Partially sanitized case study 06
Quality Control System for a Pear-Fruit Sparkling Water Production Line
Rejection reversed in reexamination
The response separated algorithm words from actual control-system architecture.
FIELD: AI/control systems / food productionSTAGE: PARTIALLY SANITIZED TECHNICAL REVIEWRESULT: Rejection reversed in reexamination
Review boundary
AI/control systems / food production
This is a public-record-based and partially sanitized technical-prosecution note prepared for peer-agency due diligence. Full file histories, claim amendments, cited references, client-identifying details, and client documents are shared only after NDA and conflict clearance.
Past outcomes illustrate reasoning methods only. They do not predict or guarantee any result in a future matter.
EXAMINER LOGIC
How the rejection framed the case
The examiner combined a dairy contamination detection model, an agricultural multi-source sensing method, and alleged common knowledge such as potential functions and gradient descent, treating the invention as an obvious transfer from detection optimization to beverage-line control.
FIP RECONSTRUCTION
How the response rebuilt the case
We dismantled the false equivalence: D1 outputs risk levels, while the invention outputs actuator action vectors; D1's gradient optimizes model hyperparameters, while the invention's gradient updates control actions; D1's disturbance is algorithm-search disturbance, while the invention's disturbance is an industrial-process disturbance. We also exposed the hindsight pattern: starting from the invention and searching backward for similar words.
OUTCOME
What changed procedurally
Reexamination was filed within 28 days after rejection; the rejection was revoked in about three months and the case proceeded to grant.
Deep technical note
Detailed English-only prosecution analysis.
This section expands the case beyond the homepage summary so foreign counsel can assess the reasoning pattern, not just the outcome.
Diagnostic read
- The examiner combined a dairy contamination model, an agricultural sensing method, and alleged common knowledge such as potential functions and gradient descent.
- The rejection treated shared vocabulary as shared technical teaching.
- The real mismatch was output and control meaning: risk-level prediction is not actuator-vector optimization for an industrial production line.
Response architecture
- Separate the outputs: D1 produced risk levels, while the invention produced control-device action vectors.
- Separate the gradients: model hyperparameter optimization is not the same as updating actuator actions.
- Expose hindsight: the rejection started from the invention and then searched backward for similar words.
Due-diligence takeaways
- AI/control inventions must be defended at the level of variables, outputs, and system effect.
- Common mathematical terminology should not be allowed to erase technical architecture.
- Reexamination can reverse a rejection when the combination depends on vocabulary matching rather than technical teaching.
What a peer firm can test
For a live matter, we normally ask for the relevant patent office or jurisdiction, prosecution stage, core rejection issue, principal cited references, current deadline, and a neutral technical summary. Client names and unpublished full documents can wait until NDA and conflict clearance are complete.
The first review focuses on whether the examiner has mis-modeled the technical problem, overstated a motivation to combine, relied on unsupported common knowledge, or missed an allowance route available through disciplined claim amendment.