Returns delivery and deliverability counts grouped by the specific IP address used to send each message. Rows are ranked by the `sort` field (default `delivered`) descending and capped at the requested `limit` (default 50, hard maximum 200). Use this to identify per-IP reputation issues — block bounces concentrated on a single IP usually indicate a reputation problem on that IP, and `sort=bounces.block` surfaces those IPs first. Rows are computed against event time (not send time), so deliveries and bounces received during the period for messages sent earlier are included. **Per-IP scope.** Bird only learns which sending IP a message used after the upstream mail-transfer system reports delivery, bounce, deferral, or late bounce. Earlier lifecycle states — accepted and processed — cannot be attributed to a specific IP, so per-IP responses omit the `accepted` / `processed` counts and the `processing` latency percentile. For workspace-wide accept-and-process counts and processing-latency percentiles, use `GET /v1/email/stats/daily`. The `delivery` and `total` latency percentiles are computed across deliveries on each IP and are null if no deliveries occurred for that IP in the period. The maximum window is 365 days; requesting a longer range returns 422. Start date (inclusive) in YYYY-MM-DD, interpreted as a calendar day in `timezone` (a UTC day when `timezone` is omitted). Defaults to 30 days before `to` when omitted; with `include_trend=true` and `trend_grain=hourly` the default tightens to 29 days before `to`, keeping the defaulted window within the 720-hour trend cap. End date (inclusive) in YYYY-MM-DD, interpreted as a calendar day in `timezone` (a UTC day when `timezone` is omitted). Defaults to today in that timezone when omitted. Window may not exceed 365 days. IANA timezone identifier (for example `Asia/Kathmandu` or `America/New_York`) to report the statistics in. It is the single source of timezone: day and hour boundaries, and the relative window defaults used when `from` and `to` are omitted, are computed in this timezone instead of UTC, so a timezone with a sub-hour offset (such as India at +05:30 or Nepal at +05:45) still gets correct local-day and local-hour totals. When it is set, a `from` or `to` given as a calendar day names a local day in this timezone, and one given as an instant stays an absolute point in time but is rounded down to its hour and bucketed in this timezone (so the hour boundaries are local, not UTC). To avoid specifying the zone twice, a `from` or `to` that carries its own numeric UTC offset (for example `+05:45`) is rejected when `timezone` is set; use a calendar day or a `Z` (UTC) instant instead. Defaults to UTC. Email statistics are aggregated across all categories. Per-category filtering is not yet available; supplying this parameter returns 422. Metric to rank IPs by, applied descending. Any count or rate in the response may be used; `bounces.block` surfaces the IPs whose reputation is most likely degraded. Rows whose rate is undefined (zero denominator) sort last. Defaults to `delivered`. The engagement metrics (opens, clicks, unsubscribes and their rates) and the pre-delivery metrics (`processed`, `rejected`, `oob_bounces`) are not available for this breakdown -- a sending IP carries no engagement and is unknown until delivery -- so they are absent from the enum below; supplying one returns 422. Maximum number of IP rows to return, ranked by the `sort` field descending. When true, each row also carries a `trend` array: a short per-bucket series of that IP's delivery rates over the window (per-IP rows have no engagement, so each trend point's open and click rates read 0 in buckets that had deliveries and null in buckets that had none). Returned only when `limit` is 50 or fewer and the window is at most 90 days (trend_grain=daily) or 720 hours (trend_grain=hourly); a larger request returns 422. When `from` is omitted and `trend_grain=hourly`, the default window is 30 days (720 hours), so a request built entirely from defaults always fits the cap. Bucket grain for the `trend` series. Has no effect unless `include_trend=true`. Inclusive start date the response covers (YYYY-MM-DD). Inclusive end date the response covers (YYYY-MM-DD). The instant the statistics in this response are current to: events recorded up to roughly this time are reflected, while more recent events may not be yet. Statistics are served from a rolling aggregation that refreshes every few seconds, so a response is near-real-time but not live; use this field to label data freshness (for example "as of 14:03") rather than assuming the numbers are to-the-second. Null when the freshness boundary is not being reported. Sending-IP breakdown rows, ranked by the `sort` metric (default `delivered`) descending. Empty when no per-IP-attributable activity (delivery, bounce, deferral, or late bounce) occurred in the period. The IP address used to send messages aggregated in this row. The dedicated IP pool this address sent through, or null when the messages went through the shared pool. Recorded when each message was sent, so it reflects the pool used at send time even if the IP has since moved between pools or been released. Distinct recipients whose message the receiving mail server accepted. Distinct recipients whose delivery failed. Approximately the sum of the five `bounces.*` sub-counts (hard, soft, admin, block, undetermined); the totals are computed independently so they may differ slightly at the approximation error. Distinct recipients who reported the message as spam. Distinct recipients in transient delivery deferral that is still being retried. Out-of-band bounce events attributed to this IP: failure notifications received after the receiving server had initially confirmed delivery. Counted as deduplicated events, not unique recipients. Recipients on this IP who remain in-inbox after all bounce signals resolve, computed as `delivered - oob_bounces`. Clamped to 0 when `oob_bounces` exceeds `delivered`. Total recipients on this IP who did not receive the message, computed as `bounced + oob_bounces`. Share of this IP's delivery attempts that resulted in an out-of-band bounce, computed as `oob_bounces / (delivered + bounced)`. Null when `delivered + bounced` is zero (no attempts). Distinct recipients with a permanent delivery failure (invalid address or non-existent domain). Distinct recipients with a transient delivery failure (mailbox full or server temporarily unavailable). Distinct recipients bounced by an upstream policy block (relaying denied, blocklisted domain). Distinct recipients bounced because the receiving mail server blocked the sending IP for reputation reasons. Distinct recipients bounced where the receiving server's response did not allow precise classification. Fraction of bounced recipients that hard bounced, computed as `hard / bounced`. Null when `bounced` is zero. Fraction of bounced recipients that soft bounced, computed as `soft / bounced`. Null when `bounced` is zero. Fraction of bounced recipients that admin bounced, computed as `admin / bounced`. Null when `bounced` is zero. Fraction of bounced recipients that block bounced, computed as `block / bounced`. Null when `bounced` is zero. Fraction of bounced recipients with undetermined classification, computed as `undetermined / bounced`. Null when `bounced` is zero. Share of this IP's delivery attempts that resulted in a message remaining in-inbox, computed as `effective_delivered / (delivered + bounced)`. Null when `delivered + bounced` is zero (no attempts). Share of this IP's delivery attempts that ultimately failed (inband or out-of-band), computed as `all_bounces / (delivered + bounced)`. Null when `delivered + bounced` is zero (no attempts). Share of effectively delivered recipients on this IP who reported the message as spam, computed as `complained / effective_delivered`. Null when `effective_delivered` is zero. p50, p95, and p99 latency percentiles in milliseconds for one latency family over the bucket. Percentiles are approximate (computed from a high-volume aggregation pipeline). All three are null together when no qualifying event contributed a latency measurement in the bucket. Median (50th percentile) latency in milliseconds. Null when no qualifying event contributed a measurement. 95th percentile latency in milliseconds. Null when no qualifying event contributed a measurement. 99th percentile latency in milliseconds. Null when no qualifying event contributed a measurement. p50, p95, and p99 latency percentiles in milliseconds for one latency family over the bucket. Percentiles are approximate (computed from a high-volume aggregation pipeline). All three are null together when no qualifying event contributed a latency measurement in the bucket. Median (50th percentile) latency in milliseconds. Null when no qualifying event contributed a measurement. 95th percentile latency in milliseconds. Null when no qualifying event contributed a measurement. 99th percentile latency in milliseconds. Null when no qualifying event contributed a measurement. Per-bucket delivery-rate series for this IP over the window. Present only when `include_trend=true`. Open and click rates are null on every point, since engagement is not attributed to a sending IP. The day (YYYY-MM-DD) or hour (ISO 8601, on the hour) this point covers, matching the requested `trend_grain`. Delivered recipients in this bucket. Bounced recipients in this bucket. Delivery rate for this bucket, as a fraction. Null when nothing was delivered or bounced. Bounce rate for this bucket, as a fraction. Null when nothing was delivered or bounced. Complaint rate for this bucket, as a fraction; event-time attribution can push it above 1 when complaints outrun the bucket's deliveries. Null when nothing was delivered in the bucket. On a sending-IP row complaints are not attributed to the IP, so this reads 0 in buckets that had deliveries and null in buckets that had none. Open rate for this bucket, as a fraction; event-time attribution can push it above 1 when opens outrun the bucket's deliveries. Null when nothing was delivered in the bucket. On a sending-IP row engagement is not attributed to the IP, so this reads 0 in buckets that had deliveries and null in buckets that had none. Click rate for this bucket, as a fraction; event-time attribution can push it above 1 when clicks outrun the bucket's deliveries. Null when nothing was delivered in the bucket. On a sending-IP row engagement is not attributed to the IP, so this reads 0 in buckets that had deliveries and null in buckets that had none. Total number of distinct sending IP addresses with activity in the period, regardless of `limit`. When it exceeds the number of rows returned, the ranking was capped; raise `limit` (up to 200) or narrow the window to see more.