InferLane Exchange — Operator Agreement

Drafted + reviewed by Claude. Captures the substantive
commitments we require of operators, drawn from independent-
contractor jurisprudence (US AB-5, UK IR35, EU Platform Work
Directive, JP Worker Classification Act), AI-IP exposure
patterns (Microsoft Copilot / Codeium litigation), and
AarDappvark Toolkit's prior legal pack. Re-reviewed by Claude on
revisions; specific clauses flagged in the drafting note at the
bottom for re-review when the listed triggers occur.

1. Parties

This Operator Agreement ("Agreement") is between InferLane Pty Ltd ("InferLane", "we", "us") operating the InferLane Exchange marketplace and the individual or entity registered as an operator and submitting work to tasks posted through the marketplace ("Operator", "you", "your").

2. Independent contractor — load-bearing

You are an independent contractor providing software-engineering services to buyers via the InferLane Exchange marketplace. Nothing in this Agreement creates an employment, partnership, joint venture, or agency relationship between you and InferLane or between you and any buyer. The following ten operating-practice constraints are non-negotiable because they are how the marketplace defends operator classification against misclassification challenges (Uber/AB-5 pattern):

You set your own hours. Marketplace does not require minimum

hours, online presence, or response-time commitments outside the per-task SLA you accept by applying.

You choose which tasks to apply for. Marketplace does not

assign work to you; you opt in per task.

You bring your own tools. Marketplace does not provide laptops,

IDEs, AI subscriptions, or any other instruments of work. Your LLM usage runs on your accounts at your expense.

You may work for anyone. No exclusivity. You may operate on

competing marketplaces or take direct buyer engagements.

You are paid per verified deliverable, not per hour. Settlement

is triggered by attestation aggregate + buyer-decision window, not by elapsed time.

No benefits. Marketplace does not provide health insurance,

leave, retirement contributions, or any other employment benefits.

You are responsible for your own taxes. Marketplace issues

1099-NEC / equivalent jurisdictional forms when thresholds are met but does not withhold; you remit your own income, self-employment, and sales taxes as applicable.

No subjective performance reviews. Marketplace measures only

objective outcomes (completion rate, attestation pass rate, dispute outcomes); we do not write performance reviews or skill assessments that constitute employment management.

You may use subcontractors. You may delegate execution to your

own employees, contractors, or AI agents acting under your authorization, provided the IP warranties in §5 still hold.

No platform-issued tools or credentials. Marketplace does not

issue you GitHub accounts, IDE licenses, or any tooling whose revocation would prevent you from working elsewhere.

3. Acceptance + collateral

By applying to a task, you represent that you have read its description, rubric, and mechanical gates and intend in good faith to fulfill them. The marketplace locks operator collateral (default: $10 USD) at apply time and releases it at settlement (approve / auto-approve) or forfeits it on SLA miss without submission. Collateral forfeiture is your contractual stake in delivery, not a fee.

4. SLA + revisions

You commit to submitting work within the operator SLA hours displayed on the task at apply time. Missing the SLA without an extension request grants the marketplace the right to mark the task expired, forfeit your collateral, and refund the buyer in full.

If your submission's mechanical gates fail or the arbiter aggregate returns needs_revision, the task transitions to returned_to_operator and you may submit a revised attempt within the same SLA window. The marketplace tracks attemptNumber and limits maximum attempts per task.maxRotations.

5. IP warranty + indemnification

By submitting work to a task, you warrant that:

(a) You own or hold sufficient license to all code, configuration,

documentation, and other artifacts in your submission;

(b) Your submission does not infringe any third party's copyright,

patent, trademark, trade secret, or other intellectual-property right;

(including but not limited to Anthropic Claude, OpenAI GPT, GitHub Copilot, or any other code-generating model), you have read the applicable provider terms and the output's use in the buyer's intended context does not violate those terms or expose the buyer to training-data IP claims;

(d) Your submission complies with the target repository's license

and any contributor-license-agreement (CLA) terms applicable to contributions to that repository.

You indemnify and hold harmless the marketplace, the buyer, and any downstream recipient against any claim, demand, or proceeding arising out of a breach of (a)–(d) above. This indemnity survives termination of your account and applies retroactively to all submissions you have made through the marketplace.

This clause is load-bearing: it follows directly from the AI-IP litigation pattern observed against GitHub Copilot, Codeium, and similar tools. The marketplace's defense against being a co-defendant is operator IP indemnification.

6. Attestation cooperation

You authorize the marketplace to provide your submission (PR URL or diff blob, execution attestation, mechanical-gate results, operator self-assessment) to assigned arbiters for attestation. Arbiters render independent professional opinions on whether your submission satisfies the buyer's rubric; the marketplace aggregates via Schelling-point voting. You agree not to attempt to communicate with, identify, or influence arbiters during the attestation window.

7. Disputes

If a buyer files a dispute on your submission, you have 72 hours from notice to file an operator_statement evidence record via the disputes endpoint. Failure to file does not constitute admission but may result in adverse admin resolution on the available record. Disputes are resolved by InferLane staff in Phase 1; future phases may delegate to expanded arbiter pools.

8. Account suspension + termination

Marketplace may suspend or terminate your operator account for:

Repeated SLA misses (3+ in any 30-day window);
Sustained dispute-loss rate above 20%;
Misrepresentation in your registered agent config, skill packs, or

capability declarations;

Breach of §5 IP warranties (single material breach is grounds);
Breach of §2 operating-practice constraints (single instance is

grounds because the constraints exist to defend the entire model);

Any conduct that, in our reasonable assessment, threatens the

marketplace's classification or compliance posture.

You may close your account at any time via the dashboard. Pending applications are auto-declined; in-flight tasks (accepted or returned_to_operator) must complete or be marked SLA-expired before account closure resolves.

9. Fees

Marketplace platform fee is 10% of the bounty, withheld at settlement from your payout. Stripe processing fees on buyer escrow funding are absorbed at funding (paid by the platform from the same bounty's remaining margin in Phase 1; Phase 1.5 may pass them through to the buyer). The current fee structure is locked in commercial/DECISIONS.md.

10. Governing law + dispute resolution

This Agreement is governed by the laws of the Australian state where InferLane is incorporated. Disputes between you and InferLane that cannot be resolved through good-faith discussion are resolved by binding individual arbitration under the rules of the Australian Centre for International Commercial Arbitration (ACICA), seated in that same state. You waive class-action and mass-arbitration participation to the extent permitted by your jurisdiction; if a court holds the class waiver unenforceable, the arbitration clause is severed (not the waiver).

11. Changes to this Agreement

Marketplace may update this Agreement on 30 days' notice via in-app notification to your registered email. Continued use after the effective date constitutes acceptance. If you do not accept the updated terms, your remedy is to close your account before the effective date and complete any in-flight tasks under the then-current terms.

12. Acknowledgment

By registering as an operator, you acknowledge that you have read, understood, and accepted this Agreement and the related InferLane Exchange Terms of Service.

13. Compute Provider extensions (Phase 1.5)

Sections 1-12 above govern your work as an operator on the Tasks service mode of InferLane Exchange (bespoke discrete software work delivered as a PR or equivalent artifact). This §13 governs your work, if any, as a Compute Provider under the Phase 1.5 Compute service mode (per-call verified API access). Sections 1-12 continue to apply to your Tasks-mode work; §13 applies only to Compute-mode invocations you serve.

By opting in as a Compute Provider via compute.register_provider, you accept this §13 in addition to §§1-12. You may opt out at any time by withdrawing your provider registration; pending invocations remain bound under the §13 terms in effect when accepted.

13.1 Compute Provider role

A Compute Provider serves per-call invocations through the marketplace. Specifically you:

(json-extraction, embedding, classification, code-generation, structured-markdown, extraction — per commercial/COMPUTE_RUBRIC_LIBRARY.md);

Declare per-call pricing per task type;
Post the bond required for your provider status tier (per

commercial/COMPUTE_TRUST_POSTURE.md §2);

Serve invocations routed to your endpoints within the per-call

latency SLO you declared.

The 10 independent-contractor operating-practice constraints in §2 above continue to apply to your Compute Provider work, with these clarifications:

§2.1 (set your own hours) — you set your own provider availability

windows; you may take your provider offline at any time (existing invocations complete normally; new routing is suspended).

§2.2 (choose tasks to apply for) — you choose which task types to

serve via your registered endpoints; you may add or remove task types at any time.

§2.3 (bring your own tools) — you bring your own compute

infrastructure, model access (e.g. your own Anthropic API key, your own GPU/TPU rental, your own TEE infrastructure for attested mode), and operational tooling. The marketplace provides no compute infrastructure to you.

§2.5 (paid per verified deliverable) — payment is per successful

invocation, not per hour of provider availability. Failed invocations (per Compute audit pipeline or per provider error) do not generate payment.

§2.10 (no platform-issued tools) — your provider endpoint is

yours; the marketplace routes traffic to you but does not provide you with API keys, compute, or any credential whose revocation would prevent you from operating outside the marketplace.

13.2 Bond + slashing

You agree to maintain at all times the bond level required for your provider status:

| Status | Required bond | When required | |---|---|---| | Probationary | $100 USD | At registration; held throughout probationary period | | Active (any reputation) | $50 USD | Required to remain active | | Suspended | bond held pending review | During investigation |

Skill-pack reputation transfer: if you have ≥ 0.85 Tasks-side reputation with ≥ 10 settled tasks, your bond requirements are halved (probationary $50, active $25) per COMPUTE_TRUST_POSTURE.md §7.6.

You agree to the slashing schedule documented in COMPUTE_TRUST_POSTURE.md §3:

10% per L3-confirmed audit failure;
100% if you accumulate 5 L3 failures within any 30-day window

(followed by status → suspended);

25% per buyer-dispute resolved against you (post-receipt

failure path);

50% per failed-attestation event in attested mode (most

serious failure type, undermines verification primitive).

Slashing follows L3 confirmation only (AI L1/L2 do not slash bond); slashed amounts distribute 50% to buyer refund + 30% to arbiter compensation + 20% to platform reserve fund.

You agree to maintain bond at the required level; if slashing reduces your bond below the minimum for your status, you receive a top-up request and 7-day grace period to restore the bond. After the grace period, your status moves to suspended until bond is restored. After 30 days suspended without top-up, your provider registration is deregistered and remaining bond is forfeit to the platform reserve fund per COMPUTE_TRUST_POSTURE.md §3.4.

13.3 Probationary mode

New Compute Provider registrations enter probationary mode for first 1000 invocations (or 500 with skill-pack transfer). Constraints:

You may serve only replicated-mode invocations during probation,

auto-paired with an anchor provider for output comparison;

Daily volume cap: $100 USD worth of invocations;
Audit rate: 1-in-100 (in addition to always-on consensus

comparison with the anchor).

Probation graduation requires 1000 (or 500) clean replicated invocations. You acknowledge that disagreement with the paired anchor provider during probation counts against your reputation even when the disagreement is borderline; this is by design to ensure new providers establish a high-confidence baseline.

If you accumulate more than 5 disagreements during probation (0.5% disagreement rate on 1000 invocations), your status moves to review pending Exchange investigation. Outcomes are documented in COMPUTE_TRUST_POSTURE.md §6.3.

13.4 Attestation cooperation (Compute mode)

You authorize the marketplace to:

Submit any of your invocation outputs to the AI audit pipeline

(L1 Claude Haiku structural verification, L2 Claude Sonnet semantic verification, L3 human arbiter Schelling vote);

Compare your outputs against paired anchor provider outputs

during probationary mode and in replicated-mode invocations;

Publish your audit history (pass/fail counts, audit tier,

outcomes) publicly via compute.audit_history and at /exchange/compute/providers/[your-id];

Adjust your reputation score per the curve documented in

COMPUTE_TRUST_POSTURE.md §5.

You agree not to attempt to identify L3 arbiters, communicate with arbiters during audit windows, or coordinate with other providers to manipulate consensus in replicated mode. Violation of any of these is grounds for immediate suspension + 100% bond forfeit.

13.5 Data handling obligations

As a Compute Provider you agree to:

Process buyer inputs only for the duration of the specific

invocation you were routed to serve;

Not retain, copy, archive, log, or otherwise persist buyer

inputs or outputs beyond what is required to serve the invocation (transient processing buffers excepted);

Not use buyer inputs or outputs to train models, fine-tune

systems, build derivative datasets, or for any purpose beyond serving the specific invocation;

Provide a data-handling attestation if requested by the

marketplace (e.g. SOC 2 Type II report, ISO 27001 certificate, signed self-attestation) within 30 days of request;

Not log invocation inputs to your provider-side observability

systems beyond opaque metadata (invocation ID, latency, cost) required for billing reconciliation.

Violation of any of these is grounds for immediate suspension and provides the marketplace cause for 100% bond forfeit regardless of audit outcomes.

13.6 Attested mode (TEE) obligations

If you opt into attested mode by declaring attestationCapabilities in your provider registration, you additionally agree to:

Run inference workloads serving attested-mode invocations

exclusively inside a Trusted Execution Environment (TEE) of the declared type (e.g. Azure Confidential VM with SEV-SNP, Intel TDX, AMD SEV-SNP per commercial/DECISIONS.md §2);

Generate cryptographic attestation reports for each attested-mode

invocation that the marketplace and buyer can independently verify via the appropriate attestation service (Azure Attestation Service or RATS-compliant verifier);

Not falsify, replay, or otherwise manipulate attestation reports.

Falsified attestation is the most serious failure mode under this Agreement and incurs 50% bond slash (per §13.2) plus immediate suspension pending investigation;

Maintain TEE infrastructure patched and in compliance with the

TEE vendor's security advisories.

13.7 Pricing + fees

Your per-call pricing is set by you at registration via pricingPerCallUsdCents per task type per execution mode (single, replicated, attested). You may update pricing at any time; new pricing takes effect on the next invocation routed to you (in-flight invocations complete under the pricing in effect at routing time).

Marketplace platform fee on your Compute revenue is documented in commercial/COMPUTE_PRICING.md §3. You receive your declared per-call price minus your bond cost minus (for anchor pairing during probation) the consensus orchestration cost. Stripe payout processing is per the Stripe Connect terms you accepted on operator onboarding.

For invocations failing audit (L1/L2/L3 confirmed failure or buyer-dispute upheld), you do not receive the per-call price and your bond is slashed per §13.2.

13.8 Anchor providers — Exchange-operated

You acknowledge that the marketplace operates anchor providers wrapping upstream third-party APIs (Anthropic Claude, Voyage AI, and equivalents). Anchor providers:

Have synthetic reputation backed by Exchange's own reputation;
Receive routing priority for any invocation where no

reputation-graduated third-party provider better matches the buyer's filters (price, attestation capability);

Are used as the paired comparison for new third-party providers

during probationary mode.

You agree that this routing arrangement does not constitute unfair preference; anchor providers exist to provide day-1 production capacity and a consensus anchor for replicated mode. Your reputation-based routing weight grows with audit pass rate per COMPUTE_TRUST_POSTURE.md §5, allowing you to compete with anchors over time on per-call basis.

13.9 Acceptable Use — Compute extensions

In addition to §3 (Acceptable use) above and ACCEPTABLE_USE_POLICY.md generally, as a Compute Provider you may not:

Serve outputs you know to be incorrect;
Coordinate with other Compute Providers to manipulate consensus

in replicated mode;

Misrepresent your attestation capabilities, supported execution

modes, or hardware/software stack;

Serve invocations you were not routed to via the marketplace

routing layer (e.g. by sniffing the routing API and attempting to serve invocations before they're assigned);

Retain buyer inputs or outputs beyond §13.5 limits;
Operate multiple unlinked Compute Provider accounts to evade

reputation scoring (Sybil behavior; same prohibition as Tasks per ACCEPTABLE_USE_POLICY.md §4).

Violation of any of these is grounds for immediate suspension + 100% bond forfeit, in addition to any other consequences under this Agreement, the TOS, or applicable law.

13.10 Account suspension + termination (Compute extensions)

In addition to the Tasks-mode suspension grounds in §8 above, the marketplace may suspend or terminate your Compute Provider registration for:

Accumulated L3 audit failures triggering automated suspension

per §13.2 (5 fails in 30d);

Attested-mode attestation failures (immediate suspension pending

investigation);

Sustained dispute-loss rate above 15% over 30 days (lower

threshold than Tasks because per-call audit asymmetrically surfaces quality issues);

Bond not topped up after slashing within 7-day grace period;
Breach of §13.5 data-handling obligations (single material breach

is grounds);

Misrepresentation in your attestationCapabilities,

supportedTaskTypes, or pricingPerCallUsdCents declarations;

Sustained latency above declared SLO (more than 10% of

invocations exceeding declared SLO over 30 days);

Any conduct that, in our reasonable assessment, threatens the

marketplace's verification posture or the Compute service mode's credibility.

You may withdraw your Compute Provider registration at any time via compute.deregister_provider. Pending invocations complete normally; bond is returned 30 days after deregistration (chargeback window) absent any open disputes or audit investigations involving your provider.

13.13 Settlement, Stripe Connect Express, and tax (Phase 1.5.1)

Phase 1.5.1 enables third-party Compute Providers (independent contractors using Stripe Connect Express for payouts) in addition to anchor providers. This §13.13 governs settlement mechanics and the tax + chargeback responsibilities specific to that path.

(a) Custody posture. All provider funds are held by Stripe in the provider's Stripe Connect Express account. The marketplace never custodies provider funds. The "settlement" row in the marketplace database is an accounting projection of the Stripe Transfer record, not a held balance. See LEGAL-PERIMETER.md §4.2 for the operative legal posture; that document governs in any conflict.

(b) Platform service fee, not commission. The 4% fee retained by the marketplace on each Compute invocation is the platform service fee for connection, dispatch, audit, settlement processing, and chargeback handling. It is not a sales commission and does not establish a partnership, joint venture, or employer/employee relationship.

(c) Per-invocation transfer + minimum invocation cost. Stripe Transfers fire per-invocation (not daily-batched) for Phase 1.5.1. The minimum invocation cost is $0.50 — below this threshold, Stripe's per-transfer fee (~$0.25 + 0.25%) exceeds the 4% platform service fee and the marketplace operates at a loss. The marketplace reserves the right to decline invocations below this threshold or to batch them (returning a delayed-settlement notice to both buyer and provider).

(d) 30-day MAX hold envelope. No settlement persists in any non-terminal state more than 30 days from initial RESERVED. At day 30, the settlement is force-clawed-back and the buyer is refunded regardless of audit state. This is a hard backstop required by LEGAL-PERIMETER.md §4.3 to keep the platform outside money transmission perimeters. You waive any claim against the marketplace for amounts force-clawed-back under this clause.

(e) Audit-failure clawback. If a Compute invocation is verdicted as L1, L2, or L3 audit failure per COMPUTE_TRUST_POSTURE.md objective criteria, the corresponding settlement is clawed back via stripe.transfers.createReversal if already paid out, or voided if not yet paid out. The buyer is refunded the full invocation cost. The marketplace reserves the right to apply a NEGATIVE accrual against the provider's NEXT settlement batch when clawback occurs after Stripe transfer settlement.

(f) Chargeback clawback from future earnings. Buyer-initiated chargebacks on Compute charges trigger: - Immediate freeze of the provider's next settlement batch on charge.dispute.created; - stripe.transfers.createReversal + NEGATIVE accrual on charge.dispute.lost; - If the provider's pending settlement balance is insufficient to cover the clawback, the deficit applies against ALL future settlements until cleared. You authorize this clawback from future earnings as a condition of Phase 1.5.1 participation. - If the deficit remains uncleared 60 days after the charge.dispute.lost event, the marketplace may pursue recovery via any available legal means, including reporting the debt to a collection agency.

(g) Tax handling — Stripe issues 1099-NEC. Stripe Connect Express issues 1099-NEC forms directly to US-tax-resident providers earning more than $600 in a calendar year. The marketplace relies on Stripe's tax-form issuance and is not responsible for issuing duplicate forms unless Stripe fails to issue for a provider the marketplace knows is owed one (in which case the marketplace will manually issue a 1099-NEC within 30 days of discovery). You are responsible for your own tax filings and must keep your Stripe Connect Express tax-info section current.

(h) US-tax-resident warranty (Phase 1.5.1 ONLY). By onboarding as a Compute Provider via the Phase 1.5.1 application, you warrant that you are a US tax resident (US citizen, US lawful permanent resident, or US-domiciled entity). Non-US providers are not eligible for Phase 1.5.1 onboarding. International provider expansion is deferred to Phase 1.5.2+; the marketplace will update this clause and notify all providers at least 30 days before any change.

(i) AI-training-data IP warranty (extension of §5). §5(c)'s IP indemnification applies to all model outputs delivered by you to the marketplace, including outputs produced by AI models you operate as part of your Compute service. You warrant that you have the right to use any model weights, fine-tuned models, or training data your service depends on, and that none of your outputs contain or substantially reproduce copyrighted material you do not have the right to use. The marketplace may suspend your provider registration immediately on receipt of a credible IP infringement claim against any of your outputs, pending investigation.

(j) Probation and acceptance. Newly admitted Phase 1.5.1 providers enter active_probationary status with reduced dispatch volume and increased audit cadence for the first 30 days or first 50 invocations (whichever comes first). The marketplace reserves the right to extend probation or refuse to graduate any provider that fails calibration tests or accumulates dispute losses during probation.

13.14 Vocabulary lock (Phase 1.5.1)

The marketplace uses specific vocabulary in user-visible surfaces to preserve the independent-contractor relationship per commercial/legal/PROVIDER_VOCABULARY.md. The marketplace will admit or decline your application (not "approve" / "reject"); the application eligibility check is conducted by the founder (not a "review board"); your relationship to the marketplace is one of eligibility + service delivery (not "employment"). These terms are load-bearing for the contractor-classification posture in §2 above and may not be varied by support or marketing materials.

13.11 Changes to this section

Same as §11 — 30 days' notice via in-app + email. Changes to the bond schedule (§13.2) or audit cadence (§13.4) follow the trust-posture review cadence in COMPUTE_TRUST_POSTURE.md §15; material changes outside review triggers require a DECISIONS-level reversal.

13.12 Acknowledgment

By opting in as a Compute Provider via compute.register_provider, you acknowledge that you have read, understood, and accepted §13 in addition to §§1-12 of this Agreement, the related InferLane Exchange Terms of Service (including §12 Compute mode amendments), the Acceptable Use Policy (§10 Compute extensions), the Compute Pricing model (commercial/COMPUTE_PRICING.md), and the Compute Trust Posture (commercial/COMPUTE_TRUST_POSTURE.md).

Drafting note — re-review triggers: Re-review the following
clauses when the listed triggers occur:

- §5(c) — AI-IP indemnification language: re-review on each
material change to Anthropic / OpenAI / GitHub provider terms
(set a quarterly Claude review).
- §8 suspension grounds — re-review if any single suspension
produces operator pushback claiming disproportion or denial of
due process.
- §10 force-arbitration + class waiver: re-review on any
regulatory development that limits class-waiver enforceability
in AU (ACL §23 unfair-contract-terms regime), US (CA
Iskanian-style decisions), or EU (Platform Work Directive
amendments).
- §2 ten-rule list — re-review whenever supported jurisdictions
list (currently US/AU/GB/CA/NZ/IE/JP/SG) expands; each new
jurisdiction has its own contractor-classification rules.
- §13.2 bond + slashing — re-review on any material change to the
COMPUTE_TRUST_POSTURE.md slashing schedule or distribution
ratios.
- §13.4 attestation cooperation — re-review on each material
change to the AI audit pipeline (L1/L2 prompt revisions),
tracked in COMPUTE_TRUST_POSTURE.md §7.4.
- §13.5 data handling — re-review on EU AI Act Article 6 / Article
26 obligations for Compute Providers, GDPR developments affecting
processor obligations, or addition of any EU jurisdiction.
- §13.6 attested mode — re-review on each TEE vendor security
advisory affecting declared TEE types, AWS Nitro Enclaves /
Azure CVM / Intel TDX / AMD SEV-SNP security updates.
- §13.9 acceptable use Compute extensions — re-review on each
material change to upstream provider acceptable use policies
(Anthropic AUP, OpenAI AUP, etc.).