Prediction scorecard
Every falsifiable call from “What We’re Watching”, with its deadline, the early signal to look for and what would prove it wrong. Misses are scored as openly as hits.
Each issue ends with one prediction. It goes on this page with the deadline pinned to a date, and it is scored when the deadline passes or the question resolves, whichever comes first. The wording here is a one-line summary; the issue link has the original. Entries are never removed or edited after the fact. The page is reviewed on the first Tuesday of every month.
8 Open0 Hit0 Partly hit0 Miss
-
At least one top-20 pharma company joins Biohub’s Virtual Biology Initiative as a named funding partner.
- Early signal
- A drugmaker contributes perturbation screens or joins the initiative’s data-standards work, or Biohub creates an industry membership tier.
- Wrong if
- The funder list still has no pharma company by mid-2027, and pharma keeps paying for proprietary perturbation data instead, through deals like GSK–Relation or startups like Rivercell.
-
Xaira publishes a preprint or peer-reviewed paper reporting X-Design hit and progressability rates across at least five targets, rather than two case studies.
- Early signal
- A bioRxiv submission or conference abstract from the X-Design team, most plausibly at an antibody engineering meeting in the first quarter.
- Wrong if
- A partnership with a large pharma is announced first, with no accompanying data release.
-
At least one top-20 pharma publicly announces a dedicated in-house rapid-validation wet lab, scoped to confirming machine-generated hypotheses rather than running a therapeutic programme.
- Early signal
- Job postings for molecular biology and automation staff reporting into a computational or data-science organisation, or an AI-platform company signing a per-hypothesis validation-capacity deal with a CRO.
- Wrong if
- A published discovery-agent result reproduces cleanly across independent reruns, so validation stops being the bottleneck.
-
The European Commission’s November health-AI package includes an explicit mechanism for verified access to frontier bio-capable models for European researchers, either an EU verification scheme or access negotiated with US frontier labs.
- Early signal
- The Commission’s meeting with frontier labs is scheduled with life sciences on the agenda, or AI Office or DG SANTE consultation language mentions “trusted access” or “verification” for biological capabilities.
- Wrong if
- The package sticks to funding, compute, data spaces (EHDS) and sandboxes, with no access-governance element for frontier bio models, and no such mechanism appears by year-end.
-
FDA finalises its January 2025 draft guidance on AI for drug and biologic regulatory decision-making as a cross-agency framework tied to the new Deputy Commissioner for Technology and AI.
- Early signal
- The guidance appears on CDER’s 2027 guidance agenda or is named as a priority by the deputy commissioner before year-end, or the final text moves toward QMS-style requirements such as change control and lifecycle monitoring.
- Wrong if
- It is still in draft on March 31, 2027, or is finalised with no reference to the deputy commissioner’s office or any cross-centre scope.
-
At least two more top-20 pharmas license an agentic research or regulatory platform on undisclosed terms, and at least one of them also runs a large milestone-based AI discovery partnership.
- Early signal
- A press release naming a therapeutic area and a programme count but no upfront, milestones or royalties, or an EU pharma publishing an AI risk classification covering a licensed third-party agent.
- Wrong if
- Agentic platform deals arrive with disclosed values and milestone structures instead, or one of the three current K Pro customers declines to expand or renew.
-
At least one more FDA warning letter or Form 483 cites AI-generated content in GMP documentation, and at least one names an agentic or LLM-based system by function.
- Early signal
- A citation under 21 CFR 211.22(c) or 211.100 whose observation describes who reviewed the AI output and against what criteria, not just that AI was used.
- Wrong if
- EMA finalises Annex 22 with a workable pathway for generative models, FDA harmonises and enforcement stays quiet through Q1 2027, or the next citations land in discovery or clinical rather than manufacturing.
-
The first EMA or FDA action against a large sponsor, not a small contract manufacturer, finding that AI-generated regulatory documentation was reviewed by people without the depth to catch what was missing, most likely as a Form 483 observation.
- Early signal
- Large pharma hiring reviewers and quality staff faster than it announces AI efficiency programmes.
- Wrong if
- No such action within twelve months, with review and quality headcount still falling while agent deployments scale.