LLM models
New Claude, GPT and Mistral versions are tested on internal fixtures (50+ FR briefs) then deployed after QA. No customer action required.
Updates included. No hidden surcharge.
LLM model improvements, prompt evolutions and data migration on schema changes are included in the per-seat price. The SLA is documented here, not buried in a premium tier.
01
Evergreen commitment
Every Scope subscription includes all engine evolutions (LLM models, prompts, extraction and scoring pipelines) at no extra cost. No AI feature is billed separately, no update is gated behind a higher tier.
Prompts and pipelines
Prompt engineering improvements (clarify, scoping, estimation, multi-piece extraction) are deployed continuously with no service interruption.
Schema migration
If Scope changes the data schema (new scoping axes, refactor of scoping objects), the migration of existing projects is free and orchestrated by the Scope team.
02
Versioning system
Each new model or prompt deployment follows the same ritual: fixtures, QA, canary, generalisation.
1 · Re-test on fixtures
The new model/prompt replays 50+ reference briefs (anonymised, sector-diverse). Metrics: clarify coverage, hallucination rate, person-day envelope alignment.
2 · Scope-team QA
Human review of significant deltas (>5% gap on any metric). Explicit validation or rollback documented in the audit log.
3 · 10% canary
Progressive deployment on 10% of traffic for 48-72h. Runtime error monitoring via GlitchTip and response-time monitoring via the observability stack.
4 · Generalisation
100% rollout. Public changelog published on the Changelog page. Email notification to org admins for significant changes.
03
Documented SLA
The SLA is not a marketing promise: it is measurable, contractualisable and published.
| Metric | Target | Scope | Contractual commitment |
|---|---|---|---|
| Application uptime | 99.5% | Application routes (/projets, /api/*) | Best-effort on Discovery/Solo/Team — contractual on Business/Enterprise |
| MTTR critical incident (S1) | < 4 hours | Incidents blocking usage during FR business hours | Contractual on Enterprise — best-effort below |
| Maintenance windows | EU weekends 02:00-06:00 | Major stack updates | Notified 7 days ahead — cancellable on Enterprise request |
| Incident procedure | Alerting <30 min, mitigation <4h, RCA <72h | All S1/S2 incidents | Published process — see Incident Response Process |
04
Roadmap of tested and planned models
Which LLM model versions have been tested on Scope fixtures, which are in production, which are coming.
| Model | Status | Tested on | In production since |
|---|---|---|---|
| Anthropic Claude Opus 4.6 | In production | 50 FR fixtures | May 2026 |
| Anthropic Claude Sonnet 4.6 | In production | 50 FR fixtures | May 2026 |
| Anthropic Claude Opus 4.7 | In test (canary) | 20 FR fixtures (extending to 50) | Target June 2026 |
| Anthropic Claude Sonnet 4.7 | Planned | — | Target Q3 2026 |
05
Free data migration
If we impose a schema change (new scoping axes, refactor of scoping objects, export restructuring), the migration of existing projects is entirely free and orchestrated by the Scope team.
Covered scope
Any schema change triggered by Scope (refactor, new mandatory axis, change in export format). Out of scope: voluntary client-side migrations (e.g. Stripe org change).
Process
30-day notification before deployment. Background migration, transparent to the user. No access interruption to historical documents. Audit log of each transformation retained.
Rollback
If the migration introduces a blocking regression, automatic rollback within 24 hours. Pre-migration snapshot retained 30 days.