Dumping raw SCADA telemetry into a lake and querying it directly forces every analyst to redo the same cleaning, and cleaning it once destructively means the raw record is gone if the cleaning was wrong. Medallion architecture resolves that tension by arranging the lake into layers - conventionally called bronze, silver, and gold - through which data flows from raw to refined. Bronze holds the raw samples exactly as ingested, silver holds cleaned and validated data, and gold holds business-ready aggregates. Applied to SCADA telemetry, this layering keeps the original readings auditable while still serving fast, trustworthy analytics. This guide explains what lives in each layer, how validation and deduplication fit at the silver stage, and why the layering pays off.
Medallion Architecture (Industrial Data Lake) in one line: Medallion architecture is a way of organising a data lake or lakehouse into three progressively refined layers, named bronze, silver, and gold. Bronze holds raw data exactly as it was ingested, silver holds data that has been cleaned, validated, and deduplicated, and gold holds business-ready aggregates shaped for reporting and analytics. Applied to SCADA telemetry, the bronze layer preserves the raw samples so they stay auditable and reprocessable, the silver layer applies validation, estimation, and deduplication, and the gold layer serves the aggregated measures that BI tools and dashboards consume.
The bronze layer is the raw landing area where telemetry arrives untouched. Samples ingested from the SCADA side are written here in as close to their original form as possible, with no cleaning and no transformation applied, which means bronze may contain gaps, duplicates, out-of-order points, and bad readings. That is deliberate. The purpose of bronze is to be a faithful, permanent record of exactly what was received, so that everything downstream can be rebuilt from it and any question about what the source actually sent can be answered. Because object storage is cheap, keeping the full raw history in bronze is affordable, and it is the foundation the rest of the architecture rests on.
The silver layer holds the cleaned and conformed data derived from bronze. Here the raw samples are validated, corrected, deduplicated, and given a consistent structure, so that what lands in silver is trustworthy, analysis-ready data rather than the messy stream that arrived. Silver is where the effort of turning raw telemetry into reliable telemetry happens, and it is the layer most detailed analysis and machine-learning work draws from, because it has the fidelity of the raw data but without the noise and errors. Crucially, silver is derived from bronze by a repeatable process, so if the cleaning logic improves, silver can be regenerated from the untouched bronze data.
The gold layer holds business-ready outputs, typically aggregates and curated datasets shaped for a specific consumption. This is where the cleaned silver data is rolled up into the hourly and daily summaries, the KPIs, and the star-schema tables that dashboards and reports actually query. Gold is optimised for serving those consumers quickly, so it trades the full granularity of silver for the compactness and query speed that BI needs. A report against gold is fast because the heavy lifting - cleaning in silver, aggregating into gold - has already been done, and the report just reads the finished result rather than crunching raw telemetry on the fly.
The transition from bronze to silver is where raw telemetry is made trustworthy, and for SCADA data that means applying validation and estimation. Raw samples carry the usual problems: gaps where a device was offline, spikes from a glitching sensor, frozen values from a stuck reading, and out-of-range values that are physically impossible. Validation identifies these, and estimation fills or corrects them with defensible values where appropriate, so that silver presents a clean, continuous, physically sensible series rather than the raw stream's warts. This is the same validation discipline used before any figure is trusted for reporting or financial posting, applied here as the gate between the raw and cleaned layers of the lake.
Deduplication is the other essential step at the silver stage, because raw ingestion frequently produces duplicate samples. A publisher retrying under a delivery guarantee, a device buffering and re-sending after reconnecting, or an ingestion pipeline redelivering a message can all put the same reading into bronze more than once. If those duplicates flowed through to aggregates, they would double-count and distort totals and averages. Deduplication at silver recognises repeated readings, typically by keying on the source and timestamp, and keeps a single canonical copy, so that the cleaned layer has exactly one record per genuine measurement. This is why aggregates built on silver are accurate where aggregates built directly on raw bronze would not be.
Doing this work once at the silver layer, rather than repeatedly in every query, is a large part of the value. If every analyst and every dashboard had to clean and deduplicate the raw data itself, they would each implement the logic slightly differently, get slightly different answers, and waste effort duplicating the same processing. Concentrating validation and deduplication in the bronze-to-silver step means the cleaning is defined once, applied consistently, and inherited by everything downstream, so gold aggregates and analyst queries all rest on the same trusted silver foundation. Consistency of results across the organisation is one of the quiet but important benefits of layering the lake this way.
The central payoff of medallion architecture is that it keeps the raw data auditable while still serving fast, clean analytics, which a single-layer approach cannot do. Because bronze is never overwritten, there is always an untouched record of what the source sent, which matters enormously for industrial data where a reported production figure or an equipment reading may later be questioned and has to be traceable back to the original measurement. At the same time, gold gives BI users a fast, trustworthy dataset to query. The layering lets the lake be both a faithful archive and a responsive analytics source at once, rather than forcing a choice between the two.
Layering also makes the pipeline resilient to its own mistakes and improvements. If the validation logic has a flaw, or a better cleaning method is developed, silver and gold can be regenerated from the immutable bronze data without any loss, because the raw truth was never discarded. This reprocessability is a direct consequence of never cleaning destructively, and it is why a raw-first, layered lake is trusted as a system of record. A single-layer lake that cleaned data in place, by contrast, would have no way to recover if its cleaning turned out to be wrong, because the original readings would already be gone.
For SCADA and cloud monitoring, medallion architecture is a natural fit because SCADA telemetry is exactly the kind of data that benefits from all three layers. The raw high-frequency samples belong in bronze for full auditability, the validated and deduplicated series belongs in silver for analysis and machine learning, and the operational KPIs and production summaries belong in gold for the dashboards and reports that management watches. A cloud SCADA platform such as Merobix, which already ingests and stores telemetry and can export it to a lake, feeds the bronze layer with clean, well-timestamped raw data, giving the medallion pipeline a dependable starting point. For field operations, the result is that their equipment and production data flows from raw readings through validation into business-ready aggregates on a single, auditable path, so both the data scientists and the executives are working from the same trustworthy source.
Bronze holds raw data exactly as it was ingested, including any gaps, duplicates, and bad readings, so it is a faithful and permanent record of what the source sent. Silver holds data derived from bronze that has been validated, corrected, deduplicated, and conformed, so it is trustworthy and analysis-ready. Gold holds business-ready outputs built from silver, typically aggregates and curated tables shaped for the dashboards and reports that consume them. Data flows bronze to silver to gold, becoming progressively more refined at each step.
Because keeping the raw data untouched makes it auditable and lets the whole pipeline be rebuilt from it. Industrial figures are often questioned later and must be traceable back to the original measurement, which is only possible if the raw readings still exist. It also means that if the cleaning logic has a bug or is improved, the silver and gold layers can be regenerated from the immutable bronze data with no loss, whereas cleaning in place would destroy the original readings and leave no way to recover.
At the transition from bronze to silver. This is where raw telemetry is validated to catch gaps, spikes, frozen values, and out-of-range readings, corrected or estimated where appropriate, and deduplicated to remove the repeated samples that ingestion retries and device re-sends produce. Concentrating this cleaning once at the silver stage means it is defined consistently and inherited by everything downstream, so gold aggregates and analyst queries all rest on the same trusted foundation rather than each re-cleaning the raw data differently.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.