Service Level Agreement

Last updated Jun 14, 2026

This page is the authoritative SLA for the Matter API. It is the same document the legal contract references; this page is plain English; the legal document is at https://mattermode.com/legal/sla (kept in sync by the engineering lead).

Availability commitment

For every GA endpoint: ≥ 99.95 % availability per calendar month, measured by external probes (US + EU + APAC) at 1-minute intervals.

For Beta endpoints: ≥ 99.5 % availability. Beta is opt-in; the lower SLA is explicit on the endpoint's docs page and in the Matter-Maturity: beta response header.

For Preview endpoints: no SLA. Preview is design-partner-only.

Credits when the SLA is missed

If monthly availability falls below 99.95 % on GA, the customer earns credits against their invoice for the month:

Credits are issued automatically by the SLA enforcement engine (P11.32) on the first of the following month. Customers do not need to file a claim; the engine measures + issues. Credits appear on the next invoice.

Webhook delivery SLA

Receiver-side delivery latency (the time from the resource mutation to the customer's webhook handler receiving the signed payload):

The "receiver-side" measurement is at the customer's endpoint, not at Matter's exit. We measure with synthetic test endpoints in the same regions as the customer's deployed regions.

Credits. Receiver-side p99 above 10 s sustained for ≥ 30 minutes triggers a SEV2; credits apply per the availability table.

Out of scope. Receiver-side outages (customer's endpoint returning 5xx). Matter retries per the retry policy and the dead-letter queue holds the delivery; this is not a Matter SLA event.

CDC lag SLA (strict-consistency reads)

A read declaring Matter-Consistency: strict (or routing through an endpoint with x-matter-consistency: strict) returns data reflecting every write committed before the read began. The platform achieves this by routing strict reads to the primary.

A read declaring Matter-Consistency: eventual (or the operation-class default) returns data with a CDC lag bounded by:

Response carries X-Matter-Read-Lag-Window: <ms> advertising the worst-case current lag, never the SLA ceiling.

Credits. CDC lag p99 > 5× SLA for ≥ 15 minutes triggers a SEV2; credits apply.

What counts as downtime

Downtime for the availability calculation: a 1-minute window in which ≥ 50 % of synthetic probes from at least two regions failed with status ≥ 500, OR with a response shape that diverges from the spec (per the request-validation gate).

Not downtime.

• Customer-side errors (4xx, with rare exceptions). 429s above 5 % across the customer base do count.
• Maintenance windows pre-announced ≥ 7 days in advance via the status page + email. (Matter's commitment is no-downtime deploys by default; pre-announced windows are advertised "potential brief degradation".)
• Force majeure (regional cloud provider outage with no failover region available). Matter operates across providers per the capacity plan; a single-provider regional outage is normally absorbed.

Maintenance windows

• Pre-announced ≥ 7 days in advance via the status page + email.
• No-downtime by default. Rolling deploys, per-region. The advertised "potential brief degradation" is the worst case.
• A planned full-downtime maintenance has not occurred in Matter's history and is not currently planned. If one becomes necessary, it would be ≥ 30 days' notice.

Per-region SLA application

Once at capacity Stage 5 (multi-region; see capacity plan):

• The SLA applies per-region to customers pinned to that region. A US outage does not trigger EU credits.
• Cross-region writes (rare; migration events only) carry a relaxed p99 documented in the consistency contract.

Measurement

External synthetic probes from US (us-east) + EU (eu-central) + APAC (ap-southeast) run at 1-minute intervals. Probes:

• Cover every GA endpoint at least once per probe cycle.
• Hit cached + uncached + computed reads.
• Hit at least one sync mutation per cycle.
• Use test-mode tokens — never live mode.
• Report to the status page and the SLA enforcement engine.

Probe infrastructure runs on a different cloud provider than the primary API hosting, to avoid measurement-and-system colocation failure modes.

Webhook retry policy

For completeness — receiver-side delivery delays may originate at the customer's endpoint, not at Matter's exit. Matter retries on receiver failure:

After the 7th attempt, the delivery is dead-lettered. The customer can:

• View the dead-letter at GET /v1/events/{id}/deliveries.
• Replay via POST /v1/events/{id}/replay within the 30-day event-replay window.

Credit issuance

The SLA enforcement engine runs on the first of every month. For each customer:

1. Computes per-endpoint monthly availability from probe data.
2. Computes webhook delivery p99 from delivery metrics.
3. Computes CDC lag p99 from CDC consumer metrics.
4. For each SLO that fell below threshold, computes the credit per the credit table.
5. Posts the credit to the next invoice.
6. Emails the customer with the credit and the underlying measurement.

Credits never expire; they apply to invoices until exhausted.

If a customer disputes a measurement, they may file at support@mattermode.com. Matter responds within 5 business days with the raw probe data + the engine's calculation. Disputes are rare; the engine is deterministic.

SLA review cadence

This SLA is reviewed annually. Changes follow:

1. RFC at apps/docs/rfcs/ describing the proposed change.
2. API Council approval if the change tightens or relaxes commitments.
3. Customer comms ≥ 90 days before any reduction in commitment. Increases (tighter commitments) take effect immediately on the next month.
4. Legal document update in lockstep.

See also

• Matter API SLOs — internal performance budgets the SLA depends on.
• Capacity plan — how Matter scales while meeting the SLA.
• Quotas — per-tier rate limits.
• Metering — what's billable.
• Severity matrix — how outages are classified.
• Incident communications — how customers are notified.