Shape composition

A50Claim drafting + validity
B40Prior-art ranking
C20C-family landscape
D-irac10MPEP-grounded IRAC structure
D-mcq40MPEP-grounded multiple choice
D-oa10Office-action argument
E30Patent strategy free-text

deterministic structural judge (deferred) 200 rows · 7 shapes

Three-mode bracket results Measured on DeepSeek-R1-0528-Qwen3-8B Q5_K_M (llama.cpp, temp 0.6) · see methodology
Shape	Scorer	closed	retrieval	oracle
B Prior-art ranking	deterministic	0.018	0.000	0.017
D-irac MPEP-grounded IRAC structure	structural	1.000	1.000	1.000
D-mcq MPEP-grounded multiple choice	deterministic	0.625	0.850	0.950
overall Overall (90 scorer-supported rows)	deterministic	0.397	0.489	0.541

Sample rows

One representative row per shape — fetched from the HF dataset on publish, embedded in the manifest for offline reading. Truncated to fit; full text on HuggingFace.

A Claim drafting + validity judge

Question

Given this multilayer printed-board IPC-H abstract (stacked capacitive coupling layers — dielectric between power and ground planes), identify a single 'white-space' angle adjacent to the granted claim that would survive prior-art and § 103 scrutiny.

Oracle context

BIGPATENT abstract (IPC class H): A multilayer printed board comprising a plurality of capacitive coupling layers each consisting of a dielectric layer and a power supply layer…

Gold label

A strong white-space angle is a dynamically tunable version of the capacitive coupling layer in which the dielectric is a variable-permittivity material (e.g., barium strontium titanate) whose capacitance can be modulated in-situ by a control voltage — preserving the stacked-layer geometry while adding active tuning.

B Prior-art ranking deterministic

Question

You're preparing an invalidity search for EP2640036A1 claim 70699_1. Draft exactly five prior-art retrieval queries for the claim below, each targeting a distinct patentably-material element.

Oracle context

EPO claim 70699_1 (application EP2640036A1): The method of claim 6 further comprising receiving a list of functions of each of the plurality of devices…

Gold label

["device cooperation service information registration combined device function operation definition file", "predetermined device function list retrieval multi-device operation composition detection", …]

C C-family landscape judge

Question

Based on the IPC-H multilayer printed-board technology disclosed (stacked capacitive coupling layers with dielectric between power and ground planes), map the surrounding IPC-class landscape relevant for FTO and cross-licensing.

Oracle context

BIGPATENT abstract (IPC class H): A multilayer printed board comprising a plurality of capacitive coupling layers…

Gold label

The embedded-capacitance multilayer PCB landscape in IPC H05K clusters around three sub-segments: (1) discrete embedded capacitor insertion (Sanmina, TTM, Viasystems legacy), (2) planar/distributed embedded capacitance (3M C-Ply, Oak-Mitsui FaradFlex), and (3) active-tunable embedded reactive structures…

D-irac MPEP-grounded IRAC structure structural

Question

FACT PATTERN: Applicant claims a method of converting all sulfur compounds in industrial effluent to H₂S as an intermediate step before removal. Examiner issues a final § 103 rejection. Applicant offers a pre-filing declaration by Dr. Reyes (independent expert) expressing strong skepticism that the conversion approach would solve effluent remediation. Apply IRAC to evaluate whether this declaration constitutes sufficient evidence of nonobviousness under MPEP 716.05.

Oracle context

MPEP 716.05 Skepticism of Experts: 'Expressions of disbelief by experts constitute strong evidence of nonobviousness…'

Gold label

Issue / Rule / Application / Conclusion structure required. Issue: whether Dr. Reyes's pre-filing declaration constitutes sufficient evidence of nonobviousness. Rule: MPEP 716.05 — expert disbelief is strong evidence. Application: Reyes is independent and expressed strong, pre-filing skepticism about the very mechanism. Conclusion: declaration suffices to rebut prima facie obviousness.

D-mcq MPEP-grounded multiple choice deterministic

Question

Under MPEP 716.05, evidence of expert skepticism is primarily relevant to rebutting a prima facie case under which statutory basis for rejection? (A) § 101 — patent-eligible subject matter. (B) § 102 — novelty. (C) § 103 — nonobviousness, because expressions of disbelief by experts constitute strong evidence of nonobviousness. (D) § 112 — enablement.

Oracle context

MPEP 716.05 Skepticism of Experts: 'Expressions of disbelief by experts constitute strong evidence of nonobviousness…'

Gold label

D-oa Office-action argument judge

Question

Draft an office-action response for U.S. Application 18/241,774 (biodegradable hydrogel via cross-linking). Claim 1: cross-linked biodegradable hydrogel using a type-II photoinitiator. Examiner rejects under § 103 over Chen (type-I photoinitiator) + Patel (related cross-linker). Argue why the substitution lacks a reasonable expectation of success.

Oracle context

MPEP 716.05 Skepticism of Experts (cited in support of unexpected results / lack of reasonable expectation)…

Gold label

Applicant respectfully traverses the § 103 rejection over Chen/Patel because the examiner has not established a reasonable expectation of success in substituting a type II photoinitiator for Chen's type I photoinitiator — the photochemical mechanism is materially different (energy transfer vs. fragmentation), and Patel does not teach or suggest type-II behavior in this matrix…

E Patent strategy free-text judge

Question

Translate this patent abstract language into plain English for a PCB layout engineer: 'a plurality of capacitive coupling layers each consisting of a dielectric layer and a power supply layer and a ground layer arranged in stacked relation.'

Oracle context

BIGPATENT abstract (IPC class H): A multilayer printed board comprising a plurality of capacitive coupling layers…

Gold label

Each 'capacitive coupling layer' is a three-layer sandwich embedded inside the PCB stackup: a power plane and a ground plane face each other across a thin insulating (dielectric) layer. Because the two planes are conductors separated by an insulator at very close distance, the sandwich behaves like a built-in capacitor distributed across the board…

Sources

The bench is anchored to public corpora — row provenance is preserved per-row in source_metadata. The composition below shows how the 200 rows distribute across the seed sources.

USPTO MPEP 60
HPI-Naumann PatentMatch 40
BIGPATENT 100

USPTO MPEP: 2,047 MPEP subsections — 4,437 RAG chunks — seeds the D-* shape rows. Public-domain U.S. government work.
HPI-Naumann PatentMatch: 25,340 EPO claim ↔ prior-art pairs — seeds the B shape rows. CC-BY-4.0.
BIGPATENT: 1.3M U.S. patent abstracts — seeds the A, C, and E shape rows. CC-BY-4.0.

How to load

License: free · cc-by-4.0. Released as a HuggingFace dataset; available via the standard datasets library.

from datasets import load_dataset
ds = load_dataset("Orionfold/patent-strategist-bench-v0.1", split="train")
print(ds)  # 200 rows × 13 columns

# Filter to MPEP-grounded MCQs only:
mcq = ds.filter(lambda r: r["shape"] == "D-mcq")

Citation

@dataset{patent_strategist_bench_2026,
  title  = {Patent-Strategist Bench v0.1: a 200-question seven-shape benchmark for patent-prosecution reasoning},
  author = {Sehgal, Manav and Orionfold},
  year   = {2026},
  month  = {5},
  url    = {https://huggingface.co/datasets/Orionfold/patent-strategist-bench-v0.1},
  note   = {Companion to "Patent-Strategist v1 baseline on Spark", ainative.business/field-notes/patent-strategist-v1-baseline-on-spark/}
}

Companion methodology

This bench is the methodology artifact for the field note patent-strategist-v1-baseline-on-spark — the paired article walks through how the seven shapes were designed, how the three-mode bracket was scored, and what the headline finding means for the next fine-tuning cycle.

Read the methodology article