Binding effect: how a mechanical contract acquires legal force

The gap, stated plainly

The standard says the signed artifact is a prose-plus-parameters-plus-specification tuple in which prose governs on conflict, and that the machine is a referee. Put those together and today’s honest position is:

A mechanical contract’s verdict is evidence about the parties’ agreement. It is not the agreement, and nothing in the standard makes it bind.

That is a respectable position - it is what referee mode means. But the headline claim is “executable and enforceable”, and the second half currently rests on nothing more than the ordinary enforceability of a signed document. Every other consequential thing in this language is declared explicitly. This one is silent.

There is a second silence behind it: the machine that judges is not a party. The fault model attributes infrastructure faults to “the host” and even suspends late-payment interest during an attributed host fault - a carve-out granted against someone who signed nothing and owes nothing.

The ladder

One declaration in the header, from a closed set, each rung an ordinary legal construct that long predates this standard:

binding-effect: expert-determination
  scope:     [ invoice-pricing, pool-ceilings, cadence ]   # which verdicts get this force
  carve-out: manifest-error | fraud                        # the standard escape
  prose:     #klausul-19                                   # the clause that says it

Rung	What the parties agree	The legal construct
evidential (default)	receipts are admissible evidence of what was submitted and judged; the tribunal weighs them freely	ordinary documentary evidence, strengthened by qualified electronic signatures (equivalent legal effect to handwritten) and qualified timestamps (a presumption of accuracy)
expert-determination	the machine’s verdict on declared questions binds the parties absent manifest error or fraud	expert determination: a named third party decides a defined question and the parties agree to be bound - a century of commercial practice, distinct from arbitration
conclusive-evidence	the receipt is conclusive as to the facts it records, absent manifest error	the conclusive-evidence certificate of commercial lending practice
self-executing	the verdict causes the transfer rather than recording it	a payment mandate - direct debit authority, payment initiation with strong authentication. The authority to move money comes from the mandate, never from the contract’s cleverness

Two disciplines make the ladder safe:

• Scope, not blanket. A rung applies to named verdict classes. Machine-checkable arithmetic - was the invoiced unit price equal to the contractual indexed rate? - is exactly what an expert determination should cover. “Was the delay excusable?” is a soft term and must never acquire mechanical force. The linter can enforce this: a rung above evidential may only scope verdicts whose rules are fully executable terms.
• Profiles forbid rungs that mandatory law would strike. A consumer profile refuses everything above evidential; a public-sector profile constrains the rest to declared classes.

The ceiling is a profile property

The obvious next question is whether the ladder can open up as this standard travels beyond Sweden. It can, and the shape matters:

Define the whole vocabulary in the core; let profiles permit rungs. If the core knew only two rungs, a future profile wanting a third would have to add one - and the covenant is that profiles constrain, never extend. Inverting it keeps the covenant: the core defines four rungs and permits none by itself, and each profile publishes its permitted subset.

mk-core        defines: evidential | expert-determination | conclusive-evidence | self-executing
  ├─ se-public      permits: evidential, expert-determination (scoped to executable verdicts)
  ├─ se-b2b         permits: evidential, expert-determination
  ├─ se-b2c         permits: evidential
  └─ xx-*           permits: whatever that jurisdiction's law actually supports

Why the ceiling genuinely varies, in one paragraph of honest generalization: common-law systems travel better up this ladder than civil-law ones. England and the jurisdictions following its commercial practice have long enforced both expert determination and conclusive-evidence certificates in commercial contracts, subject to manifest-error and unconscionability limits; the United States enforces determination and appraisal clauses in commercial dealings on a similar footing. Civil-law systems with free evidentiary assessment - Sweden’s, and the same principle across much of continental Europe and in Japan and Korea - give a court latitude that a private “conclusive” label cannot remove. The top rung is not a capability being withheld; it is a legal fact that varies, and a profile is exactly where varying legal facts belong.

Two consequences to design for now rather than retrofit: the verdict’s force must be readable off the artifact, because a counterparty abroad will ask what a receipt actually does to them; and cross-border pairs inherit the governing law’s ceiling, so where parties want more than that law supports, the answer is the ordinary one - arbitration at a seat that supports it - not a declaration the machine pretends to enforce.

The referee is a party

Add host (equivalently referee) to the party role classes, with:

• A seal and an identity, pinned like any counterparty.
• A practice statement as a governed artifact - what the host promises about ordering, receipt issuance, retention, archival re-timestamping, disclosure handling, and its conduct on its own faults; pinned by hash in the contract. The precedent is exact: certification authorities have published policy and practice statements under a standard framework for two decades, and qualified trust service providers carry statutory liability plus conformity assessment. A host that issues sealed judgments is that shape of animal.
• Its own ports and obligations - heartbeat emission, receipt delivery windows, archival attestations - judged and receipted like everyone else’s.

Then the control-liability carve-out stops being a hopeful sentence: an infrastructure fault suspends an obligation because a signatory undertook availability and failed, which is ordinary contract law rather than a favour the document grants an outsider. And “hosting without hostage” becomes true in a checkable way: a host whose duties are declared and receipted can be replaced by another conformant host without renegotiating the deal, because the practice statement is the swappable part.

Why this is not a power grab by the machine

Every rung above evidential is the parties choosing to give a mechanism force they already routinely give to third parties: valuers, adjudicators, certifying engineers, bank certificates, calculation agents. The mechanical contract’s advantage over all of those is not authority, it is auditability - the determination is reproducible, the inputs are signed, and manifest error is checkable by re-running the fold instead of cross-examining a professional. An expert determination you can replay is a strictly better expert determination.

The inverse is the real argument for declaring the rung at all: an undeclared force gets decided later, under pressure, by whoever has the better lawyer. Silence does not make the machine humble. It makes the outcome unpredictable.

The legal-review brief

Written to be handed to a lawyer as-is.

Scope. Does a declared binding-effect rung achieve its intended effect, per segment, under Swedish law - and what drafting is required in the prose clause it points at?

Facts. The artifact is a signed document (qualified electronic signatures and seals) containing prose clauses that govern on conflict, machine-executable terms, an append-only hash-chained log of submitted documents and machine verdicts each sealed and timestamped by a declared host, and a declaration naming what force those verdicts have between the parties. Verdicts are reproducible: any party can re-run the same inputs and obtain the same verdict bit-for-bit.

1. Expert determination. Can parties validly agree that a machine’s determination of a defined arithmetic question binds them absent manifest error - and is that agreement outside the arbitration act, so it neither requires arbitration formalities nor triggers the consumer invalidity rule? What must the clause say for the determination to bind contractually rather than merely evidentially?
2. Manifest error. Is “manifest error” workable in Swedish drafting, and does replayability change the analysis - can a reproducible computation be attacked more or less easily than a human expert’s opinion?
3. The evidential rung. Given free evidentiary assessment, what weight does a sealed, timestamped, hash-chained receipt realistically carry - and does an evidence agreement purporting to fix that weight bind a Swedish court at all? (Our assumption: not fully, which is why the default rung claims only admissibility.)
4. Consumers. Confirm that everything above evidential is unavailable against a consumer, so a consumer profile is right to refuse it at lint time.
5. Public sector. Does a public buyer face additional constraints in agreeing to expert determination on performance questions - limits on delegating decisions, or public-access implications of a determination whose record is a public document?
6. The host. If the host is a declared party with an availability undertaking, does the interest-suspension carve-out hold as ordinary contract law? Does issuing sealed verdicts make the host a trust service provider, with the liability that follows - and is that desirable or to be avoided?
7. Scope discipline. Is the linter rule sufficient - that a rung above evidential may only cover verdicts whose rules are fully executable, never ones touching soft terms? Where exactly does a determination stop being technical and start being adjudication?
8. Later: the same questions for one common-law jurisdiction and the United States, to test the profile-ceiling thesis.

Deliverable sought: per rung, a yes / yes-with-drafting / no for each of B2B, B2G and B2C, plus model prose for the two rungs intended to ship.

Honest creaks

1. Swedish evidence law is not English evidence law. Free evidentiary assessment means an evidence agreement has weaker footing here than a conclusive-evidence certificate has in an English commercial contract. The first two rungs are on solid ground; conclusive-evidence is the rung most likely to be trimmed by a real lawyer, and this document does not pretend otherwise.
2. Several rungs are simply unavailable against consumers, and a linter refusal is better than an argument after the fact.
3. Scope creep is the failure mode. The moment a rung’s scope quietly widens to cover a verdict that depends on a soft term, the machine is adjudicating reasonableness - the exact thing the executable/soft tier split exists to prevent.
4. A declared host is a named dependency. Pinning a practice statement makes host-switching an amendment, and a contract whose host disappears needs a declared succession path: who re-seals, who holds the archive.
5. This document needs a lawyer, not a review pass - an actual opinion, jurisdiction by jurisdiction. The constructs it leans on all exist; whether these words achieve that effect is not a question a specification can answer by itself.

This RFC is exploratory. It is published as a basis for discussion, not as normative text; where it touches the standard, the standard governs until a decision is recorded.