What Exactly Does This Skill Do? A Quick Word First
Anyone working in the salmon processing industry will surely understand this: once a grader machine is up and running, the biggest fear is "Punto Cero"—the rate of rejected product being discarded skyrocketing. When the rejection rate climbs, the entire line's profit vanishes, and operators often only realize there's a problem by looking at reports after the fact. The skill introduced today is specifically designed to address this pain point. It enables operators to see real-time data during their shift, allowing for early intervention instead of regretting things only after clocking out.
This skill is called Analisis Grader — Guia de uso en turno. Simply put, it's an operational guide plus a decision-support tool that walks you through how to best utilize this analysis module over the course of an entire production shift. It comes from a repository on GitHub called mantenimiento-planta, authored by orelcain. Although the name is in Spanish, its functionality is very practical—it helps you understand the grader's data, letting you know when to think and when to act.
Core Workflow: Six Steps, No Need to Rush
When you open this skill, you'll find it breaks down the entire shift workflow into six clear steps. Following them myself, it felt like having someone guide you by hand. The first step is exporting Excel files from the Matrix software. Among these, PIEZA_PIEZA is essential—it's the "source" of all data, containing information on each fish (or fillet), such as weight, grade, quality, gateway, and batch. The other file, PUERTA_0, is optional and mainly supplements rejection reasons, like photoelectric sensor triggers or out-of-spec items.
The second step is dragging and dropping these two files into the upload area of /analisis-grader. The system will automatically recognize and start the analysis, with minimal manual button-clicking required—a godsend for those who prefer convenience. The third step is checking the KPI cards at the top. Here, the most critical metric is % Punto Cero, the rejection rate. It gives you three color bands: green for below 2%, yellow for between 2% and 3.5%, and red for anything above 3.5%—once it turns red, it signals a major problem requiring immediate attention.
The fourth step is reviewing the Trends tab. I strongly recommend every operator keep an eye on this page during their shift. It uses a linear regression algorithm to predict the end-of-shift situation based on currently accumulated data. You can see how many minutes remain, the predicted rejection rate, forecasted total production, and whether the weight trend is moving up or down. If the predicted rejection rate exceeds 3.5%, the page will display a warning directly and automatically suggest recommended actions, such as adjusting the configuration for certain gateways.
The fifth step is execution. If the rejection rate is indeed above the threshold, you go to the Compuertas (Gateways) or Sugerencias (Suggestions) tab, click the "Apply" button, and the system updates the gateway configurations in real-time. Finally, the sixth step: remember to click "Save Session" at the end of the shift. This allows you to compare data across different shifts later, helping you see whether the adjustments were effective.
Data Architecture and KPI Interpretation: Don't Stay a Novice
Going a bit deeper, this skill also thoroughly explains the data architecture. PIEZA_PIEZA is the core table, where each record represents a fillet's weight, grade, quality, gateway, batch, and timestamp. PUERTA_0 supplements rejection reasons, such as whether the photoelectric eye detected an irregular shape or the item was ejected for exceeding weight limits. Gateway 0 is the "zero point"—the rejected batch—while gateways 1 through 12 are the normal sorting outputs, each assigned specific weight grades and quality standards.
Regarding KPIs, besides the rejection rate mentioned above, there's also the weight coefficient of variation (CV), which reflects the stability of fillet weights. A CV below 15% is green, 15% to 25% is yellow, and exceeding 25% suggests the sensors might be malfunctioning and need inspection. Another metric is the HHI concentration index, which checks whether production is overly concentrated on a single output. If one gateway is consuming all the output, it indicates a problem with the sorting logic that needs adjustment.
I tried importing a real dataset myself and found the trend panel genuinely useful. It displayed "Time remaining: 3 hours 12 minutes," "Predicted rejection rate: 4.8%," and next to it a small red badge—making it instantly clear that immediate action was required. The system's suggestion was "lower the weight threshold for Gateway 3 by 2.5 grams," backed by historical data evidence, noting that a similar adjustment previously reduced the rejection rate by 0.8%. These data-driven recommendations are far superior to guessing on a whim.
When to Trust the AI, and When Not to Take It at Face Value
This skill also dedicates a section to AI analysis. In the trend panel, there's an "Analyze Now" button. Clicking it runs an analysis and outputs prioritized recommendations (high/medium/low), each supported by data evidence. It also logs historical runs, allowing you to check the consistency of previous analyses. If consistency exceeds 70%, it indicates the AI's judgment is stable and can be trusted confidently; if it falls below 40%, treat it as a mere reference and rely on the actual on-site situation.
I see it as having an experienced senior supervisor watching over your shoulder—but even a veteran isn't always right, so you need to keep your own judgment. During my shifts, I typically check the trend panel every half hour. As soon as I notice the predicted rejection rate approaching the yellow zone, I proactively inspect the gateway configurations rather than waiting until the red light flashes and scrambling.
Conclusion
Overall, this skill is a highly practical tool for production staff in salmon processing plants. It transforms complex production data into actionable guidance, and the learning curve is low—if you can drag files and read numbers, you can get started. If you're a shift lead or operator responsible for graders, I strongly recommend installing it and giving it a try. It just might help you bring a shift's rejection rate down from 4% to under 2%—and that... well, that's a game-changer worth having.
