One disclosure, from intake to score.
Every field of the intake is followed to the records it resolves to, the context found around it, the nine measures and the score. Nothing is invented between the layers.
- 1Intake
One office record. For Cancer T-cell therapy, a University of Pennsylvania disclosure dated 2011-12-09, that is a title, nine keywords, one paper, one patent, three registered trials, five inventors and a grant.
Nothing else is needed to start. - 2Resolved records
Each field is checked against the register that can confirm it. The paper resolves to a 2011 article in The New England Journal of Medicine, the patent to US 8,916,381 with 28 claims, the trials to their phase and status, and the funding to $6.6 million across four NIH grants.
A field that does not resolve stays empty and is reported empty. - 3The context
Around the disclosure: 73 registered trials in the same use, the companies working on it, and products already approved. This layer says whether an idea is early, crowded or late.
Anything dated after 2011-12-09 is context only. The FDA approval of the licensed product in August 2017 is shown, never scored. - 4Nine measures
Each reading is ranked among earlier disclosures with the same reading. Its 28 claims sit in the 86th percentile, its 70 cited earlier patents in the 96th. Readings average into nine measures.
IP defensibility 0.49, inventor alignment 0.95, translational funding 0.82, market signal 0.95: the four with weights learnt from history. - 5One score
The fitted weights turn the measures into one deal chance: 5.7% for a licence or startup within about three years, the 59th percentile among 48,991 earlier records.
The demonstration rule says Review. Every number traces back down the layers to its record.
Five layers, and nothing invented between them.
Intake, then resolution
Each field of the intake record is queried against the source family that can confirm it. A field that resolves becomes a node with its own identifier; a field that does not stays empty and is reported empty.
The context around it
Competing trials in the same indication, companies working on it, and any product already approved. This is the layer that decides whether a disclosure is early, crowded or late.
Dimensions, then the composite
Readings are placed among their peers, graded 0 to 5, combined by weight into nine dimensions, and reduced to one number with a tier and its gates.
Eight stages from a JSON record to one shared graph.
What the repository does today, in order, when a batch of disclosures goes in. Each stage writes nodes; nothing is scored until the graph is complete.
- 01Intake and validation
Each record is checked against the intake model before any network call.
- 02Publications and authors
OpenAlex for works, venues and authors; a local journal impact table by ISSN.
- 03Federal grants
USASpending and NIH RePORTER, for award, funder, amount and investigator.
- 04Clinical trials
ClinicalTrials.gov, for phase, status, sponsor and enrolment.
- 05Patents and challenges
The USPTO Open Data Portal for the file, PTAB for validity challenges, bulk data for claims and citations.
- 06Who is who
Inventors and assignees resolved in order, so the same person on two records becomes one node.
- 07Career history
An inventor's earlier patents are pulled and tagged as history, kept apart from the disclosure's own.
- 08One graph
Records that share an institution or a paper join, so the second disclosure starts where the first left off.
Three phases, each unlocked by a constraint.
We run it for you
We operate the pipeline over your existing backlog and return scored assessments. Nothing to install, no procurement, no security review.
Your own instance
A dedicated, isolated deployment for your institution, operated by us. Your disclosures sit under your rules. Isolation is what makes the first contract possible.
Comparison across institutions
Once three or more institutions are live, each can see how a disclosure compares with similar ones elsewhere, anonymised. This cannot be bought early, only earned.
What an office receives
A report card
The composite, its tier, the nine dimensions with their readings, the rule that combined them, the one question to resolve first, and every source with the date it was read.
A ranked backlog
Every disclosure in the batch scored the same way, sortable by tier, dimension or gap, with the ones held back by a single missing fact listed separately.
Your own record of results
As disclosures resolve, the office's own outcomes join its instance under its rules, and the scores are checked against them. Nothing leaves the instance but distributions.