Office Action Analysis — App 19334772 (public record)

Full Analysis

Office Action Response Analysis · Final (CTFR)

App. No. 19/334,772

Art Unit: 2128
Examiner: STORK, KYLE R
Mailed: 06/23/2026
Response period stated in the OA: “3 MONTHS FROM THE MAILING DATE OF THIS COMMUNICATION”
Rejections: §103 ×2
Claims: 20 rejected
Generated: Aug 10, 2026

Final rejection — after final, amendments are not entered as of right (37 CFR 1.116); response options include an after-final response, AFCP 2.0, an RCE, or appeal.

The record presents a mix of one comeback-resistant gap resting on the examiner's own admission (arg 2, though it may be cured by a new reference and sits in a dependent claim) and a crux argument covering all three independent claims (arg 1) whose strength cannot be fixed until the truncated Khan paragraphs (¶¶0040-0041, 0050) and the complete claim-5 mapping are obtained — considerations that favor confirming the record before committing to a pure-argument posture. Counsel may weigh that arguments 3 and 4 test weakly against established §2143/§2141.01(a) doctrine and are better used as reinforcement than as lead points, and that the teaching-away point (arg 5) turns entirely on scope-matching the Wonus quote to the mapped limitation. Where the verify-first items (args 1 and 6) resolve unfavorably after record confirmation, an amend-and-argue posture — tying the add/exclude and temporal-graph features into the independent claims to force the combination gaps — is a path for counsel to consider alongside continued argument, and this examiner's documented interview propensity may inform how these considerations are best surfaced.

Examiner Kyle Stork (AU 2128): allowance rate 44% (n=59); avg 3.54 OAs to allowance; interviews held in 58% of cases, and when an interview was held allowance followed 59% of the time (correlation, not causation); RCE filed in 59% of cases. Based on n=97 applications; USPTO public data, 2016-01-01..2022-12-31. Correlational — it informs, it never decides.

Generated on a published USPTO office action — no confidential disclosure involved. First-pass analysis for attorney review — not a drafted response.

Indicated Allowable Subject Matter & Examiner Interview

Examiner interview (MPEP 713) — a consideration. The strongest candidate arguments below are close calls (see the likely examiner responses in the Argument Bank), so an examiner interview to test the arguments and probe what would put the case in condition for allowance may be worth weighing before filing a written response.

Per-Claim Strategy

An at-a-glance recommendation per rejected claim, composed deterministically from the analysis below. A triage summary for counsel to weigh, not a decision.

The application stands under a final rejection — amendments are not entered as of right (37 CFR 1.116); see the finality note above.

Claim	Rejections	Recommended path	Basis	Fallback amendment	Confidence
Claim 1	§103 (obviousness)	Argue	Mischaracterized reference (#3)	—	high
Claims 2, 3	§103 (obviousness)	Review — no argument identified	—	—	low
Claim 4	§103 (obviousness)	Argue	Missing element (#1)	—	moderate
Claim 5	§103 (obviousness)	ArgueStrategy check: re-ranked — For claim 5 specifically this self-contained missing-element gap is more dispositive than the inherited (currently higher-effective) Khan attack, subject only to the evidentiary caveat that the office action was truncated before the claim-5 mapping — counsel must confirm against the complete action.	Missing element (#2)	—	moderate
Claims 6–9	§103 (obviousness)	Review — no argument identified	—	—	low
Claim 10	§103 (obviousness)	Argue	Mischaracterized reference (#3)	—	moderate
Claims 11, 12	§103 (obviousness)	Review — no argument identified	—	—	low
Claim 13	§103 (obviousness)	Review — no argument identifiedStrategy check: re-ranked — Rank 2's temporal-graph gap applies equally to claim 13 (which recites the identical limitation as the CRM parallel of claim 4) but the argument bank tags Rank 2 only to claim 4; this examiner-admitted, strong-survival gap should be the top path for claim 13 as well.	—	—	low
Claim 14	§103 (obviousness)	ArgueStrategy check: re-ranked — For claim 14 (CRM parallel of claim 5) this self-contained missing-element gap is more dispositive than the inherited Khan attack, subject to the caveat that the office action was truncated before the claim-5/14 mapping and must be confirmed.	Missing element (#2)	—	low
Claims 15–18	§103 (obviousness)	Review — no argument identified	—	—	low
Claim 19	§103 (obviousness)	Argue	Mischaracterized reference (#3)	—	moderate
Claim 20	§103 (obviousness)	Review — no argument identified	—	—	low

Argument Bank

Candidate arguments for counsel, ranked strongest-first — brainstorming inputs for counsel to evaluate, not a drafted response.

No asserted reference reaches the temporal dependency graph of claim 4 (examiner's own admission)

Missing elementClaim 4Rebuts: §103 rejection of claims 1, 2, 3, 4, 5, 10, 11, 12, 13, 14, 19, 20

generate a temporal dependency graph consistent with the time-series data, wherein the temporal dependency graph identifies at least one of a temporal dependency constraint, a sequence progression constraint, or a time-based correlation

The office action itself states that Kursun 'fails to specifically disclose' the temporal dependency graph and its temporal/sequence/time-based constraints, and relies on Dang only for a generic 'dependency graph.' Both Kursun and Dang are fully grounded. Per OA2, Dang's available text is directed entirely to symptom-disease conditional probabilities computed via a TF-IDF/Bayesian process and a Gini-index inquiry step, and does not address temporal sequencing or time-based data. On the grounded record, neither asserted reference supplies a temporal dependency constraint, a sequence progression constraint, or a time-based correlation, leaving a combination gap on claim 4 for counsel to weigh.

Office action (Rejection 1, claim 4): 'Kursun fails to specifically disclose: generate a temporal dependency graph consistent with the time-series data ... wherein the temporal dependency graph identifies at least one of a temporal dependency constraint, a sequence progression constraint, or a time-based correlation'
OA2 (Reference 3): Dang's available text 'is directed entirely to symptom-disease relationships and conditional-probability edges; it does not, in the portion provided, address temporal sequencing or time-based data'
Dang claim 9: inference 'based on hyponymy relationships ... and ... a confidence value' — no temporal element

MPEP § 2131 / § 2143.01 — every limitation must be supported by the cited art; a §103 combination cannot rest on a limitation absent from all asserted references

Risk The examiner may map Dang's knowledge-graph structure or Kursun's time-stamped historical data (paragraph the examiner cites for claim 4 timestamps) to the temporal graph; counsel should confirm the Dang 'dependency graph' passage the examiner relies on actually identifies a temporal/sequence/time-based constraint and note that Dang's conditional-probability edges are not temporal.

Likely examiner response✓ survives — strong

The examiner would lean on the claim's 'at least one of' structure: the limitation is satisfied by any single alternative, so the examiner need only map a time-based correlation OR a sequence progression constraint OR a temporal dependency constraint. The examiner could argue Dang's conditional-probability/hyponymy edges, or Kursun's iterative pattern analysis, reasonably read on one alternative under BRI, or — consistent with this examiner's documented interview propensity and multi-OA path to allowance — issue a new ground citing an additional reference for the temporal graph rather than concede the limitation, since the current admission that Kursun 'fails to specifically disclose' it does not preclude curing the gap in a next action.

How to adjust Strong on the current record because it rests on the examiner's OWN office-action admission plus OA2's confirmation that Dang's grounded text is directed to symptom-disease conditional probabilities with no temporal sequencing. Reinforce by pinning the exact office-action sentence and by pressing that neither Kursun nor Dang supplies ANY of the three alternatives (foreclosing the 'at least one of' escape). Caveat for counsel: because the examiner can cure a combination gap with a new reference, weigh whether the temporal-graph features are worth pulling into an independent claim to force the issue rather than leaving them in a dependent claim.

No reference supplies resource-usage-based constraint prioritization of claim 5 (subject to verifying the truncated mapping)

Missing elementClaim 5Claim 14Rebuts: §103 rejection of claims 1, 2, 3, 4, 5, 10, 11, 12, 13, 14, 19, 20

Strategy check: re-ranked from #5 — For claim 5 specifically this self-contained missing-element gap is more dispositive than the inherited (currently higher-effective) Khan attack, subject only to the evidentiary caveat that the office action was truncated before the claim-5 mapping — counsel must confirm against the complete action.

determine a system status indicating a computational resource usage level associated with the computing system; determine, using the system status, a priority threshold level; ... input the first constraint subset into the data generation model

None of the four fully-grounded reference excerpts — Kursun's GAN synthetic-data generation, Wonus's enrichment graph, Dang's medical-diagnosis knowledge graph, or Khan's PLNN inference — discloses selecting a constraint subset based on a computational-resource-usage system status and a derived priority threshold. This is a candidate missing-element gap for claim 5 (and parallel claim 14). Important caveat: the provided office action was truncated before the claim-5 mapping, so the examiner's actual claim-5 rationale and reference reads could not be reviewed; counsel must confirm this gap against the complete office action before relying on it.

OA3 (claim 5): 'None of the four fully-grounded reference excerpts ... discloses resource-usage-based constraint prioritization'
OA3 caveat: 'the provided office action was truncated before the claim-5 mapping'

MPEP § 2143.01 / § 2131 — a limitation absent from the cited art cannot support the rejectionEvidence needed: Obtain and review the complete office action's claim-5 (and claim-14) mapping.

Risk Because the claim-5 mapping is not in the reviewed record, the examiner may have cited additional support or an additional reference; counsel must read the complete office action before asserting this gap.

Likely examiner response⚠ fragile — the comeback likely defeats it

The examiner could point to the claim-5 mapping that counsel has not yet seen — the office action was truncated before it — and assert that resource-usage-based constraint prioritization was addressed there with a pin-cite (e.g., to Kursun's iterative refinement or model-ensemble subset selection, which OA2 confirms Kursun discloses). Because the actual rationale is off-record for counsel, the examiner is free to have mapped a system-status/priority-threshold read that counsel has not rebutted.

How to adjust Procedurally premature: the substantive gap may be real (none of the four grounded excerpts shows resource-usage-based prioritization), but the office action's claim-5 mapping is not in the record reviewed, so counsel cannot rebut a rationale it has not read. Obtain and review the complete office action before advancing this. If, once reviewed, no reference supplies the system-status/priority-threshold selection, this rises toward a clean missing-element candidate for claims 5 and 14; until then it should not be relied on.

Khan does not reach the add/exclude relationship-modification limitation on the amended independent claims

Mischaracterized referenceClaim 1Claim 10Claim 19Rebuts: §103 rejection of claims 1, 2, 3, 4, 5, 10, 11, 12, 13, 14, 19, 20

generating the inferred entity-relationship network comprises modifying the relationship dataset based on the inferred statistical dataset, including adding an inferred relationship based on the at least one correlation coefficient value when it is not included in the relationship dataset, and excluding a relationship when an explicit correlation associated with the relationship is not statistically significant

Khan is the only reference asserted for the amended add/exclude limitation that is the crux of the response, and Khan is fully grounded. On the available Khan text — the claims, abstract, and early detailed description — Khan discloses a probabilistic logical neural network whose novel 'relative correlation coefficient' (the 'J' parameter) is bounded in [−1,1] and 'modulate[s] the Fréchet inequalities' at logical operational nodes to tighten belief bounds during upward/downward inference (Khan claims 4-5). That available text is directed to inference and interpretability within a neural network, not to constructing a graph by adding relationships for correlated nodes and excluding relationships that are not statistically significant, and not to graph-based synthetic/simulated data generation. The examiner's support rests on Khan ¶¶0040-0041 and 0050, which are truncated and do not appear in the retrieved excerpt, so counsel should obtain and confirm those paragraphs before finalizing; on the record actually available, the contextual gap between belief-bound tightening and relationship-dataset modification is a distinction counsel may press for all three independent claims.

Khan claim 4: 'the logical operational nodes incorporate relative correlation coefficients bounded in a range of [−1, 1] that modulate the Fréchet inequalities'
Khan claim 5: relative correlation coefficients 'interpolate between a maximum anti-correlation represented by −1, statistical independence represented by 0, and maximum correlation represented by 1'
OA2 (Reference 4): Khan's available text 'is directed to inference and interpretability in machine-learning models, not to synthetic/simulated data generation; the detailed-description excerpt is truncated before the examiner-cited ¶¶0040-0041 and 0050'

MPEP § 2141.02 — a reference must be read for what it actually teaches; see also § 2143.01 (articulated reasoning with a rational underpinning)Evidence needed: Obtain and confirm the exact text of Khan ¶¶0040-0041 and 0050 before relying on this distinction.

Risk The examiner will likely quote Khan ¶¶0040-0041 and 0050 to show the correlation/anti-correlation reads as add/exclude; counsel must verify those paragraphs first because the reference is fully grounded and the examiner may have accurate support there. Prosecution-history caution: characterizing the claimed 'adding/excluding' step as fundamentally distinct from correlation-driven edge selection may narrow the scope later asserted for that step — frame the distinction carefully.

Likely examiner response◐ survives — moderate

The examiner would point to the very paragraphs counsel cannot yet see — Khan ¶¶0040-0041 and 0050 — as the express basis for the add/exclude mapping, and would argue that Khan's 'relative correlation coefficient' (the 'J' parameter) already supplies the raw operation: J expresses statistical independence at 0 and correlation toward ±1 between input nodes (per OA2/OA4 on Khan claims 4-5), so a node pair whose correlation is not statistically significant (J at/near 0) maps to 'excluding a relationship,' and a significant correlation coefficient maps to 'adding an inferred relationship.' The examiner would frame the difference between 'modulating Fréchet inequalities at operational nodes' and 'modifying a relationship dataset' as a mere labeling/context distinction, not a structural absence, and would note that under §103 Khan need only teach the recited operation, not the synthetic-data context.

How to adjust This survives only as far as the truncated paragraphs allow — it is genuinely VERIFY-FIRST. Counsel should obtain Khan ¶¶0040-0041 and 0050 before pressing; if those paragraphs disclose adding/excluding relationships for graph construction, the mischaracterization framing collapses and the better lever shifts to distinguishing the claimed relationship-dataset-modification-within-graph-generation as a whole (arrangement/context) or to amending to tie the add/exclude step to the downstream constraint-generation and simulated-data pipeline. Do not present the contextual gap as dispositive until the cited paragraphs are on the record.

Wonus teaches away from a machine-learning graph generation model

Teaching awayClaim 1Claim 10Claim 19Rebuts: §103 rejection of claims 1, 2, 3, 4, 5, 10, 11, 12, 13, 14, 19, 20

input the inferred statistical dataset and the node dataset into a graph generation model to generate an inferred entity-relationship network

Wonus is relied on for generating an inferred entity-relationship network via an enrichment memory graph, yet Wonus's own specification excerpt states that machine-learning approaches to associating node values are costly and unnecessary and that configuring the graph based on user input and human expertise is 'less costly and more effective than using machine learning methods.' The claims require a graph generation model that operates on the inferred statistical dataset. Counsel may argue Wonus discourages the very automated/model-driven graph construction the claims recite, undercutting the motivation to adapt Wonus's user-configured enrichment graph into the claimed graph generation model.

OA2 (Reference 2): Wonus states 'that machine-learning approaches to associating search terms are costly and unnecessary, and that user input/human expertise is "less costly and more effective than using machine learning methods"'
Wonus claim 9 / claim 18: 'node entries and edge relationship entries in the enrichment memory graph are configurable based on user input'

MPEP § 2145 (teaching away) and § 2141.02 (reference considered in its entirety, including passages that discourage the claimed path)

Risk The examiner will likely respond that Wonus's preference for user configuration does not criticize or disparage using a model, and that a preference is not a teaching away; counsel should tie the passage to the specific claimed 'graph generation model' step rather than to graph construction generally.

Likely examiner response◐ survives — moderate

The examiner would narrow the teaching-away quote to its context: Wonus's statement that user input is 'less costly and more effective than using machine learning methods' concerns associating SEARCH TERMS in an enrichment datastore, not graph-generation models generally, so it does not criticize or discredit the claimed graph generation model. The examiner would add that a mere preference for one approach as cheaper/more effective is not a teaching away under §2141.02/§2145 unless it discourages the claimed path, and that the combination draws its model-driven generation from Kursun's GAN, so Wonus need not itself supply the ML graph model.

How to adjust Viability depends on scope-matching. Pin the exact Wonus language and confirm from the office action which limitation Wonus is actually mapped to (the 'graph generation model' vs. the 'inferred entity-relationship network'). If Wonus is relied on for the graph-generation function, the disparagement of ML in that context is a genuine motivation-to-combine problem; if Wonus supplies only the graph structure and Kursun supplies the model, the teaching-away is blunted. Distinguish 'preference' from 'discouragement' proactively and be ready to argue the statement does more than express cost preference.

Four references from disparate fields assembled by impermissible hindsight

Improper hindsightClaim 1Claim 10Claim 19Rebuts: §103 rejection of claims 1, 2, 3, 4, 5, 10, 11, 12, 13, 14, 19, 20

input the inferred statistical dataset and the node dataset into a graph generation model to generate an inferred entity-relationship network ... generate ... a set of constraints for generation of simulated data ... to generate a simulated node dataset

The rejection stitches together four references drawn from markedly different fields — Kursun's GAN synthetic-data generation for anomaly detection, Wonus's enrichment-provenance datastore for search recall, Dang's medical-diagnosis knowledge graph, and Khan's probabilistic logical neural network for inference — to reconstruct the claimed pipeline. OA4 flags hindsight reconstruction as a strong combination weakness. The stated motivations are generic and field-specific to each reference in isolation (improved search recall for Wonus, likelihood-of-relatedness graphs for Dang, inductive reasoning for Khan) rather than an articulated reason a PHOSITA would have assembled all four into the specific claimed sequence of statistical inference → graph generation → constraint generation → constrained simulated-data generation. Counsel may argue the only roadmap connecting these disparate teachings is the applicant's own claim.

OA1 motivations: Wonus 'for improved searching and identifying related data through data enrichment'; Dang 'for generating a graph based on likelihood of relatedness of graph nodes'; Khan 'for using inductive reasoning to infer relationships between nodes'
OA2: Wonus is a provenance/lineage datastore, Dang is medical diagnosis, Khan is PLNN inference/interpretability — three different problem spaces from Kursun's GAN synthetic-data generation
OA4: 'hindsight_reconstruction (strong)'

MPEP § 2143.01 — the reason to combine must be articulated with a rational underpinning and cannot be drawn only from the applicant's disclosure; see also § 2145

Risk The examiner will respond that each combination step carries its own stated motivation and that KSR permits combining known elements for predictable results; counsel should be prepared to show why the aggregate sequence, not each pairwise step, lacks a rational underpinning.

Likely examiner response⚠ fragile — the comeback likely defeats it

The examiner would answer that KSR and MPEP §2143 do not require references from the same field, and that §2145 forbids attacking references individually where the rejection rests on their combination — each reference was mapped to a discrete limitation with a stated rationale, and 'disparate fields' alone is not a defect. The examiner would characterize the 'only roadmap is the applicant's claim' assertion as conclusory and note that an articulated per-reference motivation, even if field-specific, is precisely what §2143.01 requires; absent identification of a specific missing rational underpinning, the hindsight label is attorney argument.

How to adjust A generic 'hindsight/disparate fields' framing rarely carries on its own. To make it viable, counsel must anchor it to a SPECIFIC defect in the record — e.g., identify a limitation for which the office action gives no articulated reason to combine, or show the combination changes a reference's principle of operation or lacks a reasonable expectation of success (§2143.01/§2143.02). Better deployed as support for the concrete gaps (args 1, 2, 5) than as a standalone attack; consider folding it into the teaching-away point on Wonus.

Khan is arguably non-analogous art

Non-analogous artClaim 1Claim 10Claim 19Rebuts: §103 rejection of claims 1, 2, 3, 4, 5, 10, 11, 12, 13, 14, 19, 20

wherein the set of statistical metric values includes at least one correlation coefficient value corresponding to a correlation between a first and second node of the set of nodes

Khan's available text is directed to a probabilistic logical neural network that performs upward/downward inference by tightening belief bounds via activation functions set to a generalization of the Fréchet inequalities — an inference-and-interpretability technique for machine-learning models. The claimed field, as framed for argument in OA4, is computer-implemented generation of simulated datasets from graph-structured data to validate or train AI models. Counsel may argue Khan is neither in the same field of endeavor (it does not generate simulated/synthetic data) nor reasonably pertinent to the inventor's problem of building an inferred entity-relationship network to constrain a generative model. If Khan is non-analogous, it cannot support the §103 rejection of the amended correlation-coefficient and add/exclude limitations.

OA2 (Reference 4): Khan's available text 'is directed to inference and interpretability in machine-learning models, not to synthetic/simulated data generation'
Khan abstract: 'Inferencing is performed with a probabilistic logical neural network' governed by 'Fréchet inequalities'
OA4 field statement: claimed field is 'generation of simulated/synthetic datasets from graph-structured (node-and-relationship) data'

MPEP § 2141.01(a) — a reference is available only if same field of endeavor or reasonably pertinent to the inventor's problem

Risk The examiner will likely argue Khan is reasonably pertinent because both concern relationships between nodes expressed as correlation coefficients; the 'reasonably pertinent' prong is fact-sensitive and the examiner may frame the shared problem broadly. Prosecution-history caution: framing the field narrowly to exclude Khan may constrain later claim-scope arguments about what the invention encompasses.

Likely examiner response⚠ fragile — the comeback likely defeats it

The examiner would argue Khan IS analogous under both prongs of §2141.01(a): Khan is a machine-learning/graphical-model reference whose central 'J' parameter quantifies correlation between nodes — the same subject as the claimed 'correlation coefficient value corresponding to a correlation between a first and second node' — making it at minimum reasonably pertinent to the inventor's problem of characterizing inter-node correlations to build a relationship network, regardless of whether Khan itself generates synthetic data. The examiner would also note that the narrow 'field' framing (simulated-dataset generation from graph data) appears in OA4 argument commentary, not necessarily as the inventor's field of endeavor from the specification, so the framing is contestable.

How to adjust Non-analogous-art is hard where the references overlap in ML/graph modeling and the borrowed teaching is correlation-between-nodes itself. Confirm the inventor's actual field of endeavor from the as-filed specification (not the OA4 argument framing, which is analysis, not record) before asserting it. This argument is weaker than the direct content gap; if pressed at all, pair it with arg 1 (both target Khan) rather than run it independently, and treat the 'reasonably pertinent to the problem' prong as the likely losing point.

Generic, conclusory motivations to combine

Conclusory rationaleClaim 1Claim 10Claim 19Rebuts: §103 rejection of claims 1, 2, 3, 4, 5, 10, 11, 12, 13, 14, 19, 20

input at least a portion of the set of constraints and the inferred entity-relationship network into a data generation model to generate a simulated node dataset consistent with the set of constraints and the inferred entity-relationship network including the inferred relationship

The stated reasons to combine recite only each reference's own general benefit (improved search recall for Wonus, likelihood-of-relatedness graphs for Dang, inductive reasoning for Khan) followed by a boilerplate 'reasonable expectation of success' recital, without an articulated line of reasoning explaining why a PHOSITA would integrate these teachings to reach the specific claimed data-generation pipeline. Counsel may argue the motivations do not connect the borrowed feature to the claimed combination as a whole and therefore fall short of the articulated-reasoning requirement.

OA1 motivations recite generic benefits: Wonus 'improved searching and identifying related data,' Dang 'generating a graph based upon likelihood of relatedness,' Khan 'using inductive reasoning to infer relationships'
Office action combination statements each end with the boilerplate 'with a reasonable expectation of success'

MPEP § 2143 / § 2143.01 — each KSR rationale requires factual findings and an articulated rational underpinning, not a conclusory assertion

Risk The examiner may cure by supplementing the reasoning in the next action; a conclusory-rationale argument standing alone rarely disposes of a rejection and is strongest paired with the substantive missing-element and hindsight arguments above.

Examiner's Characterization of the Cited Art

Note

What each cited reference actually discloses, checked against what the examiner said it teaches — limited to the reference text available to the analysis.

Kursun

US 2021/0049455Claim text retrieved

Kursun (US 2021/0049455) discloses an iterative synthetic-data-generation system that analyzes input data with a machine learning model to identify 'emerging patterns' of abnormality, extracts common data characteristics to define a 'data scenario,' broadens the scope of that scenario using encoding/decoding logic, and generates a synthetic data set using a generative adversarial neural network (GAN). The synthetic data is refined by testing it against the identified pattern, may be stored in a repository, and is used to retrain machine learning models (including model ensembles built from subsets of models). The available text (abstract, claims 1-20, and a truncated description excerpt) confirms the GAN-based synthetic data generation, iterative refinement, and retraining/ensemble concepts, but does NOT contain many of the specific paragraphs the examiner pin-cites (e.g., ¶¶0003, 0041, 0046-0050, 0052, and the Fig. 4 item-level details).

Claim element	Examiner asserts	Reference discloses	Evidence
input the dataset into a statistical inference model to generate an inferred statistical dataset including a set of statistical metrics and statistical metric values	Kursun's machine learning model analyzes a dataset and identifies patterns; statistical metrics are associated with the dataset and stored in a policy database (¶0041); metadata is analyzed to identify emerging patterns (¶¶0046-0049).	Partially supported	The available text supports analyzing input data with a machine learning model to identify an emerging pattern (claim 1). However, the specific 'statistical inference model,' 'statistical metrics,' 'metric values,' 'policy database,' and metadata pin-cites (¶¶0003, 0041, 0046-0049) are not found in the available text (claims, abstract, and truncated description); the full specification should be checked. The available text does not use the terms 'statistical metric' or 'statistical inference model.'
generate, using the inferred entity-relationship network, a set of constraints for generation of simulated data	Kursun identifies inferred entity-relationship data (abnormal patterns based on interactions and characteristics) and uses it to generate synthetic data via a GAN (Fig. 4 items 414/416; ¶0050).	Partially supported	The GAN-based generation of synthetic data from identified patterns is supported (claim 1; abstract; Fig. 4 is described in the drawing list as 'identification of abnormal data patterns and generation of synthetic data'). However, the specific ¶0050 text and Fig. 4 item-level detail (414, 416) are not in the available excerpt, and — notably — the examiner's own rejection concedes Kursun 'fails to specifically disclose' the inferred entity-relationship network and instead relies on Wonus/Dang/Khan for it. Whether Kursun independently teaches 'constraints' derived from an 'entity-relationship network' is not established in the available text.
transmit the simulated dataset to a user device to cause validation or training of an artificial intelligence model	Synthetic/simulated data is transmitted to the data repository for retraining models (Fig. 4 item 422).	Partially supported	Storing synthetic data for training and retraining models is supported (claim 3: "store the multiple synthetic data sets in a data repository for training"; claim 4: "the machine learning model is retrained"). However, the specific step of transmitting to a 'user device' (as opposed to a data repository) and the Fig. 4 item 422 pin-cite are not verifiable in the available text; the examiner's own citation refers to 'the data repository,' not a user device — a distinction counsel may wish to weigh against the claim language 'transmit the simulated dataset to a user device.'
claim 2 — specialized generative model set, model specialization set, subset of models (portion set)	Kursun's GAN is specialized based on identified emerging patterns, and a subset of one or more models determined to be most accurate is used to update the model (¶¶0004, 0008, 0041).	Partially supported	Claim 5 recites generating "a machine learning model ensemble using a subset of the one or more additional machine learning models," and claim 6 recites the ensemble is "continuously updated to include a subset...determined to be most accurate." This supports a 'subset of models' concept. However, the available text does not describe a 'specialized generative model set' where each specialized model corresponds to a 'model specialization,' nor a 'portion set' keyed to portions of an entity-relationship network; ¶¶0004 and 0008 in the granularity the examiner asserts are not in the available excerpt.
claim 4 — determine time-series data associated with the dataset, each data point associated with a timestamp	Kursun collects data by monitoring a source over a period of time (¶0023), and historical data includes an associated timestamp (¶0052).	Partially supported	The 'monitoring...over a period of time' concept is in the available text (the 'To monitor' definition: "to watch, observe, or check something for a special purpose over a period of time"). However, the characterization of collected data as 'time-series data,' the 'data point / timestamp' mapping, and ¶0052 (historical data with associated timestamp) are not found in the available text; the full specification should be checked.
one or more processors and one or more non-transitory computer-readable storage media storing instructions (claim 1 preamble)	Kursun discloses a processor configured to execute computer-executable code (¶0053) and a computer-readable medium storing instructions (¶0054).	Supported	Claim 1 recites "a memory storage device, a communication device, and a processor, with computer-readable program code stored thereon"; claim 15 recites "at least one non-transitory computer-readable medium having computer-readable program code portions embodied therein." (The specific ¶¶0053-0054 are not in the available excerpt, but equivalent content is present in the claims.)
input at least a portion of the constraints and the inferred entity-relationship data into a data generation model to generate a simulated dataset (GAN)	Abnormal pattern data is used to generate synthetic/simulated data using a GAN architecture (Fig. 4 items 414/416; ¶0050).	Supported	Claim 1: "based on the expanded scope of the data scenario, generate a synthetic data set using a generative adversarial neural network." The GAN synthetic-data-generation mechanism is squarely in the available text. (The 'constraints and inferred entity-relationship data' framing is the examiner's overlay; the underlying GAN generation is supported.)

Claim 1"receive input data for analysis and expansion; analyze the input data using a machine learning model to identify an emerging pattern in the input data; extract common data characteristics from the identified emerging pattern in order to determine a data scenario; utilize encoding and decoding logic to expand scope of the data scenario; and based on the expanded scope of the data scenario, generate a synthetic data set using a generative adversarial neural network." (Kursun, claim 1)This is the core disclosure the examiner relies on for the model-based analysis and GAN synthetic-data-generation limitations; it is verbatim in the available text.
AbstractKursun's abstract describes generating synthetic data via an iterative process, analyzing data to determine emerging patterns, broadening scope, generating a synthetic data set with a generative neural network, and using it 'to further train one or more machine learning models for further pattern detection.'Supports the examiner's characterization of GAN synthetic data generation and using the synthetic data to train/retrain AI models.
Claims 3 and 4Claim 3: "generate multiple synthetic data sets; and store the multiple synthetic data sets in a data repository for training a set of one or more additional machine learning models." Claim 4: "the machine learning model is retrained using a combination of real input data and data from the synthetic data set."Supports the examiner's assertion that Kursun transmits/uses simulated data to cause training or retraining of a model.
Claims 5 and 6Claim 5: "generate a machine learning model ensemble using a subset of the one or more additional machine learning models..."; Claim 6: the ensemble "is continuously updated to include a subset of one or more additional machine learning models determined to be most accurate in identifying emerging patterns."This is the available-text support for the dependent-claim-2 'specialized generative model set' / 'subset of models' mapping the examiner attributes to ¶0008.
Detailed Description (GAN definition paragraph)Kursun defines a GAN in which "the generative network learns to map from a latent space to a data distribution of interest" and "the generator is seeded with randomized input that is sampled from a predefined latent space, such as a multivariate normal distribution," concluding "a GAN architecture is a useful tool in generating synthetic data that mimics historical training data."Confirms the GAN-based synthetic-data-generation mechanism but frames the synthetic data as mimicking a distribution — not as being constrained by an inferred entity-relationship network with correlation coefficients.
Detailed Description ('To monitor' definition)"To 'monitor' is to watch, observe, or check something for a special purpose over a period of time. The 'monitoring' may occur periodically over the period of time, or the monitoring may occur continuously over the period of time."Bears on the dependent-claim-4 assertion that Kursun collects data over a period of time; the monitoring-over-time concept is present, but the specific 'time-series data' and 'timestamp' pin-cites (¶¶0023, 0052) are not verifiable in the available text.

Wonus

US 2023/0418878Claim text retrieved

Wonus discloses a multi-model datastore combining an 'enrichment memory graph' with an 'enrichment catalog' to maintain provenance and lineage of enriched data and to improve search recall. Raw data values are transformed by enrichment functions into enriched data values; both are registered as nodes in the enrichment memory graph, and an edge relationship is created between them based on the enrichment type, with the edge also indicating a confidence level (abstract). The graph can infer edge relationships transitively (an edge between a raw value node and an enriched value node combined with an edge between the raw value node and an intermediate node), and node/edge entries are configurable by user input. Notably, the specification excerpt states that machine-learning approaches to associating search terms are costly and unnecessary, and that user input/human expertise is 'less costly and more effective than using machine learning methods.'

Claim element	Examiner asserts	Reference discloses	Evidence
generate an inferred entity-relationship network including a set of inferred node identifiers and a set of inferred node relationships (graph generation model)	Wonus (¶¶0023-0024) discloses generating an inferred entity-relationship network with inferred relationships (e.g., deriving citizenship from birthplace) using an enrichment memory graph that augments an enrichment catalog with provenance/lineage and is leveraged for graph traversal to identify relationships.	Partially supported	Supported as to a graph with inferred relationships: claim 3/12 recites inferring an edge relationship transitively, and the description recites 'deriving citizenship data from place of birth data' and an enrichment memory graph used to find related terms. Not clearly supported as to a 'graph generation model': the available text describes an 'enrichment memory graph constructor' executing enrichment functions and a graph that is 'configurable based on user input,' and affirmatively states machine-learning is 'not needed' and that user input is 'more effective than using machine learning methods.' The inference in Wonus is transitive/edge-combination based, not statistical. Whether an enrichment-function/user-configured graph constructor is a 'graph generation model' is a characterization point for counsel, and the ML disclaimer is a possible teaching-away consideration under MPEP 2141.02.
receive a node dataset comprising an entity dataset and a relationship dataset, wherein the entity dataset comprises a representation of a set of nodes and associated node values, and the relationship dataset includes relationships between at least two nodes	Wonus (¶0005) discloses a node dataset consisting of entities having a value matching an initial search value, where each entity/node includes a node value and an edge relationship value.	Partially supported	The abstract and claims 1/4 support nodes with values and edge relationships ("Enriched data is stored in graph nodes with edge associations"; claim 8/17: "one or both of the first raw data value and the first enriched data value comprise a numerical value"). However, in the available text Wonus builds/queries an internal enrichment memory graph from raw data values rather than 'receiving' a pre-formed node dataset containing entities and relationships as input; the closest support (claim 4) is retrieval of node values and edges from an existing graph in response to a search value. Whether this reads on 'receiving a node dataset' is a nuance for counsel. Paragraph numbering (¶0005) is not present in the provided text, but the substance appears in claim 4 / the description excerpt.

Abstract"Enriched data is stored in graph nodes with edge associations indicating the enrichment type and confidence level." Additional enriched values and associations between graph nodes can be entered in the graph by a user.Shows nodes with values and edges (relationships) carrying a type and a confidence level — the basis for the examiner's node-dataset / relationship mapping.
Claim 1 (and method claim 10, medium claim 19)Register a node for a raw data value and a node for an enriched data value, and "create an edge relationship between the first raw data value node and the first enriched data value node in the enrichment memory graph based on the first enrichment type."Supports that Wonus builds a graph of nodes connected by edge relationships derived from enrichment functions.
Claim 3 (and claim 12)"infer an edge relationship between the first enriched data value node and an intermediate node ... wherein the inferred edge relationship is based on a combination of (1) the edge relationship between the first raw data value node and the first enriched data value node and (2) an edge relationship between the first raw data value node and the intermediate node."Directly supports the examiner's 'inferred relationships / inferred entity-relationship network' mapping — but the inference is transitive/combinatorial over existing edges, not statistical.
Claim 4 (and claim 13)Identify an enrichment memory graph node matching an initial search value and "determine a set of enrichment memory graph node values that includes at least the enrichment memory graph node value matching the initial search value and any enrichment memory graph node value having an edge relationship or an inferred edge relationship."Corresponds to the examiner's paraphrase (attributed to ¶0005) of 'a node dataset consisting of entities having a value matching an initial search value ... a node value and an edge relationship value.'
Description (Section II, 'deriving citizenship data from place of birth data')"a data transformation may include deriving citizenship data from place of birth data, mapping Internet protocol (IP) address data to geolocation data ... or translating term data or name data from a first human language to equivalent term data or name data in a second human language."This is the source of the examiner's citizenship-from-birthplace example used to characterize inferred relationships/enrichments.
Description (Section II, enrichment memory graph configurability)"Utilizing machine learning algorithms to associate search terms may involve use of excessive resources (i.e., costs) to train the data. However, such resources are not needed for the enrichment memory graph model ... configuring or updating the enrichment memory graph based on user input and human expertise can be less costly and more effective than using machine learning methods."Potential MPEP 2141.02 teaching-away consideration for counsel: Wonus expressly favors user-configured/human-expertise graph construction over machine-learning, whereas the claim recites inputting data into a 'graph generation model.' Counsel may weigh whether characterizing Wonus as supplying a model-based graph generator is consistent with the reference read as a whole.

Dang

US 2022/0208373Claim text retrieved

Dang's available text (abstract, claims 1-16, and a truncated description) describes a machine-guided medical-diagnosis system that generates a symptom-disease knowledge graph from electronic medical guideline documents, detects factual nodes (symptoms/tests/contexts), and infers evidential, feature, and disease nodes based on confidence values and hyponymy relationships. Connections between nodes carry conditional probabilities P (e.g., the likelihood that one symptom occurs given another is observed). Likelihood values for suspected diseases are computed using a combined TF-IDF and Bayesian modeling process, with a Gini-index step used to generate a 'best next inquiry question.' The available text is directed entirely to symptom-disease relationships and conditional-probability edges; it does not, in the portion provided, address temporal sequencing or time-based data.

Claim element	Examiner asserts	Reference discloses	Evidence
claim 4 — a (temporal) dependency graph identifying at least one of a temporal dependency constraint, a sequence progression constraint, or a time-based correlation	Dang's dependency graph identifies at least one of a temporal constraint, a sequence progression constraint, or a time-based correlation, and the dependency graph is used for a determination such as identifying a disease within a probability (¶¶0003-0005, 0041-0045).	Not found in available text	The available text describes a symptom-disease knowledge graph built on conditional probabilities between symptoms/diseases; none of the three claimed alternatives (temporal dependency constraint, sequence progression constraint, or time-based correlation) is found in the available text, and the passages the examiner cites (structural graph, semantic matching, confidence score at ¶¶0003-0005) affirmatively concern non-temporal properties. For counsel to weigh whether the cited evidence maps to any temporal/sequence/time-based alternative; the full specification should be checked.
claim 3 — statistical metric including at least one of a univariate metric, a multivariate metric, etc., determined between node values	Dang uses a set of univariate and multivariate equations (metrics) to determine confidence values for inferring (¶¶0047-0051).	Not found in available text	The available text describes TF-IDF, Bayesian modeling, feature prioritization, and a similarity analysis "made using distances of data sets considering mutual difference measures," and yields confidence scores/values, but the specific 'univariate' and 'multivariate equation' terminology does not appear in the available (partly truncated) text; the full specification at ¶¶0047-0051 should be checked.
the indication of structural, semantic, and statistical properties is consistent with the inferred statistical dataset	Dang's graph is generated from medical guideline documents by calculating likelihoods, optimizing symptom weights, and applying modeling, so the structural/semantic/statistical properties are consistent with the inferred statistical dataset (¶¶0003-0005).	Partially supported	The available text supports the underlying likelihood/confidence/TF-IDF-Bayesian computations (Claims 3-4, 8-9), but 'consistent with the inferred statistical dataset' is a claim-mapping conclusion tying Dang to the other references' constructs rather than language found in Dang; for counsel to weigh whether the cited passages actually establish this consistency limitation.
a graph generation model to generate an inferred entity-relationship network (set of inferred nodes)	Dang generates a graph including a set of inferred evidential nodes in the set of factual nodes (¶¶0003, 0010).	Supported	Claim 7 / SUMMARY: "detecting a set of factual nodes ...; inferring a set of evidential nodes in the set of factual nodes; inferring a set of disease feature nodes ...; and inferring a set of possible disease nodes."
each relationship indicates a particular relationship label between at least two particular nodes	In Dang, each relationship between two nodes is labeled with a conditional probability indicating the probability that one symptom occurs when another is observed (Fig. 1B; ¶¶0030-0032).	Supported	DETAILED DESCRIPTION (FIG. 1B): "Each connection to another symptom has a corresponding conditional probability P ... indicates ... the likelihood of the symptom s m1 occurring when the symptom s m2 is observed."
indication of structural properties of the node dataset	Dang's knowledge graph data structure is generated based on inferred data and likelihood values (¶0003).	Supported	Abstract: "Generate a knowledge graph data structure using one or more electronic medical guideline documents."
indication of semantic properties of the node dataset	Dang uses semantic matching to determine similarity between input text and nodes in the graph (¶¶0009, 0047).	Supported	Claim 8: "measuring, using semantic matching, a similarity between an input text one or more nodes in the knowledge graph data structure ..."
indication of statistical properties of the node dataset	Dang determines a confidence score (statistical property) for the inferred set of nodes (¶¶0009, 0010).	Supported	Claim 8: "the measuring yields a confidence score"; Claim 9: inference is "based on a confidence value for each of the one or more candidate evidential nodes."

Abstract"Generate a knowledge graph data structure using one or more electronic medical guideline documents. Evaluate a likelihood value for one or more diseases based on the knowledge graph data structure."Supports the examiner's use of Dang for graph generation (structural) and likelihood/statistical properties.
Claim 7 / SUMMARY"detecting a set of factual nodes in a knowledge graph data structure; inferring a set of evidential nodes in the set of factual nodes; inferring a set of disease feature nodes in the set of evidential nodes; and inferring a set of possible disease nodes in the set of diseases feature nodes."Supports the assertion that Dang generates a graph containing inferred nodes derived from an initial node set.
Claim 8 / SUMMARY"measuring, using semantic matching, a similarity between an input text one or more nodes in the knowledge graph data structure, the input text comprising one or more observed symptoms, wherein the measuring yields a confidence score."Supports the examiner's mapping of the 'semantic' property (semantic matching) and 'statistical' property (confidence score).
Claim 9 / SUMMARY"identifying one or more candidate evidential nodes ... based on hyponymy relationships between at least two factual nodes ... and is further based on a confidence value for each of the one or more candidate evidential nodes; adding to the ... set of evidential nodes, at least one of the one or more candidate evidential nodes based on its corresponding confidence value."Supports inference of new nodes/relationships governed by confidence values.
DETAILED DESCRIPTION (FIG. 1B discussion)"Each connection to another symptom has a corresponding conditional probability P. ... A given conditional probability P(s m1 |s m2 ) indicates, for a pair of symptoms (s m1 , s m2 ), the likelihood of the symptom s m1 occurring when the symptom s m2 is observed." Disease-to-feature and disease-to-disease connections likewise carry conditional probabilities.Supports the examiner's mapping that each inferred relationship carries a label (conditional probability) between at least two nodes.
DETAILED DESCRIPTION (TF-IDF/Bayesian discussion; FIGS. 3-5)A combined TF-IDF and Bayesian modeling process is applied to the symptom-disease knowledge graph; feature recognition, feature selection, feature prioritization via information retrieval, and a similarity analysis "made using distances of data sets considering mutual difference measures" are performed to compute likelihood values.Bears on the claim 3 assertion that statistical metrics are computed between node values; note the specific 'univariate/multivariate equation' terminology does not appear in the available text.

Khan

US 2025/0111206Claim text retrieved

Khan describes a probabilistic logical neural network (PLNN) — a probabilistic graphical model of propositional nodes and logical operational nodes (AND, OR, NOT, implication, conditional, equivalence) joined by directed edges, which performs upward and downward inference by tightening belief bounds via activation functions set to a generalization of the Fréchet inequalities. Its central novel parameter is a 'relative correlation coefficient' (the 'J' parameter) bounded in [−1, 1] that modulates the Fréchet inequalities at operational nodes and expresses maximum anti-correlation (−1), statistical independence (0), and maximum correlation (1) between input nodes. The available text (claims, abstract, and an early portion of the detailed description) is directed to inference and interpretability in machine-learning models, not to synthetic/simulated data generation; the detailed-description excerpt is truncated before the examiner-cited ¶¶0040-0041 and 0050.

Claim element	Examiner asserts	Reference discloses	Evidence
adding, based on the at least one correlation coefficient value between the first and second nodes, an inferred relationship between the first and second nodes when the inferred relationship is not included in the relationship dataset	Khan (¶¶0040-0041, 0050) calculates correlation/anti-correlation and the examiner interprets nodes with a positive correlation value as those added to the relationship.	Not found in available text	The available text (claims, abstract, and truncated detailed description ending at the J-parameter discussion) does not contain ¶¶0040-0041 or 0050; the available text describes the J coefficient as modulating Fréchet-inequality bounds at operational nodes and describes 'node spawning' during training, but does not describe adding a relationship/edge to a relationship dataset based on a positive correlation value. The full specification (¶¶0040-0041, 0050) should be checked to confirm whether it supports the examiner's reading.
excluding, from the inferred entity-relationship network, a relationship of the relationship dataset when an explicit correlation associated with the relationship is not statistically significant	Khan (¶¶0040-0041, 0050) calculates correlation/anti-correlation and the examiner interprets nodes with an anti-correlation value as those excluded from the relationship.	Not found in available text	The available text does not contain ¶¶0040-0041 or 0050. Available passages describe anti-correlation (J=−1) as a coefficient value modulating probability bounds and describe mutually-exclusive disjunction (⊕) nodes, but do not describe excluding a relationship from a relationship dataset because a correlation is not statistically significant. For counsel to weigh: the examiner equates 'anti-correlation' with 'not statistically significant,' whereas the available text distinguishes anti-correlation (−1) from statistical independence (0) — the full specification should be checked. The claim language keys exclusion to lack of statistical significance, not to anti-correlation.
the generated simulated node dataset exhibits the at least one correlation coefficient value of the inferred statistical dataset for the node dataset	Khan (¶¶0040-0041, 0050) — nodes having a positive correlation value are those added to the relationship, such that the dataset exhibits the correlation coefficient value.	Not found in available text	The available text does not contain ¶¶0040-0041 or 0050 and does not describe generation of a simulated/synthetic dataset at all; Khan's available text is directed to inference within a probabilistic logical neural network (Abstract; Claim 1). Whether the full specification connects the J coefficient to any generated dataset should be checked; the available text does not support this mapping.
the set of statistical metric values includes at least one correlation coefficient value corresponding to a correlation between a first and second node of the set of nodes	Khan (¶¶0016, 0050-0051) expresses relatedness as correlation/anti-correlation of nodes, defined using a correlation coefficient.	Supported	Claims 4-5 and the description recite a 'relative correlation coefficient' (J parameter) bounded in [−1, 1] that expresses maximum anti-correlation (−1), independence (0), and maximum correlation (1) 'between two input nodes'; the specific ¶¶0016/0050-0051 pin-cites are outside the available excerpt but the concept is affirmatively present in the available claims and description.
motivation to combine — using inductive reasoning to infer relationships between nodes	Khan (¶0016) supports using inductive reasoning to infer relationships between nodes.	Partially supported	The description states the model 'integrates inductive reasoning (e.g., in a manner of a neural network and/or based on statistics), deductive reasoning, and probabilistic reasoning' and 'make[s] inferences on systems for which little training data is provided.' The available text supports inductive/probabilistic inference generally, but does not, in the available portion, describe inferring or constructing relationships (edges) between nodes as opposed to inferring probability bounds on a pre-specified graph structure.

Claim 4"the logical operational nodes incorporate relative correlation coefficients bounded in a range of [−1, 1] that modulate the Fréchet inequalities, and wherein the relative correlation coefficients are taken as input and provided as output at the logical operational nodes"This is the textual basis for the examiner's 'correlation coefficient value' mapping — the J parameter is a correlation coefficient associated with operational nodes.
Claim 5"the relative correlation coefficients interpolate between a maximum anti-correlation represented by −1, statistical independence represented by 0, and maximum correlation represented by 1"Supports that Khan expresses both correlation and anti-correlation between nodes as a numeric coefficient, which the examiner reads onto the amended correlation/anti-correlation limitations.
Detailed Description (SUMMARY/DETAILED DESCRIPTION excerpt)"The enhanced machine learning model directly expresses degrees of uncertainty among variables, e.g., correlation and/or anti-correlation and independence. This expression occurs via a new parameter (e.g., a relative correlation coefficient, that is herein called the 'J' parameter)."Confirms the J parameter expresses correlation/anti-correlation among variables — the concept the examiner relies on for the amended limitations.
Detailed Description (excerpt, discussing the J element of the PLNN tuple)"J=1 refers to a condition of maximum correlation between two input nodes given fixed marginal probabilities for the input nodes. J=−1 refers to a condition of maximum anti-corre[lation]"Ties the correlation coefficient to a relationship between two input nodes, but describes it as a coefficient modulating probability bounds at operational nodes — not as a mechanism for adding or removing relationships from a dataset.
Abstract; Claim 1"Inferencing is performed with a probabilistic logical neural network... implements upward and downward inference... nodes are coupled with respective belief bounds... each logical operational node comprises a respective activation function set to a probability-respecting generalization of the Fréchet inequalities."Frames Khan's actual subject matter as probabilistic inference within a fixed logical graphical model — for counsel to weigh against the examiner's use of Khan for adding/excluding relationships and for generating simulated data.

Element-by-Element Claim Chart

Claim 1 — §103 (Kursun in view of Wonus and Dang and Khan)

Status glyph	Claim element	Status	Disclosure / notes	Location
✓	one or more processorsKursun	Taught	Processor is disclosed in grounded Kursun claim text. The OA pin-cite to ¶0053 is not present in the available Kursun excerpt, but the concept is independently grounded in the claims.	Kursun, claim 1 ('a module containing a memory storage device, a communication device, and a processor'); Kursun, claim 15
✓	non-transitory computer-readable storage media storing instructions executed by the processorsKursun	Taught	Grounded in Kursun claim 15.	Kursun, claim 15 ('at least one non-transitory computer-readable medium having computer-readable program code portions embodied therein')
◐	receive a node dataset comprising an entity dataset (set of nodes and node values) and a relationship dataset (relationships between at least two nodes)Wonus	Arguably taught	Wonus's enrichment memory graph has value-bearing nodes and edge relationships (OA2). For counsel to weigh whether Wonus's raw/enriched-value graph nodes map to the claimed 'entity dataset' with 'associated node values' plus a separate 'relationship dataset'; the reading is stretchable but contestable.	Wonus, claim 1 ('register a node for the first raw data value and a node for the first enriched data value ... create an edge relationship ... based on the first enrichment type'); Wonus, claim 4
◐	input the node dataset into a statistical inference model to generate an inferred statistical dataset having a set of statistical metrics and metric values, each associated with one or more nodes or relationshipsKursun	Arguably taught	Per OA2, the OA's specific pin-cites for this element (Kursun ¶¶0003, 0041 'policy database'/'statistical metrics') are NOT present in the available Kursun excerpt. Grounded Kursun text describes an ML model identifying 'emerging patterns' and extracting 'common data characteristics' — for counsel to weigh whether that equates to a 'statistical inference model' producing 'statistical metrics and metric values.' Verify the ¶0041 statistical-metric disclosure against the actual reference.	Kursun, abstract ('continuously analyzing the data to determine emerging patterns'); Kursun, claim 1 ('extract common data characteristics from the identified emerging pattern')
◐	the set of statistical metric values includes at least one correlation coefficient value corresponding to a correlation between a first and second node (amended limitation)Khan	Arguably taught	Khan's 'J' correlation coefficient is grounded in the claims. For counsel to weigh a mapping mismatch: per OA2, Khan's coefficient modulates Fréchet inequalities between propositional/logical-operational nodes of an inference network (belief bounds), not a 'statistical metric value' of a data node dataset describing correlation between two data nodes. Contestable whether Khan's inference-node coefficient reads on 'correlation between a first and second node of the set of nodes.'	Khan, claim 4 ('relative correlation coefficients bounded in a range of [−1, 1]'); Khan, claim 5 ('maximum anti-correlation represented by −1, statistical independence represented by 0, and maximum correlation represented by 1')
◐	input the inferred statistical dataset and node dataset into a graph generation model to generate an inferred entity-relationship network with inferred node identifiers and inferred node relationshipsWonus, Dang	Arguably taught	Wonus infers edge relationships in an enrichment memory graph; Dang generates a knowledge graph with inferred nodes. For counsel to weigh whether either supplies a 'graph generation model' taking BOTH an inferred statistical dataset and the node dataset as input.	Wonus, claim 3 ('infer an edge relationship ...'); Dang, claim 2 ('generating the knowledge graph data structure')
◐	the inferred network includes an indication of structural, semantic, and statistical properties of the node dataset, consistent with the inferred statistical datasetDang	Arguably taught	Dang provides a graph structure (structural), semantic matching (semantic), and a confidence score (statistical) — grounded in claims. For counsel to weigh whether these three properties, in Dang's medical-diagnosis context, are shown to be 'consistent with the inferred statistical dataset' as claimed.	Dang, claim 8 ('measuring, using semantic matching, a similarity ... wherein the measuring yields a confidence score'); Dang, claim 2 (knowledge graph data structure)
✓	each inferred node relationship indicates a particular relationship label between at least two particular inferred nodesDang	Taught	Grounded description discloses labeled conditional-probability edges between node pairs. The OA's Fig. 1B/¶¶0030-0032 pin-cite corresponds to the grounded conditional-probability discussion.	Dang, description ('A given conditional probability P(s_m1 \| s_m2) indicates, for a pair of symptoms ... the likelihood of the symptom s_m1 occurring when the symptom s_m2 is observed')
◐	generating the network includes ADDING, based on the correlation coefficient, an inferred relationship between the first and second nodes when it is not in the relationship dataset (amended limitation)Khan	Arguably taught	The grounded Khan text (claims, abstract, early description) discloses the J correlation coefficient but does NOT show modifying a relationship dataset by adding a relationship for positively-correlated nodes. Per OA2, the examiner's cited ¶¶0040-0041 and 0050 are truncated and NOT in the available Khan excerpt — counsel should obtain and verify what those paragraphs actually disclose before relying on this mapping. Note contextual gap: Khan is directed to inference/belief-bound tightening, not to relationship-dataset construction for synthetic data generation.	Khan, claims 4-5 (relative correlation coefficient in [−1,1])
◐	generating the network includes EXCLUDING a relationship when an explicit correlation associated with it is not statistically significant (amended limitation)Khan	Arguably taught	Examiner reads anti-correlated nodes as 'excluded' (OA cites Khan ¶¶0040-0041, 0050 — not in available excerpt per OA2). For counsel to weigh: the claim ties exclusion to a relationship NOT being 'statistically significant,' whereas Khan's grounded text describes anti-correlation (J=−1) and independence (J=0) as coefficient values, not a statistical-significance test used to exclude a relationship from a graph. Verify the cited paragraphs.	Khan, claim 5 ('maximum anti-correlation represented by −1, statistical independence represented by 0')
◐	generate, using the inferred network, a set of constraints for generation of simulated dataKursun	Arguably taught	Per OA2 the OA's Fig. 4/¶0050 pin-cites are not in the available excerpt. Grounded Kursun 'data scenario' / scope-expansion concept is the closest disclosure; for counsel to weigh whether that constitutes a 'set of constraints.'	Kursun, claim 1 ('extract common data characteristics ... determine a data scenario; utilize encoding and decoding logic to expand scope of the data scenario')
✓	input constraints and the inferred network into a data generation model to generate a simulated dataset consistent with the constraints and network, including the inferred relationshipKursun	Taught	GAN-based synthetic/simulated data generation is grounded in Kursun claims/abstract. For counsel to weigh whether Kursun's GAN output is shown to be 'consistent with' an inferred entity-relationship network 'including the inferred relationship' (the network/inferred-relationship framing comes from Wonus/Dang/Khan, not Kursun).	Kursun, claim 1 ('based on the expanded scope of the data scenario, generate a synthetic data set using a generative adversarial neural network'); Kursun, abstract
◐	generated simulated node dataset exhibits the correlation coefficient value of the inferred statistical dataset (amended limitation)Khan, Kursun	Arguably taught	This limitation combines Khan's correlation coefficient with Kursun's GAN output. Per OA2, Khan is directed to inference/interpretability, NOT synthetic-data generation, and Kursun's grounded text does not mention correlation coefficients. For counsel to weigh whether the combination shows a simulated dataset that 'exhibits' the correlation coefficient.	Khan, claims 4-5; Kursun, claim 1 (GAN synthetic data)
◐	transmit the simulated dataset to a user device to cause validation or training of an artificial intelligence modelKursun	Arguably taught	Retraining/validation using synthetic data is grounded. For counsel to weigh whether 'transmit ... to a user device' specifically is shown; Kursun stores in a data repository and describes user devices generally, but the transmit-to-user-device step is a contestable read (OA cites Fig. 4 item 422, not in available excerpt).	Kursun, claim 3 ('store the multiple synthetic data sets in a data repository for training a set of one or more additional machine learning models'); Kursun, claim 4 (retraining)

Claim 10 — §103 (Kursun in view of Wonus and Dang and Khan)

Status glyph	Claim element	Status	Disclosure / notes	Location
✓	non-transitory computer-readable storage media with instructions (independent media claim mirroring claim 1)Kursun	Taught	Claim 10 tracks claim 1 in media form; the shared substantive limitations are charted under claim 1 and not repeated here except the contested ones below.	Kursun, claim 15
◐	correlation coefficient value corresponding to correlation between first and second node (amended)Khan	Arguably taught	Same contestable mapping as claim 1 — Khan's J coefficient operates on inference-network nodes, not clearly on data-node correlation. Note: claim 10 deleted the earlier 'network is consistent with the inferred statistical dataset' phrasing and replaced it with 'the indication of the structural, semantic, and statistical properties is consistent with the inferred statistical dataset' — for counsel to confirm the examiner's mapping addresses the amended phrasing.	Khan, claims 4-5
◐	modifying the relationship dataset by adding/excluding relationships based on the correlation coefficient / statistical significance (amended)Khan	Arguably taught	Same posture as claim 1: grounded Khan text does not show relationship-dataset add/exclude; OA relies on Khan ¶¶0040-0041, 0050 which are not in the available excerpt — verify before relying.	Khan, claims 4-5
◐	simulated node dataset exhibits the correlation coefficient value (amended)Khan, Kursun	Arguably taught	Same combination concern as claim 1 — Khan is not directed to synthetic data generation (OA2).	Khan, claims 4-5; Kursun, claim 1

Claim 19 — §103 (Kursun in view of Wonus and Dang and Khan)

Status glyph	Claim element	Status	Disclosure / notes	Location
◐	receiving a node dataset representing a set of nodes and a set of relationships, each relationship associated with at least two nodes (method claim)Wonus	Arguably taught	Method-form counterpart to claim 1; same Wonus mapping and same contestability.	Wonus, claim 1; Wonus, claim 4
◐	each statistical metric associated with one or more nodes or one or more relationships; and at least one correlation coefficient value between a first and second node (amended)Kursun, Khan	Arguably taught	Same mapping and same contestability as claim 1. Claim 19 recites providing 'at least one of the inferred statistical dataset or the node dataset' to the graph generation model (an alternative rather than both) — for counsel to note the narrower/alternative input phrasing when comparing to the examiner's combination.	Kursun, claim 1; Khan, claims 4-5
◐	modifying the set of relationships by adding/excluding relationships based on correlation coefficient / statistical significance (amended)Khan	Arguably taught	Same posture as claims 1 and 10; grounded Khan text does not show relationship add/exclude, and the OA's cited ¶¶0040-0041, 0050 are not in the available excerpt — verify.	Khan, claims 4-5
◐	generated simulated node dataset exhibits the correlation coefficient value (amended)Khan, Kursun	Arguably taught	Same combination concern as claim 1.	Khan, claims 4-5; Kursun, claim 1

Claim 2 — §103 (Kursun in view of Wonus and Dang and Khan)

Status glyph	Claim element	Status	Disclosure / notes	Location
◐	determine a specialized generative model set associated with the data generation model, each specialized model associated with a particular model specializationKursun	Arguably taught	Grounded Kursun ensemble/subset concept is the closest disclosure. OA cites Kursun ¶¶0004, 0008 which are not in the available excerpt. For counsel to weigh whether an ML-model ensemble equates to a 'specialized generative model set' with 'model specializations.'	Kursun, claim 5 ('generate a machine learning model ensemble using a subset of the one or more additional machine learning models'); Kursun, claim 6
◐	generate, based on the inferred network, the constraints, and the model specialization set, a portion set, each portion corresponding to a portion of the network or constraints and associated with a model specializationKursun	Arguably taught	Significant stretch for counsel to weigh: OA's ¶0008 mapping of a model 'subset' to a 'portion set' corresponding to portions of the entity-relationship network is not grounded in the available Kursun excerpt. The 'portion of the inferred entity-relationship network' framing has no clear counterpart in grounded Kursun text (Kursun does not build an entity-relationship network).	Kursun, claim 6 (subset of most-accurate models)
◐	provide each portion to the corresponding specialized generative model to generate a model output set, and generate the simulated node dataset including the model output setKursun	Arguably taught	For counsel to weigh whether routing 'portions' to specialized models is disclosed vs. Kursun's ensemble-selection concept.	Kursun, claim 1 (GAN synthetic data); Kursun, claim 5

Claim 3 — §103 (Kursun in view of Wonus and Dang and Khan)

Status glyph	Claim element	Status	Disclosure / notes	Location
◐	determine a statistical metric between at least two values of the node valuesDang	Arguably taught	Dang's confidence values / conditional probabilities between nodes are grounded. For counsel to weigh whether these are a 'statistical metric between at least two values of the associated node values.'	Dang, claim 9 ('a confidence value for each of the one or more candidate evidential nodes'); Dang, description (conditional probabilities between nodes)
◐	the statistical metric includes at least one of a univariate, multivariate, conditional dependency, outlier, or time-series metricDang	Arguably taught	Because the claim is 'at least one of,' the conditional-dependency alternative maps to Dang's grounded conditional probabilities. OA's cited univariate/multivariate equations (¶¶0047-0051) are not in the available Dang excerpt, but the conditional-dependency alternative is independently grounded.	Dang, description ('conditional probability P(s_m1 \| s_m2)')
◐	provide the node dataset to the statistical inference model to generate a statistical value, and generate the inferred statistical dataset including that valueDang	Arguably taught	For counsel to weigh the mapping of Dang's confidence/likelihood computation to the claimed statistical value generation.	Dang, claim 9 (confidence value); Dang, claim 3 (TF-IDF and Bayesian modelling)

Claim 4 — §103 (Kursun in view of Wonus and Dang and Khan)

Status glyph	Claim element	Status	Disclosure / notes	Location
◐	determine time-series data associated with the node dataset, each data point associated with a particular timestampKursun	Arguably taught	Grounded Kursun 'monitor ... over a period of time' text supports data collected over time. OA's timestamp pin-cite (¶0052) is not in the available excerpt; for counsel to weigh whether per-datapoint timestamps are shown.	Kursun, description ('the system reaches out to the database and watches, observes, or checks the database ... over the period of time'); Kursun, claim 14 (continuous input data feed)
—	provide time-series data plus at least two of the dataset/network/constraints/statistical dataset to a temporal coherence model to generate a temporal dependency graph identifying a temporal dependency, sequence progression, or time-based correlationKursun, Dang	Not taught	The examiner expressly states Kursun fails to disclose the temporal dependency graph and turns to Dang for a generic 'dependency graph.' Per OA2, Dang's grounded text is directed entirely to symptom-disease conditional probabilities and does NOT address temporal sequencing or time-based data. The claim's alternatives are all temporal/sequence-based, which the grounded Dang text does not reach. See missing-element finding.	Office action (examiner: 'Kursun fails to specifically disclose ... generate a temporal dependency graph ...'); Dang, description (symptom-disease conditional probabilities; no temporal content)
—	provide the network, temporal dependency graph, and constraints to the data generation model to generate a simulated dataset consistent with all threeKursun, Dang	Not taught	Depends on the temporal dependency graph element above; if that is absent from the grounded art, this dependent step is likewise not reached. For counsel to weigh.	Dang, description (no temporal graph)

Claim 5 — §103 (Kursun in view of Wonus and Dang and Khan)

Status glyph	Claim element	Status	Disclosure / notes	Location
—	generate a constraint set with a first subset at a first priority level and a second subset at a second priority levelKursun, Wonus, Dang, Khan	Not taught	IMPORTANT: the provided office action excerpt is truncated after the claim 4 mapping, so the examiner's actual claim-5 mapping was NOT available to review — do not attribute any specific claim-5 reasoning to the examiner without confirming the full office action. None of the four fully-grounded reference excerpts discloses priority-leveled constraint subsets. For counsel to verify against the complete office action.	(none in grounded reference text)
—	determine a system status indicating a computational resource usage level, and determine a priority threshold level from the system statusKursun, Wonus, Dang, Khan	Not taught	No grounded portion of Kursun (GAN synthetic data), Wonus (enrichment graph), Dang (medical diagnosis), or Khan (PLNN inference) discloses computational-resource-usage-based prioritization. Candidate gap — but confirm against the full (untruncated) office action, which was not provided.	(none in grounded reference text)
—	determine first priority satisfies and second does not satisfy the threshold, then input only the first subset to generate a simulated dataset consistent with the first and not the secondKursun, Wonus, Dang, Khan	Not taught	Depends on the resource-status/priority-threshold concept above. For counsel to weigh once the full office action claim-5 mapping is obtained. NOTE ON OMITTED CLAIMS: rejected claims 11-14 mirror claims 2-5 in media form, and rejected claim 20 mirrors claim 2 in method form; they add no distinct limitations beyond those charted here and were omitted under the 8-claim cap. Claims 6-9 and 15-18 are listed as rejected but the specific rejection text/references/rationale were not in the provided office action and could not be charted.	(none in grounded reference text)

Elements not shown by the cited art (3)

Claim 4 — “generating a temporal dependency graph (via a temporal coherence model) that identifies at least one of a temporal dependency constraint, a sequence progression constraint, or a time-based correlation”: The examiner's own office action states Kursun 'fails to specifically disclose' the temporal dependency graph and relies on Dang for a generic 'dependency graph.' Both Kursun and Dang are fully grounded. Per OA2, Dang's available text is directed entirely to symptom-disease conditional probabilities (TF-IDF/Bayesian) and does not address temporal sequencing or time-based data, so it does not supply the temporal/sequence/time-based alternatives the claim requires. Neither asserted reference reaches the temporal-graph limitation on the grounded record — a §103 combination-gap candidate for counsel to weigh.
Claim 1 — “modifying the relationship dataset by ADDING an inferred relationship based on the correlation coefficient when not already present, and EXCLUDING a relationship when its explicit correlation is not statistically significant (amended limitation; also present in claims 10 and 19)”: Khan is the only reference asserted for this limitation and is graded fully grounded, but the grounded Khan text (claims 4-5, abstract, early description) discloses only the 'J' relative correlation coefficient bounded in [−1,1] modulating Fréchet inequalities within an inference network; it does not, in the available text, disclose modifying a data relationship dataset by adding/excluding relationships for graph construction. The examiner's cited support (Khan ¶¶0040-0041, 0050) is truncated and NOT in the available excerpt. Because the specific cited paragraphs were not verifiable in the provided text, this is a VERIFY-FIRST candidate: counsel should obtain and confirm exactly what Khan ¶¶0040-0041 and 0050 disclose before treating this add/exclude limitation as absent; the contextual gap (inference/belief-bound tightening vs. relationship-dataset modification for synthetic-data generation) is grounded in the available Khan text and is contestable now.
Claim 5 — “determining a system status indicating a computational resource usage level and deriving a priority threshold level used to select which constraint subset is input to the data generation model”: None of the four fully-grounded reference excerpts (Kursun GAN synthetic data; Wonus enrichment graph; Dang medical-diagnosis knowledge graph; Khan PLNN inference) discloses resource-usage-based constraint prioritization. Caveat: the provided office action was truncated before the claim-5 mapping, so the examiner's actual claim-5 rationale and reference reads could not be reviewed — counsel must confirm this gap against the complete office action before relying on it.

Rejection Map

§103Obviousness — claims 1, 2, 3, 4, 5, 10, 11, 12, 13, 14, 19, 20MPEP §2143(G)

Kursun US 2021/0049455 Wonus US 2023/0418878 Dang US 2022/0208373 Khan US 2025/0111206

Kursun is relied on for: processors and non-transitory media; inputting a dataset into a statistical inference model to generate an inferred statistical dataset with statistical metrics and metric values; generating constraints from inferred entity-relationship data (abnormal patterns); inputting constraints and inferred data into a GAN architecture to generate synthetic/simulated data; and transmitting simulated data for retraining models. Kursun is also relied on for dependent claims 2 (specialized generative model sets using subsets of ML models, GAN specialization) and 4 (time-series data collection over a period, timestamps in historical data). Wonus is relied on for: receiving a node dataset with entities, node values, and relationship values; and generating an inferred entity-relationship network including inferred relationships (e.g., deriving citizenship from birthplace) with an enrichment memory graph. Dang is relied on for: generating a knowledge graph with inferred evidential nodes having structural (graph data structure), semantic (semantic matching), and statistical (confidence score) properties; relationship labels as conditional probabilities between nodes; and for dependent claim 3, determining statistical metrics between node values using univariate and multivariate equations. Khan is relied on for the newly-amended limitations: correlation coefficient values expressing relatedness as correlation/anti-correlation of nodes; modifying the relationship dataset by adding relationships for positively-correlated nodes and excluding relationships for anti-correlated nodes; and the simulated dataset exhibiting the correlation coefficient values. The examiner interprets positive correlation values as added relationships and anti-correlation values as excluded relationships under Khan ¶¶0040-0041 and 0050. Motivations to combine: Wonus for improved searching and identifying related data through data enrichment; Dang for generating a graph based on likelihood of relatedness of graph nodes; Khan for using inductive reasoning to infer relationships between nodes.

§103Obviousness — claims 6, 7, 8, 9, 15, 16, 17, 18No rationale articulated

The office action states claims 1-20 are rejected, and claims 6-9 and 15-18 are not within the scope of the first rejection (claims 1-5, 10-14, 19-20). The specific rejection text, references, and rationale for claims 6-9 and 15-18 are not available in the provided document; the document text is cut off before these rejections are reached.

References Cited

Record & Grounding

Grounding Summary

Note

How each cited reference was grounded. A reference the analysis could only read through the office action’s characterization is flagged — its findings are limited to what the examiner said, not the reference itself.

SYSTEM AND METHODS FOR ITERATIVE SYNTHETIC DATA GENERATION AND REFINEMENT OF MACHINE LEARNING MODELSUS20210049455: Claim text retrieved
MULTI-MODEL ENRICHMENT MEMORY AND CATALOG FOR BETTER SEARCH RECALL WITH GRANULAR PROVENANCE AND LINEAGEUS20230418878: Claim text retrieved
INQUIRY RECOMMENDATION FOR MEDICAL DIAGNOSISUS20220208373: Claim text retrieved
PROBABILISTIC LOGICAL NEURAL NETWORK WITH INTERPRETABLE PARAMETERSUS20250111206: Claim text retrieved

Data Egress Log

Note

Your uploads stay in-boundary. External retrieval was limited to public patent-number lookups: 4 fetches. No claim text, no client material left the environment.

Patents fetched by number: US20210049455 US20230418878 US20220208373 US20250111206
Documents processed: fe4d208d-ef45-4522-9467-3300b45ea49f.pdfoffice action
8e4064fb-8ea7-4b57-bdb5-7e85b4b73e77.pdfclaims
Processed in-boundary — never transmitted externally.

Obviousness Framework

Field of endeavor: Computer-implemented generation of simulated/synthetic datasets from graph-structured (node-and-relationship) data, using statistical inference and graph-generation models to build an inferred entity-relationship network that then constrains a generative model, for the purpose of validating or training artificial-intelligence models.
PHOSITA: For argument purposes (a reasonable proposed construction for counsel to adopt or adjust, not a factual finding): a person with a bachelor's or master's degree in computer science, data science, or a related discipline, plus a few years of practical experience in machine learning, including familiarity with generative models (e.g., GANs), graph/knowledge-graph data structures, and applied statistical inference (correlation and dependency metrics). The office action does not itself articulate a PHOSITA level; any PHOSITA construction appearing only in commentary must be confirmed against the actual record before it is relied on.A construction for argument — not asserted as fact.

Reference	Analogous art	Rationale
Kursun	Analogous	Same field of endeavor: Kursun is directed to iterative synthetic/simulated data generation using a GAN to produce training data for machine-learning models, which is the core field of the claims. It is also reasonably pertinent to the inventor's problem of generating simulated data to train/validate AI models.
Wonus	Contestable	Wonus is directed to a multi-model enrichment-memory-graph and enrichment-catalog datastore for maintaining provenance/lineage and improving search recall — a data-cataloging/information-retrieval field distinct from synthetic-data generation. Whether it is 'reasonably pertinent to the particular problem' the inventor faced (generating simulated node datasets from inferred relationships) is contestable, and counsel may weigh that Wonus's stated aim (search recall) differs from the claimed goal. Notably, Wonus expressly disparages machine-learning approaches, which is relevant to both the analogous-art and the teaching-away inquiries.
Dang	Contestable	Dang's available text is directed entirely to a machine-guided medical-diagnosis system that builds a symptom-disease knowledge graph and computes disease likelihoods via TF-IDF/Bayesian modeling. This is a different field of endeavor (medical diagnosis). Whether the general knowledge-graph-with-inferred-nodes-and-conditional-probabilities teaching is 'reasonably pertinent' to the inventor's synthetic-data-generation problem is contestable and framed from the inventor's perspective, not hindsight.
Khan	Contestable	Khan is directed to a probabilistic logical neural network (PLNN) for inference and interpretability using belief bounds and Fréchet-inequality activation functions; per OA2 the available text is 'directed to inference and interpretability in machine-learning models, not to synthetic/simulated data generation.' Its correlation coefficient ('J' parameter) is a device for modulating probabilistic inference at logical operational nodes, not for modifying a relationship dataset to generate synthetic data. Same-field status is weak; 'reasonably pertinent' status is contestable and for counsel to weigh.

§103 rejection of claims 1-5, 10-14, and 19-20 over Kursun in view of Wonus, further in view of Dang, further in view of Khan. Kursun supplies the processors/media, statistical-inference-to-metrics, constraint generation from inferred data, GAN-based simulated-data generation, and transmission for retraining; Wonus supplies the node dataset (entities/values/relationships) and the inferred entity-relationship network via an enrichment memory graph; Dang supplies structural/semantic/statistical graph properties and conditional-probability relationship labels; Khan supplies the newly-amended correlation-coefficient limitations (adding relationships for positive correlation, excluding for anti-correlation, and the simulated data exhibiting the correlation coefficient).

Kursun + Wonus + Dang + Khan

Motivation asserted Wonus combined with Kursun to allow 'improved searching and identifying of related data through data enrichment' (Wonus ¶0005); Dang combined to allow 'generating a graph based upon likelihood of relatedness of graph nodes' (Dang ¶0003); Khan combined to allow 'using inductive reasoning to infer relationships between nodes' (Khan ¶0016).

Teaching awaymoderate
OA2 reports that the Wonus specification states machine-learning approaches to associating search terms are costly and unnecessary, and that user input/human expertise is 'less costly and more effective than using machine learning methods.' The claimed pipeline is built on machine-learning models (a statistical inference model, a graph generation model, and a data generation model). For counsel to weigh under MPEP § 2143.01, Wonus's own disparagement of ML may cut against a PHOSITA being led to insert Wonus's enrichment-graph teaching into Kursun's ML/GAN pipeline.
Hindsight reconstructionstrong
The examiner maps the amended correlation-coefficient limitations onto Khan's 'J' parameter, but Khan uses J to modulate Fréchet-inequality belief bounds for probabilistic inference in a PLNN — not to add or exclude edges in a graph for the purpose of generating synthetic data. The examiner's reading that 'positive correlation = added relationship' and 'anti-correlation = excluded relationship' is expressly framed as the examiner's own interpretation and appears to supply a mapping that Khan itself does not articulate, which counsel may argue is reconstruction using the applicant's amended claim as a template (MPEP § 2143.01).
Otherstrong
Claim-mapping mismatch for counsel to weigh: the claims recite EXCLUDING a relationship when the explicit correlation is 'not statistically significant,' whereas the examiner maps exclusion to Khan's anti-correlation (J approaching −1). An anti-correlation is a strong negative relationship, which is not the same as a correlation that is 'not statistically significant' (near J=0 / statistical independence in Khan's own scheme). The examiner's own summary of Khan describes J=0 as 'statistical independence' and −1 as 'maximum anti-correlation,' so mapping exclusion-for-insignificance onto anti-correlation may not correspond to what Khan teaches.
No reasonable expectation of successmoderate
OA2 flags that Khan's provided text 'is directed to inference and interpretability in machine-learning models, not to synthetic/simulated data generation,' and that the detailed description is truncated before the examiner-cited ¶¶0040-0041 and 0050. For counsel to weigh under MPEP § 2143.02, whether a PHOSITA would have a reasonable expectation of success in repurposing Khan's inference-oriented correlation coefficient to drive graph edge-modification and downstream synthetic-data generation is not established on the verifiable record, particularly where the cited passages could not be confirmed.
Conclusory motivationmoderate
Each asserted motivation restates the individual purpose of the reference being added (Wonus's search recall, Dang's likelihood-based graph, Khan's inductive inference) rather than articulating why a PHOSITA would integrate these disparate mechanisms — a GAN synthetic-data pipeline, a provenance/search enrichment graph, a medical-diagnosis knowledge graph, and a PLNN inference parameter — into the single claimed system. Under MPEP § 2143/§ 2143.01, counsel may argue the reasoning lacks a rational underpinning tying the pieces to the claimed combination as a whole.
Othermoderate
Verification gap for counsel to confirm against the record: OA2 reports that numerous examiner pin-cites cannot be located in the available reference text — including Kursun ¶¶0003, 0041, 0046-0050, 0052 and Fig. 4 item-level details, and Khan ¶¶0040-0041, 0050 (Dang's description is also truncated). Because much of the element-by-element mapping (including the core GAN/constraint mapping in Kursun and the entire correlation-coefficient mapping in Khan) rests on passages not verifiable in the provided text, counsel should independently confirm each pin-cite before assessing the prima facie case; this is a record-verification concern, not a merits conclusion.

§103 rejection of claims 6-9 and 15-18. The office action states claims 1-20 are rejected, but the provided document is cut off before the rejection text, references, and rationale for claims 6-9 and 15-18 appear.

Othermoderate
The available record does not contain the rejection basis, reference combination, or articulated motivation for claims 6-9 and 15-18 — the document text is truncated before these rejections are reached. No PHOSITA/combination scrutiny can be performed on this rejection until the full rejection text is obtained. Counsel should secure the complete office action before analyzing these dependent claims (which recute, e.g., topological-sort input ordering, Gibbs/Metropolis-Hastings/Variational-Inference sampling, relationship embeddings, and validation-report feedback).