Launch SLOs & Deferrals

This page commits one measurable SLO per customer-facing launch scenario (S1–S10) and publishes the explicit list of promises that are intentionally deferred beyond launch. Every committed number is anchored to an SLI that already exists — in scale/slos.md (SLI-1..9) or the component-level catalog at trust.mnemom.ai/slos. Unmeasured promises are flagged, not invented.

This page defines per-scenario SLOs only. Tier-based contractual uptime targets (SLA), RTO, and RPO are on SLA & Incident Response. The error-budget deploy-gating policy is tracked separately at MNE-376 and is on the deferral list below.

Per-scenario SLOs

Sign-off pending (@wassimwehbi-mnemom): all committed numbers — especially S1 (99.9% / p99 < 100 ms), S9 (≤ 30 min comms target reuse), S10 (5 min erasure p99 reuse), and S2 publish-as-target vs hold. Numbers below reflect the draft; merge is gated on sign-off.

#	Scenario (customer promise)	Committed SLO	Measurement basis	Measured today?
S1	I get to a working hello-world in minutes	Quickstart path availability ≥ 99.9% monthly (gateway non-5xx); Mnemom-added latency overhead p99 < 100 ms on the first governed call	SLI-1 + SLI-2 (`scale/slos.md`); burn-rate alerts live	YES — availability and latency overhead measured. The < 10-min human time-to-value half has no prober yet; needs `mnemom_scenario_up` (MNE-357 item 3)
S2	My agent onboards itself	Agent-discovery surface availability (llms.txt / llms-full.txt / agents.txt / markdown content-negotiation) ≥ 99.9% monthly	Target only — no SLI exists yet. Measurement window opens when the `mnemom_scenario_up` scenario prober ships (MNE-357 item 3)	NO — target published; measurement deferred
S3	Protection just works	Safe House screens 100% of governed calls for agents with a published protection card; Chat-Always-Completes ≥ 99.99% (an intervened request still returns 2xx)	“Off-mode fidelity” + “Chat Always Completes” SLOs at trust.mnemom.ai/slos; gateway availability inherits SLI-1	PARTIAL — prod enforce flip not yet thrown (`RECIPE_MODE=shadow`, MNE-231): enforcement is screen/observe + verdict headers, not blocking. 1-10-60 detect/contain numbers are deferred (see deferral list item 4)
S4	Alignment just works	99% of integrity checkpoints processed within 5 minutes (rolling 7-day)	SLI-3 (Observer Lag, `scale/slos.md`); matches trust-page “Trace freshness p95 ≤ 5 min”	YES
S5	Sign in, create an org/agent, get a key — safely	Control-plane `/v1` availability ≥ 99.9% (30-day); p95 ≤ 250 ms (excl. LLM-proxy endpoints)	“API availability” + “API P95 latency” at trust.mnemom.ai/slos	PARTIAL — no dedicated signup/auth alert wired yet (MNE-357 item 3)
S6	Show my compliance team what my agents did	Every governed decision recallable as an AP-trace within 5 minutes p95	SLI-3 / trust-page Observer SLO	YES for recall. Audit export bundle not independently exercised — excluded from the committed number
S7	I can trust the number	Trust-Rating lookups p99 < 200 ms, availability ≥ 99.9%	SLI-4 (`scale/slos.md`)	PARTIAL — SLI-4 burn-rate alert documented but no checked-in alerting rule implements it. On-chain anchoring is on the deferral list (item 2)
S8	I can buy it	≥ 99.5% of billing webhook events delivered and processed within 10 minutes, driving automatic entitlement updates	SLI-5 (`scale/slos.md`) + trust-page 10-minute webhook SLO; billing webhook-health table (migration 023)	YES for webhook delivery. Checkout→entitlement E2E not exercised (trials are comped) — excluded from the committed number
S9	When it breaks, we both find out	Public status-page first update ≤ 30 minutes after SEV-1/SEV-2 declaration	Already committed on SLA & Incident Response; status.mnemom.ai is live	YES — internal-paging half is MNE-357 O3, not a customer SLO
S10	I can run it where my data must live	GDPR Article-17 erasure cascade completes within 5 minutes p99 (tombstone < 1 s)	Already committed on GDPR Data Subject Rights; internal stuck→1 h P1 alert in `scale/slos.md`	YES for erasure. Managed EU residency has no SLO — see deferral list item 3. Self-host (Enterprise) is the documented boundary path

The honest deferral list — explicitly out of launch scope

The following are not committed at launch. Each entry names the supported surface and the condition for re-evaluation.

AEGIS cross-tenant adaptive layer (Managed-Rule recipe tier). The gateway runs RECIPE_MODE=shadow; cross-tenant recipes observe, never block. The seven AEGIS SLOs on trust.mnemom.ai/slos open their measurement window at GA. Per-agent protection cards are the supported launch enforcement surface.
On-chain anchoring (Base L2 reputation registry + Merkle anchor). No SLO. The pipeline is fail-open with no alerting, runbook, reorg handling, or wallet monitoring. The off-chain Trust Rating is the supported launch surface; on-chain reads are best-effort verification, not a commitment. See On-chain verification for current capability scope.
Managed EU data residency. The managed cloud is single-region US (Supabase). An EU data boundary requires self-host (Enterprise). A named managed EU region is deferred with no committed date.
Live blocking on production traffic (the S3 enforce flip). Until MNE-231 lands, protection on production traffic is screening + verdict headers + AIP detection evidence — not synchronous blocking. No 1-10-60 detect/contain numbers are published for the blocking path.
Self-serve checkout end-to-end. Trials are comped; the checkout→webhook→entitlement path carries no E2E SLO. The webhook-delivery SLO in S8 is the committed slice.
/.well-known agent discovery. agents.txt is served at the web root only; /.well-known endpoints are deferred.
Localized (EFIGS) documentation. The marketing site ships EFIGS; docs are English-only at launch.
Per-scenario cost/COGS SLOs. Objective O9 is deferred to T1.
Risk-engine SLO calibration. Risk traffic has been dry since 2026-03-09; risk-proof SLIs inherit verdict-path targets as a placeholder (scale/slos.md OQ4).
Error-budget deploy-gating policy. Burn-rate alerts exist; the deploy-freeze policy is tracked at MNE-376 and is not yet in effect.

​Per-scenario SLOs

​The honest deferral list — explicitly out of launch scope

​See also

Per-scenario SLOs

The honest deferral list — explicitly out of launch scope

See also