Enumerate every reading of every claim. 10⁵–10⁸ interpretations.
A human cabinet analyses two to five readings of a claim. The Claim Interpretation Lattice enumerates the entire combinatorial space — six axes, every jurisdiction, every jurisprudence era — and emits the wording with the lowest vulnerability score.
Module spec'd · MVP Q3 2026
Six axes. One lattice.
Every feature of every claim is read along all six axes simultaneously. The cartesian product is the lattice.
Literal
Open-ended vs partial-closed transitions, range endpoint conventions, 'about X' tolerance windows, wherein-clause semantics.
Doctrine of equivalents
Festo / Warner-Jenkinson insubstantial differences — bioisosteres, salts, polymorphs, isomers, isotopes, prodrugs.
Markush expansion
RDKit + bioisostere KG sub-genus enumeration. Patani & LaVoie, Meanwell — chemical knowledge graph traversal.
Jurisdictional
Phillips v. AWH (US Fed. Cir.), G 1/24 (EPO/UPC), purposive construction (UK), JPO restrictive reading, CNIPA variable.
Temporal / jurisprudence-era
Pre/post G 2/21 plausibility, G 1/23 non-reproducible prior art, In re Cellect ODP×PTA, USPTO Inventorship Nov 2025.
Range cascading
Broad → preferred → picture-claim cascade with DOE-extension envelopes per jurisdiction.
Anticipation
Art. 54 CBE / 35 USC §102. XGBoost + LegalBERT calibrated on ~50 000 PTAB IPR + EPO opposition decisions.
Obviousness
KSR v. Teleflex + EPO problem-solution. Combination graph + secondary indicia weighting.
Sufficiency / enablement
Wands factors + Amgen v. Sanofi (598 U.S. 594, 2023). Coverage of claim scope by representative embodiments.
Plausibility
G 2/21 (EBA EPO, 23 March 2023). Is the technical effect 'embraced and embodied' by the application as filed?
Six modules, sequential.
- 01
Claim Parser
spaCy + LegalBERT/PatentBERT decompose the claim into atomic features (F1, F2, …) with quantifiers, ranges, Markush groups.
- 02
Interpretation Generator
Six-axis cartesian product. 10⁵–10⁸ configurations per claim, parallelised across the corpus.
- 03
Prior-Art Cross-Reference
Espacenet + USPTO + PATSTAT + PubChem + ChEMBL + Reaxys — full-text + structure + semantic search.
- 04
Validity Scorer
4-vector score per (interpretation × prior art): anticipation, obviousness, sufficiency, plausibility. Calibrated probabilities.
- 05
Robustness Optimizer
Constrained optimisation — minimise mean vulnerability, subject to scope and clarity constraints. Gurobi / OR-Tools / RL.
- 06
Report
PDF executive summary + interactive dashboard + JSON for the deposit pipeline. 3-5 candidate claims, each with a robustness score.
Module 1 — claim parser, in the open.
A pharmaceutical composition claim, parsed into atomic features F1–F4 with ranges, Markush groups, and explicit ambiguities flagged for the lattice generator.
A pharmaceutical composition comprising at least one
polymer P selected from the group consisting of P1, P2,
P3, and at least one active pharmaceutical ingredient
(API) X, wherein the composition exhibits a glass
transition temperature in the range of about 50-100°C
and a dissolution rate of at least 20% per minute at
pH 6.8.{
"claim_type": "composition",
"preamble": "pharmaceutical composition",
"transition": "comprising",
"transition_type": "open-ended",
"features": [
{
"id": "F1",
"type": "component",
"label": "polymer P",
"quantifier": "at least one",
"constraint": "selected from the group",
"markush": ["P1", "P2", "P3"],
"markush_type": "closed_list"
},
{
"id": "F2",
"type": "component",
"label": "API X",
"quantifier": "at least one",
"abbreviation": "active pharmaceutical ingredient"
},
{
"id": "F3",
"type": "property",
"label": "glass transition temperature",
"abbreviation": "Tg",
"unit": "°C",
"range": {
"min": 50, "max": 100,
"qualifier": "about",
"endpoints_inclusive": true
}
},
{
"id": "F4",
"type": "property",
"label": "dissolution rate",
"unit": "%/min",
"constraint": "at least",
"value": 20,
"conditions": { "pH": 6.8 }
}
],
"ambiguities": [
"'polymer P' — depends on Definitions section",
"'about' tolerance — assumed ±10% per pack convention"
]
}Between the landscape and the doctrine.
CIL consumes IP Radar's landscape and feeds Doctrinal Hardening with optimised claims for screening.
Stop drafting blind.
The lattice ships in Q3 2026. Lock the Scale tier now to be on the launch list.