Public methodology
What the field is paying attention to.
AI relevance and independent attention shape the order; direct evidence determines what can be stated as fact.
Editorial scope
Both AI and biology must be material.
Core coverage includes biological foundation models, molecular and protein design, genomics, single-cell and spatial biology, virtual cells, pathology, scientific agents, autonomous laboratories, clinical development, biomarkers, biological datasets, bioengineering, and directly relevant policy.
General AI, broad biotechnology, medical software, or health news enters scope only when the connection to biological research, drug development, diagnostics, clinical evidence, or scientific infrastructure is substantive. Promotional essays, rumors, low-value rewrites, and advice are excluded by default.
Source policy
Primary source first.
Sources are classified by type and authority. A paper, filing, regulator notice, repository release, or official announcement is presented before reporting and commentary when it is available. Independent reporting is valuable context, not a substitute for the underlying record.
- Source collection must respect terms, robots directives, rate limits, and attribution requirements.
- Full articles are not republished and paywalls are never bypassed.
- Full text is retained only when licensing permits it; otherwise the system uses metadata and permitted excerpts.
- Unavailable or inadequate source material produces uncertainty or rejection, not filled-in facts.
Editorial path
Select, verify, cluster, then explain.
- Identify a development with a material connection to both AI and biology.
- Locate the direct record and useful independent or technical context.
- Normalize identifiers, dates, organizations, authors, and links so related coverage can be compared.
- Remove duplicates and group reports only when their identifiers, organizations, event type, timing, and wording consistently describe the same development. Uncertain matches stay separate for review.
- Draft an explanation grounded in the selected evidence.
- Check claims, dates, amounts, identifiers, links, evidence labels, and stated uncertainty.
- Leave out claims that cannot be supported by the available record.
Event clustering
One event, one evolving page.
Exact identifiers—including DOI, arXiv ID, canonical URL, and release tag—identify direct duplicates. Title similarity, shared organizations and links, compatible event types, and date proximity help determine whether separate reports describe the same development.
Papers and releases remain connected through persistent identifiers. Deals, trials, and regulatory events may evolve over longer periods, so later material can update an existing event page rather than create a duplicate story. Unrelated developments are separated when the evidence shows they do not belong together.
Ranking
AI relevance first, then independent attention.
| Factor | How it affects selection |
|---|---|
| AI relevance | A clear, substantive AI connection is required for the lead positions. |
| Coverage breadth | Distinct independent outlets covering the same event increase its prominence. |
| Coverage velocity | Several independent reports arriving in the past 12 hours signal a story that is accelerating. |
| Original reporting | A verified exclusive receives a bounded lift; syndicated copies still count as one origin. |
| Editorial placement | A homepage lead, section lead, or journal feature carries more weight than a passing mention. |
| Scientific prominence | Cover and featured papers in leading journals receive a bounded lift. |
| Public discussion | Substantive discussion from distinct, identifiable voices can raise an item's rank. |
| Deal materiality | Larger financings receive a bounded lift. For partnerships and licenses, verified upfront cash matters far more than contingent headline milestones. |
| Consequence | Priority rises when the development could change research, clinical, company, or capital decisions. |
| Recency | New developments generally outrank older context unless the older item is more consequential. |
| Evidence | Direct records and independent reporting govern what can be stated, regardless of popularity. |
Repeated coverage is deduplicated by publisher ownership and syndication origin. Publication time—not the time a page was discovered—drives the coverage-velocity signal. Social totals never establish a claim or count as independent coverage. For an exact, source-linked AI × biology post, bounded engagement velocity may help identify unusual attention, preferably against that account's typical same-age posts. When a source or social channel has not been observed, that absence is treated as unknown rather than as zero attention. Deal values are compared only when the currency and amount are verified; the score uses a bounded curve so money cannot overwhelm AI relevance or independent attention. Unverified or purely contingent “biobucks” receive little weight. Exact source timestamps drive recency when available; date-only records remain date-level rather than receiving an invented clock time.
AI assistance
Useful for organization, never a substitute for evidence.
AI assistance may classify material, connect related coverage, and structure a draft. Source links and conservative language remain mandatory, especially for numbers, dates, outcomes, and scientific claims.
The system must not invent sample sizes, results, investors, amounts, dates, datasets, benchmarks, or availability. It does not convert correlation into causation, present a preprint as peer reviewed, or turn an author or company interpretation into established fact.
Social discussion
A window into the conversation, not evidence.
“Notable on X” considers public posts from the rolling 36 hours before an edition is published. A small selection is chosen for substantive analysis, informed reaction, or a useful technical link—not for likes, reposts, or other mutable engagement counts. Relevant author roles, affiliations, and interests are disclosed.
Each item keeps the exact public status URL and posting time, then appears as a native link rather than an embedded X widget. Engagement metrics are neither stored nor displayed. A post can add context, but it cannot substantiate a news claim; the underlying paper, filing, announcement, registry, or reporting must do that work. Posts that become unavailable are hidden.
Public discussion appears only when the exact post, time, and author context can be verified. It remains commentary rather than evidence for the reported news.
Accountability
Corrections and limits stay visible.
Material changes record the original publication time, correction time, what changed, why, and whether the change affected the headline or conclusion. Relevant employment, investment, advisory, family, or other relationships are attached to affected entities and displayed on their coverage.
Deals coverage reflects verified public announcements and filings but is not exhaustive. Company-reported values remain attributed, offering amounts are not silently converted into money raised, and undisclosed terms stay undisclosed. Selected social commentary is contextual, not an endorsement or a comprehensive account of discussion.