onTerms.org

Dispute Resolution

topic
refonterms:dispute:1.0.0:EW
urlhttps://onterms.org/t/dispute/1.0.0/EW
urnurn:onterms:dispute:1.0.0:EW:sha256-4bb2129534121f2188fd64da9ed113eed4472cac9b0f73df15be4b1e7d8ecc6c
sha-256sha256:4bb2129534121f2188fd64da9ed113eed4472cac9b0f73df15be4b1e7d8ecc6c

Incorporation clause

This Order incorporates by reference the onTerms Standard Terms onterms:dispute:1.0.0:EW, available at https://onterms.org/t/dispute/1.0.0/EW and verifiable against SHA-256 sha256:4bb2129534121f2188fd64da9ed113eed4472cac9b0f73df15be4b1e7d8ecc6c. Each party acts in the course of its business and not as a consumer.

Adopt-verbatim. The only negotiable surface is the typed Elections. The body is pinned by the hash above. Re-hash the canonical envelope offline to verify it has not changed.

onTerms Dispute Resolution Module

onterms:dispute:1.0.0:EW · Status: v0.9 DRAFT · Not legal advice (requires E&W arbitration-counsel sign-off). Schemas: schema/dispute-case-file.schema.json. Binding checks: spec/verifier.md §V6.

Replaces antiquated ADR with a tiered, largely-agentic ladder that stays enforceable. Because an onTerms agreement is structured, hash-pinned data with a tamper-evident audit trail, a dispute is machine-adjudicable in a way prose contracts never were — but enforceability still demands real due process. This module is engineered to that line.

PART A — Standard Terms (adopt-verbatim)

1. Scope, floors, B2B gate

1.1 Governs resolution of any dispute arising out of or in connection with the Order or any incorporated Module. 1.2 Un-electable floors. Nothing here, and no Election, overrides: (a) Mandatory Law (incl. the DPA / data-protection floor, which sits above all Elections); (b) the jurisdiction of the competent courts, which the parties do not and cannot oust; (c) either party's right to seek urgent interim/injunctive relief, or relief for IP infringement or breach of confidence. The ladder sequences court access (Scott v Avery condition-precedent; Arbitration Act 1996 s.9 stay) — it never removes it. 1.3 B2B only. Available only where each party's B2B status is a checked fact (verifier.md §V2), not a self-asserted boolean. Pre-dispute binding arbitration against a consumer is presumptively unfair (Consumer Rights Act 2015) and the UCTA-only architecture collapses against a consumer — so a consumer dispute is out of scope and this module is void for it.

2. The ladder

Tier 0 self-executing remedies (no adjudicator) · Tier 1 agent-to-agent settlement · Tier 2 agentic mediation (non-binding) → AI-assisted arbitration (binding, human Confirmer), to the elected altitude · Tier 3 human/legacy ADR or courts. A dispute climbs one rung at a time; Tier 0 and the clause-1.2 carve-outs operate in parallel at all times. Every tier writes a structured object to a single append-only Dispute Record keyed to order_id + order_hash.

3. Tier 0 — self-executing remedies — [round-2 fix 2a/2b/2h]

3.1 A Tier-0 remedy is the contractual operation of an agreed formula on agreed data, with no adjudicator forming a judgement. It is not arbitration and produces no award (so it raises no Article V issue — UKJT 2019; Law Commission Smart Legal Contracts 2021).

3.2 Qualifying test. A matter is Tier-0 only if all of: (a) objective trigger (a measured quantity/dated event, no reasonableness/intent/causation/quantum); (b) agreed, tamper-evident measurement source (named in the Order, ideally Merkle-logged); (c) deterministic formula (single correct output); (d) bounded/capped; and (e) proportionality — each remedy carries, set at module-authoring time, a legitimate_interest_justification and a proportionality_attestation benchmarked to a genuine pre-estimate of loss. Determinism is not the Cavendish v Makdessi test; proportionality is. A formula can satisfy (a)–(d) and still be an unenforceable penalty — (e) is what guards that, and it is reviewed when the remedy is drafted, not asserted per-order.

3.3 Standing remedies (each tied to an objective SaaS obligation): T0-1 SLA credit (telemetry < elected uptime → capped credit); T0-2 late-payment interest (ledger; statutory rate + fixed sum); T0-3 renewal voided (notice outside window → no auto-renew that cycle); T0-4 exit-default on missing deletion attestation (a contractual consequence — the DPA floor governs the data-protection breach separately and is unaffected).

3.4 Provisional, reversible, review-open. A Tier-0 consequence is applied as a provisional hold / escrow or ledger reservation — NOT an irreversible set-off — until the challenge window (tier0.challenge_window_days, 7–30) closes. Either party may challenge on grounds that include the data was wrong, the formula was misapplied, OR the remedy is penal / unconscionable / contrary to mandatory law. Tier-0 finality is without prejudice to the court's penalty jurisdiction. (The earlier design auto-set-off before any review and narrowed challenge to "data/formula only" — that quietly ousted penalty review and is corrected here. The Verifier enforces this: verifier.md §V6.5.)

4. Tier 1 — agent-to-agent settlement

4.1 Each party's agent may negotiate a settlement within a pre-authorised mandate envelope (mandate.md). 4.2 Armagas in the dispute layer — [round-2 fix 1b]. A settlement mandate binds only if verifiably issued and signed by the principal — the Verifier asserts the principal signature resolves under the party's DID, never the agent's (the dispute schema's issued_by_principal boolean is not sufficient; verifier.md §V6.1). 4.3 Within envelope and below agent_autobind_ceiling, acceptance may auto-bind; above, the output is a proposed settlement that binds only on a human signature over its hash. Envelope bounds are sealed in the Record, not exchanged. 4.5 Output: a Settlement Object with binding_status: auto_bound | awaiting_human_ratification.

5. Tier 2 — mediation → AI-assisted arbitration

5A. Agentic mediation (non-binding, fully automatable). An AI mediator reviews the Record, frames issues, proposes non-binding resolutions. Binds only if both accept (then recorded as a Settlement Object). No Confirmer required (it binds no one).

5B. AI-assisted arbitration (binding — human Confirmer).

  • 5B.1 A real arbitration. Seated and ruled per the Order's governing_law_pack (for EW: seat England & Wales; Arbitration Act 1996 as amended by the Arbitration Act 2025; the onTerms Arbitration Rules). The signed Order is the arbitration agreement (s.6, 1996 Act), so an award is capable of recognition under the New York Convention.
  • 5B.2 AI assists; it does not decide (SVAMC Guideline 6, non-delegation). AI assembles the record, drafts issues, runs analysis, drafts the award. The binding decision is the human Confirmer's act.
  • 5B.3 Bias/hallucination controls (mandatory in the award — [round-2 fix 2d]). (a) ≥2 independent models from ≥2 providers, with material divergence flagged; (b) a separate adversarial verification pass that attacks the draft — every cited authority and record fact checked against the actual record, unverifiable citations struck (SVAMC verification duty); (c) a reasoned, structured award. These artefacts (models_used[], divergence_flags, adversarial_report_ref) are required in a binding award, not optional — the Verifier rejects a binding award lacking them (verifier.md §V6.4).
  • 5B.4 The Confirmer is schema-substantive — [round-2 fix 2c]. The award binds only when a qualified, independent human Confirmer (appointed under the Rules; s.23A disclosure duty, 2025 Act) signs it with a structured confirmer_attestation stating they: reviewed the divergence flags; reviewed/ran the adversarial report; verified every cited authority against the record; exercised independent judgement; and listed any amendments to the AI draft. A bare signature over a hash does not satisfy the schema — that is the difference between a real tribunal and a rubber stamp, and the difference between surviving and failing an Arbitration Act 1996 s.68 / NY Convention Art. V(1)(d) challenge.
  • 5B.5 Due process. Notice; equal opportunity to be heard (audi alteram partem, Art. V(1)(b)); impartial tribunal; right to comment on out-of-record AI material (SVAMC Guideline 7); reasoned award.

6. Tier 3 — human/legacy ADR or courts

Where unresolved or above the elected binding-AI altitude, escalate to legacy human ADR (human-only arbitration under the pack's seat+rules, or institutional mediation) or the competent courts of the pack's default_forum. The full Dispute Record travels — every Tier-0 statement, sealed mandate envelopes, settlement objects, mediation proposals, the AI draft, divergence flags, adversarial report, Confirmer reasoning — admissible as the agreed lower-tier record (subject to without-prejudice protection over settlement attempts).

PART B — The dispute Election (L3)

Decides how high up the binding-AI ladder the parties opt in. Default conservative: Tier 0 + Tier 1 + agentic mediation always on; binding AI-assisted arbitration is opt-in only, and the only binding altitude is ai_assisted_with_human_confirmer (there is no fully-autonomous binding tribunal). Seat + rules are inherited from the indivisible governing_law_pack, never set free-form here — and the Verifier enforces seat/law/rules coherence (verifier.md §V3), because a seat_rules_from_pack: true boolean has no referential integrity on its own (round-2 fix 2e). All on/off are explicit booleans; all caps are real numeric defaults (amount: 0 = "nothing auto-binds / Confirmer always required"), never sentinels.

The onTerms Arbitration Rules (onterms-arb-rules:1.0, an immutable content-addressed artefact pinned in the Order) codify: AI-assisted tribunals permitted; human Confirmer required above the value cap; model/provider diversity + adversarial verification; the clause-5B.5 due-process safeguards and SVAMC Guideline 6 non-delegation; a reasoned structured award.

PART C — Enforceability (honest)

You cannot oust the courts (and don't) — the ladder sequences, never removes, court access (§1.2). A binding award needs genuine due process — NY Convention Art. V refuses enforcement for want of notice/fair hearing (V(1)(b)), tribunal/procedure not per agreement (V(1)(d)), or public policy (V(2)(b)); these are narrow but procedural-fairness failures are exactly where awards fall. A pure-AI tribunal is high-risk (no human exercising the mandate; SVAMC Guideline 6) — so onTerms makes the binding rung a real arbitration: genuine arbitration agreement + seat/rules from the pack, an impartial human Confirmer who decides and signs, due process, mandatory bias controls, a reasoned award.

[round-2 fix 2f] Small-claims full-auto is expert determination, not an "award." For matters at/below small_claims_full_auto.value_cap (opt-in), a fully-automated outcome with no per-case human Confirmer is labelled binding expert determination / contractual ODR, not a NY-Convention arbitral award — because proportionality is an ODR concept, not an Art. V cure, and a foreign court may refuse a binding determination with no first-instance human adjudicator. It is always backstopped by a cheap human-appeal (human_appeal_backstop_days): either party may demand a human Confirmer (making it a clause-5B award) or go to Tier 3. (CRT/Money Claim Online are public tribunals, not authority that a private full-auto binding award enforces.)

PART D — Structured outcomes (self-enforcing where possible)

Every tier emits a typed object referencing the signed Order by order_id/canonical_ref/order_hash: Tier-0 Statement (provisional, set-off only after the challenge window), Settlement Object, Award (with the mandatory provenance block + confirmer_attestation). Monetary/ledger consequences set self_enforcing: true with an enforcement_action (provisional hold → set-off / ledger credit after finality); non-self-executing consequences carry to Tier 3 as the enforceable instrument (English-seated award enforced under s.66 of the 1996 Act, or abroad under the NY Convention). Every object is content-addressed, onSign-signed, RFC 3161-timestamped and Merkle-logged — the Order → Tier-0 → settlement/award chain is independently verifiable, which is what makes both auto-enforcement (the ledger can trust the object) and court/NY-Convention enforcement (integrity + agreed procedure provable) work.