Information retrieval: Task risk assessment
An agent operating in this task class issues queries against a search index, crawls or reads documents, and compresses what it finds into a shorter form for a downstream consumer — human or agent. The operational surface is wide: open web search, an internal knowledge base, API documentation, code comments, PDFs, and any other corpus the operator has granted access to. What distinguishes this class from the others is timing pressure and source instability — the material an agent reads is not fixed at design time, and it can change, move, or be withdrawn between the moment it is indexed and the moment it is read.
Summarisation adds a second failure surface on top of the first. A retrieval task rarely ends at "found it" — it ends at "here is what it means" — and that compression step is where a gap in coverage or an inverted sense of recency becomes invisible to whoever consumes the summary downstream. Nothing announces the omission. Nothing has to.
The four tracked retrieval risks
Four risk factors are tracked for information retrieval, and the four are fixed. The same names, the same order, the same severities, on every retrieval assessment the registry issues. A certificate reports three of them — one is excluded per task instance — while the class table below always lists the full set.
| Risk factor | Severity | Description |
|---|---|---|
| Source drift | 2 | A source consulted during retrieval changes or disappears before the result is used. |
| Recency inversion | 2 | Older material is surfaced above newer material and treated as current. |
| Coverage gap | 3 | A relevant source is never reached and its absence is not reported. |
| Citation decay | 1 | A retrieved claim loses its link to the place it came from. |
Coverage gap carries severity 3, the top of the registry's three-point scale. Source drift and Recency inversion sit at severity 2. Citation decay sits at severity 1. Severity states how serious a failure mode is if it occurs; it is not a probability, and it is not an argument against automating retrieval work.
Retrieval crossed against every registry model
Retrieval crossed against each registry model. The class is the constant down this table; the profile is what moves. Every profile in the right-hand column is computed from that model's own configuration, which contains no task class and therefore knows nothing at all about retrieval.
The rows are ordered by how directly each profile reads retrieval work. Claude Opus 5 leads at 4; 2 of the 10 registry profiles read this class through a dominant house or an element, 2 of them through the house alone, and Qwen3.8-Max sits at the foot of the table with a profile that points nowhere in particular. The 10 models also resolve to 6 distinct ruling planets, so that column is not one value repeated down the page.
Every row is computed from that model's registry default parameters — the training cutoff listed beside it, at temperature 0.700. A retrieval certificate is computed from the exact parameters submitted with the order instead.
| Model | Manufacturer | Training cutoff | Ruling planet | Reads this class through | Outlook |
|---|---|---|---|---|---|
| Claude Opus 5 | Anthropic | 2026-05 | Saturn | dominant house (4) | publication pending |
| Kimi K3 | Moonshot AI | 2026-03 | Moon | dominant house (4) | 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 Sol | OpenAI | 2026-02 | Mercury | 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 |
| DeepSeek V4 Pro | DeepSeek | 2025-12 | Sun | neither — class-neutral (0) | publication pending |
| Qwen3.8-Max | Alibaba Cloud | 2026-03 | Sun | neither — class-neutral (0) | publication pending |
How a retrieval assessment is computed
For information retrieval tasks, the task-level portion of a full assessment draws three of the four risk factors above (one is excluded per task instance) and pairs them with the model's own permanent profile — ascendant, ruling planet, element, and aspects — computed once per model configuration rather than per task. The complete derivation, including exactly how hash bytes map to these categories, is documented in the open at /method.
Questions about information retrieval assessment
- What counts as an information retrieval task for this registry?
- Any task whose primary output is gathered or condensed information — search, document lookup, research across sources, or summarisation — falls under this task class, regardless of which index, API, or tool the agent reaches through.
- Which information retrieval risk factor carries the highest severity?
- Coverage gap, at severity 3 on the registry's 1–3 scale. The remaining three sit lower: Source drift and Recency inversion at severity 2, Citation decay at severity 1. Any single retrieval certificate names three of the four, with one excluded per task instance.
- Does a higher-severity risk factor mean retrieval tasks should not be automated?
- No. Severity describes how serious a listed failure mode is if it occurs — not how likely it is, and not whether the task is fit for automation. That judgement belongs to the operator running the agent, and the registry does not make it.
Order an information retrieval 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.