How does ORION handle biotech evidence?

As documented on 29 September 2026, ORION separates four evidence timestamps for selected biotech disclosures and study records. The reviewed source histories retain original issuer and ClinicalTrials.gov links; a missing timestamp is UNAVAILABLE, not an inferred date. Selected records do not establish complete company or registry coverage.

Last updated 11 Oct 2026 · Company promise record definitions, human review and corrections added; this is not a fresh source or coverage check.

What do the four timestamps mean?

Source published — published_at: The date or time the original source published that statement or record version. This is not the time ORION discovered it.
Event effective — effective_at: The date or time a stated event took effect, when the source establishes it. A planned date or estimated study completion date does not establish that an event occurred.
ORION first seen — first_seen_at: The first supported ORION observation of that record or version. Importing an older release cannot make this timestamp equal to its original publication date.
ORION ingested — ingested_at: When that record entered the ORION evidence store. A later source check or page edit does not establish an earlier ingestion time.

Date precision is preserved: a source that provides only a month or day does not establish an exact timestamp. A page’s Last updated date describes the page; a source-check date describes a check. Neither substitutes for these four record timestamps.

What does UNAVAILABLE mean?

The displayed evidence does not establish that value. Missing dates, source failures and unresolved identities stay explicit; they do not become zero, “no event,” a forecast, or an investment conclusion.

A failed or partial collection cannot establish a clean no-change result. A first observation establishes current state only; a before-and-after change needs two supported observations of the same identified record. A registered estimated completion date is not a promised announcement date.

Can ORION reconstruct what was known on a past date?

Source-time reconstruction arranges retained source statements by their own publication dates. Selected reviewed histories can support this limited retrospective view. Reading an older statement today does not show that ORION knew it then.

ORION-known reconstruction would require retained first-seen and ingestion evidence for the exact records available to ORION at a chosen time. A general replay of what ORION knew at any arbitrary past date is not currently provided. Missing observation history remains UNAVAILABLE; historical source dates cannot fill it.

What can the public pages establish today?

Changes distinguishes reviewed registry changes from historical examples and labels the observation window. Weekly preserves each issue’s publication and source-check dates. A company page describes its selected evidence, not a complete pipeline audit.

The Stale Clock report is published as a dated registry observation; our Stale Clock report measures registry dates; ORION's records track what companies said. A public data API is not currently provided. A public Proof service is planned: /proof shows its current state and lists no day that was not actually published.

ORION does not calculate investment scores, rankings, recommendations, target prices, or expected returns.

What is a company promise record?

A company names an action for a program and a time window in an SEC filing. A company promise record keeps that original sentence, its stated window, its filing date and its SEC link; every later restatement of the same program and action; and the disclosure that was actually found, each with its own filing and date. Certainty words such as “on track” are kept exactly as written. A statement without a dated window is not added as a promise.

Each record page states which SEC filings were read and through what date, which company releases were read and on what date, and the date the founder compared the displayed rows with the originals. A through-date describes the filings read; it is not a claim that every announcement or investor slide was assessed. Original filing dates stay separate from the date ORION checked the filings and from the status date.

What do the seven status words mean?

Company timing wording is retained as stated. Registry completion estimates and company readout timing are separate facts.

Window open

The company-stated timing window remains open as of the stated check date. This status does not predict whether the action will occur.

Disclosure found

A source disclosing the specified event was found within the checked scope. This does not indicate clinical success or failure.

Timing changed

A later company statement changes the stated timing for the same program and action. The earlier and later wording and sources are retained.

Withdrawn or stopped

The company explicitly states that the specified action has been withdrawn or stopped. This is not inferred from silence.

Window ended, no disclosure found

The stated window has ended and no disclosure of the specified event was found in the documented scope. This status requires founder review of the company news, checked sources and date; it is not an all-source absence claim.

Not mentioned in later filings

The specified statement was not found in the selected later filings. This does not establish cancellation.

Unable to determine

The checked material is insufficient to determine the requested timing or status within the documented scope.

How are rows checked?

Each row on a public record page carries one of two labels, shown on the row itself. Founder-reviewed rows and machine-verified rows are displayed separately and are never counted together.

Checked by founder — compared by hand with the SEC original by ORION's founder, Jiwoo Seong (Woo). The page states the date of that comparison and how many rows are shown out of how many rows exist for the company. Rows on hold, rows with a mismatch, and rows the founder cannot verify are not published with this label.
Quote matched to filing (not yet reviewed by founder) — found in an SEC filing by ORION's extraction engine. The quoted wording was matched character for character against that filing; a row whose quote does not match is not written. The founder has not reviewed the row, so the label does not mean a person checked its timing reading or status.

An assistant may prepare a candidate row; an assistant's comparison is not counted as a human check.

What does the ledger not do?

No predictions, scores, rankings or comparisons between companies. No investment advice. No judgement words: a changed window is recorded as a changed window, and a statement not found in the filings read is recorded as not found, not as a stopped program or a broken promise. The site presents the reviewed copy; it does not calculate new promise outcomes in the browser.

How are corrections made?

Report an error to hello@orionbiotechai.com. The founder checks the original. If the record is wrong, the correction is added to that record page with its date, what changed, why, and the evidence. The earlier version is kept; a correction is an addition, not a deletion, and the correction history stays on the page.

How should the earlier research-workspace documentation be read?

The retained notes below describe the earlier research workspace. Their scope is separate from current selected ORION records; retaining them is not a new test run or verification of current coverage. The company-level audit keeps its original 3 September 2026 date.

Earlier research-workspace methodology and dated coverage audit

1. Identity resolution

Company, ticker, sponsor, asset, brand, and development-code aliases are normalized. A relationship is shown only with its verification state; unresolved identities remain unresolved.

2. Source collection

ClinicalTrials.gov, SEC filings, regulatory records, issuer disclosures, and available market records are collected independently. Each fact retains record identifiers, source links, and retrieval timestamps where available.

3. Claim rendering

Source-backed facts, deterministic calculations, and model-generated synthesis are labeled separately. Unsupported narrative claims are blocked. Conflicting source values are preserved rather than silently reconciled.

4. Coverage and confidence

Coverage describes the presence of evidence across report domains. Relationship confidence describes source-match strength. Neither measure represents investment attractiveness or future performance.

5. Missing data

Missing values are displayed with reason codes and suggested follow-up identifiers. They are never converted to a favorable or unfavorable investment conclusion.

6. Automated release verification

A fixed 25-case cohort is rerun against the same analysis path used by the product. Each run records canonical company, ticker, entity confidence, returned trial count, terminal status, and error code. The release check fails when company attribution or ticker changes unexpectedly, entity confidence drops, a trial count falls below its reviewed floor or by more than half, or a previously working case changes to ENTITY_NOT_FOUND, ENTITY_RESOLUTION_FAILED, or SOURCE_UNAVAILABLE. Baselines are updated only after a reviewed successful run; a live-source outage is reported as an operational failure rather than silently accepted as a new baseline.

7. Known limitations

ClinicalTrials.gov records can name lead sponsors, collaborators, intervention aliases, and legal entities differently. ORION queries sponsor and intervention paths and deduplicates by NCT ID, but registry updates, renamed sponsors, newly added aliases, and upstream availability can still change coverage. Trial count is a coverage diagnostic, not a measure of drug quality or probability of success.

FDA CDER/CBER coverage boundary
Drugs@FDA and Orange Book cover primarily CDER-regulated NDA and ANDA records. Biologics licensed through the BLA pathway, including many antibodies, vaccines, and cell or gene therapies, require separate CBER or Purple Book coverage. Company-level FDA coverage can therefore be lower than the full regulatory record.

Company-level biologics coverage audit · 2026-09-03
Three of the 14 audited companies have a structural BLA coverage gap in the current Drugs@FDA and Orange Book workspace. AMGN has 21 BLA records among 27 directly observed applications (77.8%). REGN has nine BLA records among nine observed applications (100%) and no Orange Book connection. MRNA returns no applicant record from the Drugs@FDA query and its previously verified products are handled through the BLA/CBER pathway; it also has no Orange Book connection. These are source-scope limitations, not evidence that no FDA record exists. Companies with a mixture of NDA/ANDA and BLA records remain explicitly partial rather than being described as fully covered.

Can I request a data sample?

Tell ORION which companies, fields or format you would like in a sample. The founder reads each request by hand.

Or email ORION.