Liability Framework

Decisions made on 2026-05-20. Covers the advisory/authoritative distinction, ToS language, API response design, marketing guardrails, and jurisdiction.

1. Core position: advisory tool, not decision system

Goable is legally and architecturally a decision support tool. The operator sees the score and makes the final go/no-go decision. This separation — the operator clicking "cancel" or "go" — is the product's safe harbor under EU Product Liability Directive II (PLD II, 2024).

PLD II extends product liability to software and AI systems for the first time. The safe harbor criterion: "the damage was caused by an autonomous, informed decision of a human operator, not by automatic output of the system." Goable satisfies this by design: the score informs, the operator decides.

This position must be reflected in three places consistently. If even one is missing, the risk shifts toward Goable.

2. Terms of Service — the primary legal protection

2.1 Core liability clause

"Goable scores and verdicts are informational outputs based on weather forecast data and activity-specific models. They do not constitute operational directives, safety certifications, or guarantees of any kind. The operator is solely responsible for all go/no-go decisions regarding any activity. Goable assumes no liability for outcomes resulting from decisions made in reliance on its scores, including but not limited to personal injury, property damage, or economic loss."

The term "solely responsible" is the legally significant phrase — it establishes unambiguously that the decision chain does not include Goable.

2.2 Forecast uncertainty clause

"Weather forecast data is inherently uncertain. Score accuracy degrades with forecast lead time. Goable does not guarantee that forecasted conditions will match observed conditions at any specific location or time. The confidence field in every API response quantifies this uncertainty; operators should treat low-confidence scores with additional caution."

2.3 Governing law

Italian law (currently). Chosen for simplicity and familiarity. Review when Goable expands beyond EU or when Scale-plan enterprise clients with US/UK legal requirements appear. At that point, Irish law (Dublin) is the standard EU tech company alternative — common law, EU-compliant, GDPR-covered, used by Stripe, Intercom, HubSpot.

3. API response — the technical protection

3.1 Existing fields that already help

• confidence: quantifies forecast uncertainty. Low confidence = higher uncertainty. Already implemented.
• verdict: a label (favorable, marginal, unsafe…) not a command. Never "go" or "cancel".
• alerts: array of specific warnings. Surfaces the model's own uncertainty signals.

3.2 Fields to add

Add advisory: true as a static boolean in every score response. Serves two purposes:

1. Makes the advisory nature machine-readable for integrators
2. Creates an audit trail: every API response on record contains an explicit advisory marker

In the OpenAPI spec description for the score object, add:

"All scores are advisory. The integrating application must not present this score as a guaranteed condition assessment or as a substitute for professional judgment."

This contractually constrains the B2B client's UI framing — they cannot strip the advisory context when surfacing the score to their operators.

3.3 Phase 1 addition: safety_caveat on low confidence

When confidence < 0.6, include a safety_caveat string in the response:

"Forecast confidence is low for this window. Conditions may differ significantly from predicted. Exercise additional caution."

This was listed as a Phase 4 feature in the mission doc but should move to Phase 1 — it is a liability protection measure, not a product enhancement.

4. Marketing language guardrails

This is where liability risk accumulates invisibly over time — in pitch decks, tweets, landing page headlines, Product Hunt descriptions.

4.1 Never say

4.2 Always say

Rule of thumb: if the subject of the sentence is the operator ("you decide", "your team acts"), it's safe. If the subject is Goable or "the AI" ("it decides", "AI tells you"), rewrite it.

5. The operator responsibility chain

Weather provider (Stormglass, Open-Meteo)
    ↓ returns raw forecast data
    ↓ their own liability: forecast accuracy, uptime
Goable
    ↓ scores and interprets forecast data
    ↓ liability: model correctness, system uptime, accurate confidence signaling
    ↓ NOT liable for: forecast inaccuracy, operator decisions, student safety
Booking platform (Sealect, others)
    ↓ surfaces the score in their UI
    ↓ liability: how they present the score (must not strip advisory framing)
Operator (kite instructor, surf school)
    ↓ makes the go/no-go decision
    ↓ solely responsible for: student safety, on-site conditions assessment,
       operational judgment that cannot be delegated to software

This chain must be visible in the Goable ToS and in the DPA with booking platforms (which must not allow platforms to present Goable scores as authoritative).

6. EU Product Liability Directive II (PLD II) — key points

Effective 2024. Extends product liability to software and AI. What this means for Goable:

Exposed if:

• Goable makes autonomous decisions that cause harm without human review
• Marketing materials have warranted accuracy that was not delivered
• The system had a defect (bug, wrong formula) that caused a predictably dangerous output

Protected if:

• A trained professional reviews the output before acting (operator reviews score → decides) ✓
• The uncertainty is inherent and disclosed (confidence field, forecast caveats) ✓
• The ToS clearly allocates responsibility to the operator ✓
• No express warranty of accuracy in marketing ✓

Action required before launch: have an Italian lawyer review the final ToS against PLD II and Consumer Code (Codice del Consumo). Estimated cost: €400-600 one-time. Not optional.

7. What is explicitly not Goable's liability

• Forecast inaccuracy: weather forecasts are probabilistic by nature. Goable passes through provider data; it does not produce forecasts.
• Microclimate events: sudden local conditions (thermal gusts, localized storms) that global models cannot resolve at <10km. Disclosed in the confidence model.
• Equipment failure: score says favorable conditions, equipment fails. Entirely outside Goable's scope.
• Operator negligence: operator ignores a score of 15 ("poor") and goes out anyway. Goable provided the warning.
• Student or customer decisions: an adult student who overrides instructor judgment.

8. Next actions