence quote, rationale and relevant constraints. Assistants reviewed qualification passages and surrounding context, using identical qualification templates where appropriate and targeted scans for enrollment, duration, geography, language and authorization limits. This is not independent human verification or employer confirmation. Confidence in a visible master's route is distinct from inferred MBA suitability.

## Historical external evidence

Six immutable Simplify/PittCSC snapshots yielded 310 MBA-title/tag candidates, including 71 MBA-title records. The 239 degree-tag-only pages were recovered through public role data, with original ATS identities and description hashes. Role-specific MBA wording, internship format, cohort and exact requisition identity were then checked. Degree tags alone were insufficient. Full-time university recruitment, technical-only MS/PhD roles, stale training ads and duplicate mirrors were rejected or held.

The repository now has a 2027 name, but [immutable 2025 content](https://github.com/SimplifyJobs/Summer2027-Internships/blob/bf9e6c78fa7d76af74d51bcc83e237897d94c50e/.github/scripts/listings.json) remains attributable. Snapshot presence is archived evidence, not first-ever appearance or an employer opening. Current recovered requirements are later corroboration, explicitly labeled as such. A tracker Summer 2026 label alone does not pass the app's explicit Summer 2026 filter.

Earlier source research also recovered university-board dates, newsletters, employer pages and two verified first Git additions from Jobright history. These selection mechanisms differ; do not sum their weekly counts as if they sampled the market uniformly. research/ preserves the reviews and source URLs.

The additional pre-November database investigation assessed 17 sources. Fifty-four LinkedIn data-science snapshots contained 10,592 fall observations representing 2,198 distinct jobs, with just one MBA/internship overlap. A public USAJOBS mirror contained 168,358 announcements in 2025, including 43,826 in August–October and 53 student/intern title leads. Its bulk records lack descriptions; four official retained announcements were examined and two graduate-business routes accepted. One independently corroborated Nordstrom role was added. This is a bounded review, not exhaustive screening of all 53 leads.

Common Crawl supplied three further explicit-MBA Summer 2026 listings with original qualifications preserved before November: two Vanguard tracks and one Amazon Japan role. Their WARC target URLs, timestamps and payload hashes match the public index. Vanguard board dates remain separate from capture dates; all three employer publication dates are blank. Two stale J&J Summer 2025 captures and an evergreen Cahoot posting with May 2024 metadata were held or excluded instead of turned into new fall openings. The reports and exact byte-range/hash receipts are in research/databases/. Full WARC payloads and restricted bulk descriptions remain local scratch.

A legacy October 18 local job-only snapshot recovered 194 rows, five internship titles and no descriptions; one Airbnb graduate-business role was qualified through its exact original identity and later corroborating qualifications. The legacy DynamoDB seen_jobs table was checked by metadata only in the verified account and us-east-1 region and returned ResourceNotFoundException. No backup containing personal records was restored. This does not resolve the missing broader fall Scoutify history.

## Retention, semantics and remaining gaps

- Database created_at is first retained insertion. The audited writer previously substituted current time when a source date was missing; Greenhouse used updated_at and Workday relative dates were approximate. Raw posted_at and scraped_at remain separate from trusted employer dates. See TIMESTAMP-AUDIT.md.
- The cold archive introduced in 2026 preserves expired records now, but does not establish pre-December retention. No personal-data backup restore was used for this job-only task.
- The comparison corpus has 12,877 US-included records, 1,962 unknown countries, 438 missing source URLs and 1,462 descriptions under 100 retained characters. Missing fields stay missing.
- Stable company/requisition IDs and confirmed listing URLs drive deduplication. Company spelling variants, subsidiary mirrors and reposts can remain unresolved. Individual requisitions are evidence units, not vacancy or hire counts.
- The job-only SQL queries 01–10 are included in query/. They use READ ONLY transactions, timeouts and UTC on the current Mini via its MagicDNS hostname. No production data/code was changed and no personal/account/billing data was selected. Full descriptions remain local scratch, outside the downloadable bundle.