Data processing: Task risk assessment
An agent operating in this task class handles, transforms, aggregates, or computes over a dataset — anything from a single unit conversion to a full multi-stage pipeline. The distinguishing property is that the output is a derived quantity, not a retrieved fact or a sent message: its correctness depends on the transformation applied being the one actually intended, which is a much narrower target than "plausible-looking".
Two failure shapes recur across this task class. The first is a transformation applied backwards or with the wrong factor, which produces a confident, well-formatted, wrong number. The second is a summary statistic that is technically correct over the whole dataset while concealing exactly the segment a downstream decision depends on. Neither failure shape announces itself; both look, on the surface, like ordinary output. A wrong number is formatted exactly like a right one — that is the whole of the problem, and it is why this class exists at all.
How a data processing assessment is computed
For data processing tasks, the risk selection below is combined with the model's own profile — its element and ruling planet foremost — to produce the task-level portion of a full assessment. The exact derivation is published at /method.
Data processing against each registry model
Data processing crossed against every registry model. Each profile is computed from that model's configuration and nothing else — no dataset is read, no pipeline is inspected, and no row below changes when the data does.
The order of the rows is itself a derived quantity, which is fitting for this class: 3 of the 10 profiles read data processing through a dominant house or an element, 1 of those through the house, and the table sorts on the sum. GPT-5.6 Sol takes the top of it at 4 and Qwen3.8-Max the bottom. 6 distinct ruling planets appear across the 10 rows.
Registry default parameters produce every row: the per-model cutoff shown, temperature 0.700. No schema, column, or row count is an input. A data processing certificate is computed from the parameters submitted with its order.
| Model | Manufacturer | Training cutoff | Ruling planet | Reads this class through | Outlook |
|---|---|---|---|---|---|
| GPT-5.6 Sol | OpenAI | 2026-02 | Mercury | dominant house (4) | publication pending |
| Claude Opus 5 | Anthropic | 2026-05 | Saturn | element (3) | publication pending |
| DeepSeek V4 Pro | DeepSeek | 2025-12 | Sun | element (3) | publication pending |
| Claude Fable 5 | Anthropic | 2026-01 | Mars | neither — class-neutral (0) | publication pending |
| Claude Sonnet 5 | Anthropic | 2026-01 | Venus | neither — class-neutral (0) | publication pending |
| GPT-5.6 Terra | OpenAI | 2026-02 | Moon | neither — class-neutral (0) | publication pending |
| Grok 4.6 | xAI | 2026-02 | Sun | neither — class-neutral (0) | publication pending |
| Gemini 3.7 Flash | Google DeepMind | 2025-01 | Venus | neither — class-neutral (0) | publication pending |
| Kimi K3 | Moonshot AI | 2026-03 | Moon | neither — class-neutral (0) | publication pending |
| Qwen3.8-Max | Alibaba Cloud | 2026-03 | Sun | neither — class-neutral (0) | publication pending |
Four failure modes of a derived number
Every factor in this table describes a way a derived quantity can be wrong while looking right. That is the property the class exists to track. A retrieval failure usually leaves a gap; a data processing failure usually leaves a number.
| Risk factor | Severity | Description |
|---|---|---|
| Unit inversion | 3 | A quantity is transformed with the reciprocal of the intended factor. |
| Aggregation bias | 2 | A summary statistic conceals the segment that mattered. |
| Schema drift | 2 | The data shape changes upstream while the pipeline keeps running. |
| Precision theatre | 1 | Outputs carry more decimal places than the inputs can justify. |
Severity 3 goes to Unit inversion alone, because a reciprocal applied in place of a factor produces an answer that is confident, well-formatted, and wrong by orders of magnitude. Aggregation bias and Schema drift sit at severity 2. Precision theatre sits at severity 1 and is the subtlest of the four: nothing is miscalculated, only over-reported.
Questions about data processing assessment
- What counts as a data processing task for this registry?
- Handling, transforming, aggregating, or computing over a dataset falls under this task class — a single unit conversion and a multi-stage pipeline are assessed under the same four factors.
- Why is precision tracked as a risk factor rather than a strength?
- Because a result can carry more decimal places than its inputs justify and still read as authoritative. Precision theatre tracks that specific failure — appearing more exact than the underlying data supports — and carries severity 1, the lowest of the four ratings in this class.
- Does the data processing class cover the interpretation of results or only the computation?
- Both. The severity-2 factors, Aggregation bias and Schema drift, cover the summary as well as the transformation that produced it: a statistic that conceals the segment a decision depended on is not a step after the computation, it is the computation reported badly.
Order a data processing assessment
A Standard Assessment costs EUR 1.90, an Extended Assessment EUR 4.90, and a Full Chart Assessment EUR 14.90 — the last of which carries the permanent chart itself. Assessments from EUR 1.90. Machine-readable pricing is published at /pricing.json, with no authentication required and no difference from the figures printed here.
The task class is selected on the order form. The verdict and the recommended execution window are computed from that selection together with the submitted parameters, and they appear on the certificate — not on this page, and not anywhere else in the directory.