standards.sgit.ai / method / crosswalks

Bridges, not merges

A crosswalk says this provision over here bears on that provision over there. There are two ways to build one, they look similar on a slide, and they fail completely differently. This page argues for the one that fails visibly.

Before the argument, the fact: there are no crosswalks on this site. Zero. Not AI Act ↔ GDPR, not ISO 27001 ↔ SOC 2, not GDPR ↔ NIS2 incident reporting — all three named as obviously valuable in the project's own material, none built. FrameworkReference, the bridge node type, is specified and never instantiated. This page is a design, and the shipped page says so too.

Two ways to relate two instruments

A mergeA bridge
What it assertsThese two provisions mean the same thingThese two provisions stand in a stated relationship, on a stated basis
Where it livesInside a merged nodeAs an edge, with its own properties
Can a reader disagree with it?Not usefully — the disagreement has nowhere to attachYes — the basis is written down and can be argued with
What happens when one is amendedA silently wrong node. The merged meaning is now false and nothing indicates itA wrong edge you can find — the content hash moved under a bridge attached to the positional hash
Failure modeInvisibleVisible and locatable

A merge that is wrong is a silently wrong node. A bridge that is wrong is a wrong edge you can find and fix.

And two instruments almost never mean the same thing. GDPR Article 32 requires security appropriate to the risk; ISO/IEC 27002 describes an encryption control. Those are related. They are not the same statement, they do not have the same addressee, and one does not discharge the other. A merge asserts an equivalence that is not there; a bridge records the relationship that is.

Why bridges survive amendment

This is where the two-hash scheme stops being a storage detail and becomes the enabling decision.

A bridge attaches to the positional hash — the identity of the slot. When an instrument is amended, the wording moves and the content hash changes; the slot holds, and so does every mapping into it. Attach a crosswalk to text and you have a spreadsheet that rots. Attach it to a position and you have an artefact that survives revision and can report that the text underneath it moved.

bridge --attached-to--> positional hash   (stable)
                            |
                        content hash       (moves on amendment)
                            |
                        => the bridge is FLAGGED, not broken

That last line is the product. A mapping that knows it may be stale is worth having; a mapping that cannot tell is worse than none, because it looks current.

Every bridge carries a basis and a strength

A crosswalk that can only say yes is either wrong or useless. The project's own phrase for what is needed is “shades of compliance” — and there is no scale behind it yet. That is an open question, published as one, and it blocks the crosswalk browser.

The shape a bridge must carry, whatever the scale turns out to be:

FieldWhy it is mandatory
from / toPositional hashes, both ends. Never labels
relationFrom the published grammar — and if no established edge fits, an argument for the new one
basisProse. Why somebody drew this edge. A bridge with no basis cannot be argued with, which makes it a merge wearing an edge's clothes
strengthHow much of the target this actually discharges. Undefined — Q4
asserted_by / asserted_atA crosswalk is an opinion. Opinions have authors and dates
source_content_hashThe content hash at each end when the bridge was drawn. This is what makes staleness computable rather than assumed

The compounding argument

the AI standard already links to another of the standards, another acts that you need to be able to access as a graph… Map a second instrument and the edge resolves; map a third and two more resolve. So each mapping increases the value of the ones already done.

Concretely: the EU AI Act refers out to GDPR, to NIS2, to harmonised standards, to conformity assessment bodies and to notified bodies. Every one of those references is currently a node pointing at nothing. Model GDPR and a whole class of them resolves — and the AI Act graph gets better without anybody touching it.

The same conclusion was reached independently, years earlier, by somebody else, which is worth stating on a page arguing for a method:

A single graph can include nodes from GDPR, PCI DSS, ISO 27001, OWASP ASVS, etc. If two standards address the same concept (say “access control”), the graph can link them to a common concept node. The University of Maryland's researchers who built an integrated GDPR + PCI DSS ontology did exactly this.

The rest of the prior art →

What must be true before a crosswalk browser ships

  1. One crosswalk exists. Any pair. AI Act ↔ GDPR is the natural first — the two apply concurrently, so the relationship is real rather than academic.
  2. FrameworkReference is instantiated. Specified since July, built never.
  3. A basis and a strength on every bridge — which needs the scale that does not exist. Q4.
  4. The contract version is stamped in each instrument's MANIFEST.json, so a tool can refuse a shape it does not speak.
  5. The citation scheme is decided, because every bridge is a pair of citations. Done — the scheme is on the next page.

Five preconditions, one met. The crosswalk browser is T1 and it is the thing a multi-standard site sells; it is also the one tool that cannot be faked, because with zero bridges it renders an empty screen.