›Methodology

The subject of every mapping is a SPARTA technique: a documented adversary technique against space systems, published by The Aerospace Corporation. The object is one of four things: an article or sub-paragraph of EU regulation, an operational control from the ENISA Space Threat Landscape, a control or technique in a general framework such as NIST 800-53 or MITRE ATT&CK, or a SPARTA countermeasure.

Mappings are made at the most specific level the source supports. Where a regulation article has numbered sub-paragraphs, the mapping targets the sub-paragraph rather than the article as a whole, because the sub-paragraph is what an entity actually has to satisfy.

Canonical reference data is imported from authoritative published sources and is never hand-edited. Every source is recorded in a manifest with its retrieval date and credibility metadata, and a source that is not in the manifest does not exist for the platform. Re-importing a framework upserts on the source's own stable identifier, so an entity keeps its identity and its mappings across version updates.

Snippet-derived content, non-manifest blogs, and AI-summarised regulatory text are never treated as authoritative. Where a source is a proposal rather than adopted law, as the EU Space Act currently is, entries carry that status and the document version they were read from.

The corpus is not uniform, and the difference matters if you are citing it.

Imported mappings. Most edges are SPARTA's own published cross-reference and starter mappings, taken wholesale with SPARTA's relationship and confidence values. They carry an importer provenance note in place of a rationale, and SafeMode Space did not adjudicate them individually. They are reproduced because they are useful and attributable, not because they were reviewed here.

Reviewed mappings. The rest are SafeMode Space's own contribution. Each was staged as a proposal, checked against the cited source text, and carries the written reasoning that justified it. Every technique-to-regulation-article and technique-to-ENISA-control mapping is in this group, along with a reviewed subset of the technique-to-control mappings.

You can tell them apart without trusting this page: the rationale field in the corpus download carries either an importer provenance note or written reasoning. The exact split, with current counts, is stated at the top of llms.txt, regenerated on every build.

A candidate mapping is proposed against a named source excerpt and enters a review queue. It is not published at that point and is not a mapping yet. Review compares the claim against the excerpt and resolves it one of several ways: accept, accept with the rationale rewritten, reject, or escalate for a human decision. Only an accepted proposal becomes a published edge, and the decision is recorded with its reasoning, the evidence tier, who decided it, and when.

Most review decisions were made by an LLM curation session following a documented methodology, with escalated and contested cases decided by the maintainer. The recurring failure modes that methodology exists to catch are scope errors: a control selected because its name resembles the technique rather than because its scope covers it, a governance obligation asserted as active mitigation, or a cross-tactic mapping that violates kill-chain ordering. Scope rules for the controls that attract those errors are written down and applied consistently rather than re-litigated per case.

Rejected and withdrawn proposals are retained rather than deleted, so a mapping that was considered and refused stays distinguishable from one that was never considered.

Identifiers, names, and structural references are reproduced with attribution. Verbatim third-party text is gated on the source's reuse terms. The text of EU regulation articles and of ENISA controls is not reproduced at all; those pages carry the citation and the mappings and point you to the official source for the wording. Where a source's reuse terms are not confirmed, its identifiers are published and its descriptive text is withheld with a notice saying so.

This is a licensing obligation, not a paywall. Nothing on this site is gated on having an account, and there is no paid tier.

Read these before citing. Most of the corpus by volume is imported, not individually reviewed here. If you are citing a specific edge as adjudicated, check its rationale field first.

Most review decisions were LLM-made under a documented methodology, with human decisions on escalated and contested cases. That is a deliberate choice about throughput, and it is why every decision carries its reasoning: the reasoning is the thing you should evaluate, not the decision-maker.

A mapping is an argument, not a legal conclusion. Whether a given obligation applies to a given operator depends on facts about that operator that this corpus does not know. Nothing here is legal advice.

The regulatory surface moves. The EU Space Act is a Commission proposal, not adopted law, and its entries are marked provisional. Counts change as curation continues, so figures quoted from this site should carry the access date.

Coverage is uneven by construction. Frameworks with procedural sub-paragraph structure received a supplementary mapping pass; those without received only a primary pass. Absence of a mapping is weaker evidence than the presence of one.

Every mapping on the site shows its relationship, confidence, evidence type, and rationale inline, so a claim can be judged at the point it is made. See EX-0014 Spoofing for a technique with mappings across all four target types. The full corpus is downloadable as JSON or CSV, which is the same data the pages render.

Corrections are welcome and are the fastest way to improve the corpus. If a mapping is wrong, say which edge and why, and cite the source text you are reading it against.