How this edition was measured.

Which date controls the calculation?

Source dataTimestamp: 2026-09-28T12:00:05. Calculation reference (all four metrics): 31 Aug 2026. Source calendar date: 28 Sep 2026; source timezone: UNAVAILABLE.

All four metrics use 2026-08-31, the previous calendar month end. These are the fields observed in the snapshot on 2026-09-29, evaluated using that cutoff; this is not a reconstruction of the registry state on 2026-08-31.

The source timestamp is 2026-09-28T12:00:05. Its written calendar date, 2026-09-28, and timezone availability are preserved separately from the calculation reference. No offset is invented. Actual retrieval timestamps are recorded in UTC.

Actual page retrievals and completed ingestion were recorded separately in UTC, between 29 Sep 2026, 09:45 UTC and 29 Sep 2026, 09:46 UTC. The edition date is 2026-09-29, the completed observation date.

Which records are in the denominator?

The target population contains 3,762 unique registry records with an INDUSTRY lead sponsor, a PHASE3 label (including a labelled Phase 2/3 combination), and one of these statuses: RECRUITING, ACTIVE_NOT_RECRUITING, NOT_YET_RECRUITING, or ENROLLING_BY_INVITATION.

This global reference population is separate from ORION’s selected-study monitoring roster.

The official query is fully paginated with countTotal=true. Each record independently must meet the same phase, sponsor class and status filters.

https://clinicaltrials.gov/api/v2/studies?format=json&pageSize=100&countTotal=true&filter.overallStatus=RECRUITING,ACTIVE_NOT_RECRUITING,NOT_YET_RECRUITING,ENROLLING_BY_INVITATION&filter.advanced=AREA[Phase]PHASE3+AND+AREA[LeadSponsorClass]INDUSTRY

Open the original official query.

How are DAY and MONTH dates handled?

A DAY value, such as 2026-08-31, has the same start and end day. A MONTH value, such as 2026-08, represents the entire calendar month: 2026-08-01 through 2026-08-31. An interval qualifies only when its last day satisfies the cutoff.

YEAR-only values, missing values, invalid calendar dates, declared-precision mismatches and missing or invalid ESTIMATED/ACTUAL basis are counted separately. Unavailable values do not acquire an assumed date.

Calendar subtraction clamps a leap-day reference to February 28 in a non-leap year. It uses calendar months and years rather than a fixed duration.

What are the exact A, B, C and D rules?

All four metrics use interval end ≤ reference date. A and D use the shared reference date, 2026-08-31; B and C use that same inclusive comparison with their stated 12-calendar-month and one-calendar-year thresholds, 2025-08-31 and 2025-08-31.

A · 377 / 3,762

Primary completion basis is ESTIMATED and the whole date interval ends on or before the shared calculation reference, 2026-08-31, inclusive. Denominator: every target record.

B · 226 / 377

An A record whose entire last-update-posted interval ends on or before reference minus 12 calendar months, 2025-08-31, inclusive. Denominator: A records.

C · 82 / 377

An A record whose entire estimated primary completion interval ends on or before reference minus one calendar year, 2025-08-31, inclusive. Denominator: A records.

D · 331 / 552

A NOT_YET_RECRUITING record whose start basis is ESTIMATED and whose entire start interval ends on or before 2026-08-31. Denominator: all target NOT_YET_RECRUITING records, including unavailable dates.

Elapsed days in the CSV and table equal calculation reference date minus the registered primary completion DAY. MONTH rows show UNAVAILABLE for elapsed days.

How are unknown fields and filter errors reported?

Each row below describes one date field across all 3,762 target records. Missing basis is distinct from a missing date. Parsed DAY/MONTH dates can still have an unavailable basis and will not qualify an estimated-date test.

Source-filter mismatches: 0. Duplicate NCT IDs: 0. Registered sponsor name unavailable: 0.

D has its own 552-record date-quality denominator: missing 0, invalid 0, YEAR-only 0, basis missing 0, basis invalid 0.

How was source completeness checked?

The official version endpoint was read before and after pagination. Both original dataTimestamp and API version 2.0.5 matched. Every returned NCT identity was unique, the final count equalled the advertised total, and all 38 pages completed.

Every page has a retained response-byte hash, checked-at timestamp and projected-record hash. Response hashes describe the bytes received; projected hashes describe the fixed saved fields. They have different scopes.

No source failure is substituted with a zero count. An incomplete page sequence, changing source version, duplicate identity or filter mismatch prevents a edition release.

Official version endpoint · Official studies query.

Which times does a record retain?

• effective_at: UNAVAILABLE, because this is a current-state snapshot rather than an event occurrence.
• published_at: no precise instant is assigned. The last-update-posted source date and its interval are retained separately.
• first_seen_at: the actual UTC retrieval time of that API page.
• ingested_at: the actual UTC time when the complete edition measurement finished.

The calculation reference and source calendar date are separate from these observation times. This measurement does not reconstruct earlier registry versions.

How can the exact table and CSV be verified?

The HTML table and CSV enumerate the same 377 A records, in ascending NCT order. The CSV adds the full interval endpoints, B/C membership and observation timestamps. Missing cells use UNAVAILABLE.

CSV encoding is UTF-8; record delimiters are LF. Registry text that could be interpreted as a spreadsheet formula receives a leading apostrophe in CSV only. Registered sponsor names remain unchanged in the original edition projection and the escaped HTML.

Download CSV · Download checksum.

CSV SHA-256be429676b07fa0bd6077311c8368110b79bf43aadc9806c8d58fbf84a821cff5

Original edition report SHA-256d77fe4e6820806385a7fc68261b5aa75b67fa2e494b28250c652c4722238e7dc

Projected source SHA-256 in original page order5d420019944a719e164dbc5dae913f1df9529f98b614ba829bff822af2081478

What can this observation establish?

The checked record still lists this estimated date. The registered dates are not promises of results announcements. The snapshot establishes the stated source fields at the recorded check; it does not establish actual completion or a company-level conclusion.

Claude reviewed the frozen CSV, calculations and neutral presentation before publication. This review does not independently validate every source field. This report creates no new email consent or sending schedule.

Return to the edition report · Read existing Weekly editions.