Use the same workspace key for REST and MCP. Agents can read OpenAPI 3.1 for request and response schemas, or the complete plain-text reference for workflow instructions. API requests use https://app.pumpgtm.com/api/v1; Partner API requests use https://app.pumpgtm.com/api/partner/v1 with the same partner workspace key.
Check whether a lead follows your account
const base = "https://app.pumpgtm.com/api/v1";
const headers = {
Authorization: `Bearer ${process.env.PUMP_API_KEY}`,
"Content-Type": "application/json",
};
// Add people to a Play. They get its sequence on LinkedIn, email or X,
// whichever identities you give and the sequence uses.
const response = await fetch(base + "/plays/" + process.env.PUMP_PLAY_ID + "/leads", {
method: "POST",
headers,
body: JSON.stringify({
requestId: "crm-sync-001",
leads: [{ person: "Jane Doe", xUsername: "janedoe", email: "jane@acme.com" }],
}),
});
if (!response.ok) throw new Error(`Pump HTTP ${response.status}`);
const { addResult } = await response.json();
console.log(addResult.summary.addedCount, "added");
Read outreach analytics
GET /api/v1/analytics?range=30d returns the same report as Analytics: totals, daily activity, Plays with Sequences, message-step results, channels and eligible highest/lowest comparisons. MCP uses get_analytics with the same range and scope arguments. Ranges are 7d, 30d and 90d; the default scope is workspace. No Energy charge or outreach actions.
Use scope=team only with a personal OAuth or personal delegation key belonging to a current Team owner or admin in their home workspace. Shared workspace keys and the internal operator bearer do not identify a reporting member and cannot read a team report. Workspace scope supports authorized X-PumpGTM-Client selection or MCP workspaceId; team reporting never grants permission to manage those workspaces.
period.since is inclusive, period.until is exclusive and records the cached report's refresh time; dates use Pacific time. Enrolled is all-time current Sequence assignment. Reply rates are fractions and require a complete 14-day observation window; rankings require at least two non-tied results with 30 eligible people each. Null means no comparison is available. Clicks are unavailable and recorded meetings have partial coverage. The older /reports/funnel endpoint is unchanged.
curl 'https://app.pumpgtm.com/api/v1/analytics?range=30d&scope=workspace' \
-H "Authorization: Bearer $PUMP_API_KEY"One workspace key. REST or MCP.
Use the workspace key from the Connect MCP page as Authorization: Bearer <key>. Store it in your server's secret manager. It is the same credential used for MCP; never put it in browser code, URLs, logs, prompts or vendor requests.
Call GET /api/v1/workspaces without X-PumpGTM-Client first. A workspace key uses its own workspace by default. A Team owner key can select a returned workspace with X-PumpGTM-Client: <workspace UUID>. A selector cannot grant access to another tenant. Use the selected workspace consistently for discovery, checks, imports and webhooks.
All API calls require a key. This guide, the OpenAPI schema and Markdown reference are public and require no session. Keep numeric X user IDs as strings to avoid loss of precision.
Add people to a Play
POST /plays/{id}/leads with leads (1 to 500 people, each with any of linkedinUrl, email or xUsername) adds them to any Play in one call. They go straight into the Play's sequence, tagged with the Play, and get the first message in the next working window when the sequence is active. People already in the workspace are skipped. Pass a requestId to retry safely.
# Add people to a Play (any channel; people already in the workspace are skipped)
curl -X POST https://app.pumpgtm.com/api/v1/plays/$PLAY_ID/leads \
-H "Authorization: Bearer $PUMP_API_KEY" -H "content-type: application/json" \
-d '{"requestId": "crm-sync-001", "leads": [{ "person": "Jane Doe", "xUsername": "janedoe", "email": "jane@acme.com" }]}'Sequences and leads over REST
Every write route under /api/v1 runs the same service code as the matching MCP tool, so limits, approval rules and partner lead metering are identical. Your request is the approval: a sequence created with a workspace or partner key is live when POST /sequences returns, and adding people to a draft approves and starts it. Nothing is sent inside these calls; people enter the paced engine and are contacted within working hours, account limits and reply freezes.
Steps are the complete ordered list (1 to 20) of kinds connect, message, engage, email, x_dm, x_engage or instagram_dm. With any LinkedIn step, step 0 of a linear or tracks sequence must be connect (to engage before the connection request, use executionModel graph); a message needs messageTmpl or text on every variant; email, x_dm and instagram_dm need messageTmpl; the first email needs a subject; engage and x_engage need actions. GET /sequences returns a compact list (steps use index and message); GET /sequences/{id} returns the full shape (stepIndex, messageTmpl, executionModel). Read one sequence before PUT /sequences/{id}, send the full step list with the current revision; a 409 means nothing was overwritten. executionModel graph is a branching flow of up to 30 steps: each step has next ({ next: i }, or { yes: i, no: j } on a branch step with a condition: linkedin_accepted, x_followed_back, has_email or linkedin_first_degree). In a graph, a LinkedIn message or endorsement must come after a connection request and an 'Accepted?' (linkedin_accepted) or 'Already connected?' (linkedin_first_degree) Yes path; view, like and comment need no connection. A comment before the connection request needs the linkedin_comment_before_connect feature, runs only for people not yet connected or sent a request, and the request waits its delayDays from when the comment posts or is skipped.
Add people with POST /sequences/{id}/leads using method manual (1 to 500 LinkedIn URLs), csv (header row, at most 500 rows) or sales_navigator (a saved-list URL; poll GET /leads/imports/{batchId} until status is completed or failed). Manual/CSV/Sales Navigator need a connected LinkedIn account. Use contacts for email/X-only people; that method needs LinkedIn only when the supplied identities and sequence use it. Optional accountId selects the sender for new LinkedIn people (see the sender example below). Remove people with POST /leads/remove; reverse a removal with POST /leads/undo-remove using the exact removal ID (see the recovery example below). Find new people with POST /people/searches (run: true searches and returns a review batch; requestId is then required because a run is billed), record decisions with POST /people/reviews/{batchId}/decisions and finish: true to enroll the good fits (summary, addedLeadIds, addedLeads and sequences appear only on that finishing call). Read and decide replies with GET /replies and POST /replies/{id}/decide.
requestId matches ^[A-Za-z0-9][A-Za-z0-9._:-]{7,119}$ (8 to 120 characters, starts alphanumeric). For routes that accept requestId, it is optional except on a people-search run; when omitted a random key is generated, so a retry without it is a new request. Undo removal instead accepts exact removalIds and no requestId; retry those same IDs. Same requestId and same body returns the stored result with the same ids and reused (or replayed) true; same requestId and a different body is a 409 conflict code and changes nothing. Validation failures are 400 invalid_request with issues; a resource outside the workspace is 404 { error: "not_found", resource }; a service refusal is 409 { error, message } with the same customer-safe code the MCP tool gives, and changes nothing; an unexpected failure is 500 internal_error.
# Create a sequence (live on return; nothing is sent until people are added)
curl -X POST https://app.pumpgtm.com/api/v1/sequences \
-H "Authorization: Bearer $PUMP_API_KEY" -H "content-type: application/json" \
-d '{
"name": "Founders in fintech",
"steps": [
{ "kind": "connect", "delayDays": 0 },
{ "kind": "message", "delayDays": 1, "messageTmpl": "Hi {{firstName}}, thanks for connecting." }
]
}'# Add people to it (partners: add -H "X-PumpGTM-User: acme" to act for a client; the workspace's LinkedIn must be connected first)
curl -X POST https://app.pumpgtm.com/api/v1/sequences/$SEQUENCE_ID/leads \
-H "Authorization: Bearer $PUMP_API_KEY" -H "content-type: application/json" \
-d '{"method": "manual", "leads": [{ "linkedinUrl": "https://www.linkedin.com/in/example", "person": "Jane Doe", "company": "Acme" }]}'Move existing people safely
Preview POST /api/v1/leads/move/preview with { leadIds, sequenceId }. A move starts at the destination's beginning, so outreach must be untouched across LinkedIn, email and X. Existing sends, replies, exclusions, removals and uncertain delivery block moving. Play attribution, lead IDs and bound senders are preserved; changing a Play's Sequence for new leads does not move existing people.
POST /api/v1/leads/move with the reviewed selection, destination and one stable operationId UUID. Keep that exact request for recovery. The backend schedules execution after five seconds and checks eligibility again before moving. GET /api/v1/leads/move?operationId=... reads status without changing it. Moved people follow the destination's normal schedule, with existing personal pauses preserved. The move never approves messages or activates a sequence.
During the delay, POST /api/v1/leads/move/cancel with { operationId } requests cancellation. cancelled confirms that no people moved; running or completed means execution won the race. If it remains pending after executeAfter, the five-second cancellation window has still closed. Undo cancels pending work only and cannot restore a completed move. Inspect every confirmed per-person result, including blocked reasons. A failed operation may contain confirmed partial results. Compare requestLeadIds against results: missing people need review and must not be assumed blocked or unmoved. A 503 sequence_move_unconfirmed means read or retry the same operation and request, never create a replacement operation.
Agency partners: schedule an approved reply
Private API-only capability, enabled per partner. Add sendAt to POST /replies/{id}/decide with decision send, optional text/by and an Idempotency-Key header. LinkedIn, email and X are supported. sendAt must be a future RFC 3339 timestamp with a timezone, at most 30 days ahead. Attachments and other decisions are refused. Without sendAt, replies send immediately as before; invalid/null sendAt never falls back to sending now.
202 returns { ok, scheduledReply }. Poll GET /scheduled-replies/{id}; sent means provider acceptance, held needs review, and delivery_unknown must never be blindly retried or recreated. The minute worker treats sendAt as an earliest time, rechecks the original account and conversation, and holds stale approvals. This does not resume the Sequence or reassign the sender.
PATCH /scheduled-replies/{id} with sendAt, optional text, expectedVersion and a new Idempotency-Key edits unsent work. Held work also needs reapprove true and unchanged conversation evidence. POST /scheduled-replies/{id}/cancel is idempotent before the worker claims it; later cancellation returns 409. GET /scheduled-replies lists resources with status/channel/leadId filters, limit and opaque cursor. Read/cancel remain available when this scheduling capability is turned off.
For X without a draft, GET /leads/{id}/reply-decisions returns a verified replyVersion. POST on that path with channel x, decision send, text, sendAt, replyVersion and Idempotency-Key. The key must belong to the enabled partner; optionally select its active child with X-PumpGTM-User. Personal OAuth/member keys and operator keys cannot schedule. No frontend or MCP scheduling controls are added.
Undo a removal without saving its ID
You do not need the original remove response. For example, if John Doe was removed yesterday, call GET /api/v1/leads?q=John%20Doe without activity or stage filters. Match the correct person and sequence before using that row's removal.id. If you already know the lead ID, GET /api/v1/leads/{id} returns the same metadata under lead.removal. The customer never needs to keep track of removal IDs.
removal contains id, removedAt and undoAfter. null means no current Undo event is available; some historical removals cannot be reconstructed safely. Do not re-add a duplicate or invent an ID. A non-null removal still needs the normal Undo safety checks. If several people share a name, select the correct enrollment instead of restoring every match.
POST /api/v1/leads/undo-remove with { removalIds: [the returned removal.id] }. It restores the same lead, sender and progress, leaving paused sequences paused. It sends nothing immediately; active outreach may resume through the ordinary engine. Read restoration.results even on HTTP 200: restored means success, already_restored means that exact removal was previously undone and this call changed nothing, and blocked includes a reason/message. In-flight work can require ten minutes to settle; later replies, opt-outs, edits or uncertain sends can block Undo.
The example UUID below is illustrative: substitute the removal ID from your read response. Keep the credential and workspace selection identical for lookup and Undo. A partner managing a child workspace adds X-PumpGTM-User: <externalUserId> to both requests. MCP uses list_leads { query: "John Doe" } (or { leadId }) followed by undo_remove_leads { removalIds: [...] }. Reuse the exact removal IDs for retries; this endpoint does not accept requestId.
# 1. Find John; inspect the person, sequence and removal.id in the result.
curl --get 'https://app.pumpgtm.com/api/v1/leads' \
-H "Authorization: Bearer $PUMP_API_KEY" \
--data-urlencode 'q=John Doe'# 2. Use the returned removal.id, not the lead's id.
# Example only: replace this UUID with the value from step 1.
curl -X POST 'https://app.pumpgtm.com/api/v1/leads/undo-remove' \
-H "Authorization: Bearer $PUMP_API_KEY" \
-H 'Content-Type: application/json' \
-d '{"removalIds":["00000000-0000-4000-8000-000000000004"]}'Pick a LinkedIn sender and read the conversation
Partners list a client's accounts with GET /api/partner/v1/users/{externalUserId}/linkedin/accounts, then pass a returned id as accountId when adding leads to a sequence or adding supplied contacts to a Play. MCP gets the id from get_workspace → channels.linkedin.data.senders.items. Keep the same workspace on the account lookup, add and thread read. An account from another client, or a disconnected account, is refused. Omitting accountId retains automatic assignment.
accountId applies to manual, CSV, contacts and Sales Navigator imports. New LinkedIn leads keep that sender; duplicate or removed leads are never reassigned or revived. Email/X-only people do not acquire a LinkedIn sender. Changing the selected account with the same requestId is a conflict. Sales Navigator uses an explicit selection to fetch the list and send; existing saved imports keep their original behavior.
GET /api/v1/leads/{leadId}/thread reads the lead's LinkedIn conversation from its original sender. It sends nothing and does not mark messages read. Each page is oldest first; nextCursor reads the next older page. Prepend those older pages until nextCursor is null. Cursors expire after 30 minutes and cannot be moved between leads or workspaces. limit is 1–100 (default 50). The first version returns message text, direction, timestamp, opaque id and attachment presence; it does not download files or read email, X or Instagram threads.
available=false means no LinkedIn thread has been recorded for this lead yet. A disconnected sender or a provider error returns an error, never an empty successful history. Respect Retry-After on HTTP 429. MCP uses get_lead_thread with leadId and the same optional limit/cursor. No sendAt scheduling is introduced by these endpoints.
// Replace IDs with values returned from account and sequence reads.
await fetch(base + "/sequences/" + sequenceId + "/leads", {
method: "POST", headers,
body: JSON.stringify({ method: "manual", requestId: "sender-add-001", accountId,
leads: [{ person: "Jane Doe", linkedinUrl: "https://www.linkedin.com/in/example" }] }),
});
// leadId comes from addResult.addedLeadIds, GET /leads, or a pending reply.
const response = await fetch(base + "/leads/" + leadId + "/thread?limit=50", { headers });
if (!response.ok) throw new Error(`Pump HTTP ${response.status}`);
const { thread } = await response.json();
// Older page: append "&cursor=" + encodeURIComponent(thread.nextCursor).Events your agent can act on
Subscribe with POST /webhooks. For X workflows, use reply.received, lead.messaged and x.followed. x.followed represents an inbound follow of your connected account. Successful X DMs emit lead.messaged with channel: x. X replies have no generated draft; encrypted message text can be unavailable. Partners register linkedin.connected with the partner key to learn when an end user completes a hosted connect link.
Save the signing secret from the creation response; it is shown once. Each delivery contains { id, event, createdAt, data }. Validate X-PumpGTM-Signature (your HTTP stack hands it to you lowercased as x-pumpgtm-signature), formatted t=<unix>,v1=<hex>, where v1 is the hex HMAC-SHA256 of t + '.' + the exact raw request body. Reject deliveries whose t is older than five minutes and deduplicate on id. X-PumpGTM-Delivery also carries this event ID.
Delivery is best-effort: up to three attempts, retried on network errors, 5xx and 429. Other non-2xx responses end delivery; redirects are not followed. There is no durable replay queue or event history endpoint in this release. Poll lead state and activity to reconcile missed events. A test returning 202 means scheduled, not delivered.
Receivers must use HTTPS on port 443 and a public IPv4 DNS address. Private or mixed public/private addresses, URL credentials, fragments, IP literals and redirects are rejected. Return 2xx quickly and do long-running work in your own queue. Registration is not idempotent; list subscriptions before retrying an uncertain creation.
import { createHmac, timingSafeEqual } from "node:crypto";
// rawBody must be the exact bytes received, BEFORE JSON parsing.
export function verify(rawBody: Buffer, header: string, secret: string) {
const match = /^t=(\d+),v1=([a-f0-9]{64})$/.exec(header);
if (!match) return false;
const timestamp = Number(match[1]);
if (Math.abs(Date.now() / 1000 - timestamp) > 300) return false;
const expected = createHmac("sha256", secret)
.update(match[1] + ".").update(rawBody).digest();
return timingSafeEqual(expected, Buffer.from(match[2], "hex"));
}
// Verify X-PumpGTM-Signature, then parse the JSON.
// Persist event.id to deduplicate; enqueue work and return 2xx quickly.Handle uncertainty and retries
401: check or rotate the workspace key. 403: check workspace access or X channel availability. 404: the resource is not accessible in this workspace, or the Play has not enabled review APIs. 409: fetch current state, inspect the conflict, then decide whether a new request is appropriate. Do not blindly change the request ID and repeat a mutation.
On 429, respect Retry-After when provided. Retry read-only calls on transient 5xx with exponential backoff and jitter. Follower checks can return HTTP 200 with individual null results when a provider is unavailable: always inspect each result. Identity resolution returns 503 and status unknown when it cannot verify a profile.
Pagination uses nextCursor: candidate and enrolled-lead lists accept it as after; LinkedIn thread reads accept it as cursor. Stop when nextCursor is null. Candidate and enrolled-lead pages support at most 200 rows. Eligibility supports at most 100 inputs, but small batches avoid the bounded execution deadline. The API does not offer a bulk asynchronous job in this release.
POST /x/leads/{id}/stop halts future automation across this lead's sequence lanes. It preserves history, does not permanently suppress the person, and cannot recall actions already dispatched. automationStopped on status means the sequence is done or stopped; it does not identify the reason.
Usage and costs
The key itself is not a separate X credential. Customer costs follow your PumpGTM plan and Energy rules. Read /x/readiness for the current balance and billing state; it is not a quote or reservation for future work.
Known identities supplied through this import API are customer-provided leads, like uploads, and do not count as paid audience discovery. Public identity and relationship reads use Pump's public X data provider and are metered in the provider ledger. This reference does not promise those upstream reads are free or set a new per-request retail price; confirm your commercial allowance with Pump.
Sending still uses the existing X execution engine, its connected account, Energy checks and dispatch limits. Eligibility does not reserve Energy or activate a campaign. No official-X read fallback is used when the public-data provider is unavailable.
Partners: run PumpGTM for your clients
Agencies and platforms use one partner key (the workspace key of the account that signed up through POST /api/partner/v1/signup, shown on /api-keys, where it can also be rotated) and one isolated workspace per client. Clients never need a PumpGTM login. Partner routes live under https://app.pumpgtm.com/api/partner/v1 and identify a client by your own externalUserId (letters, digits and _ . : @ -, up to 120 characters).
Hosted LinkedIn connect and outreach for clients are enabled per partner by PumpGTM: call GET /api/partner/v1/account and confirm enterprise.linkedinConnect and enterprise.outreach are true; until then those routes answer 403 enterprise_required. Only a workspace key or a Team owner key can act for clients; a personal OAuth-issued key gets 403 partner_key_required.
1. Create the client with POST /api/partner/v1/users. It is idempotent: 201 when created, 200 when it already existed. Store mcpUrl from the response (https://mcp.pumpgtm.com/mcp/u/{externalUserId}); rest is null and outreach is "disabled" until outreach is enabled on your account.
2. Connect the client's LinkedIn from your own dashboard (enterprise, enterprise.linkedinConnect): set your brandName and successRedirectUrl/failureRedirectUrl once with PATCH /api/partner/v1/account, then mint a link per click with POST /api/partner/v1/users/{externalUserId}/linkedin/connect-links and open the returned url with window.open(url, "connect-linkedin", "popup,width=520,height=720"). Never use an iframe: LinkedIn verification prompts do not work inside frames. Links are single use and expire (7 days by default).
3. Learn when it is done three ways: a signed linkedin.connected webhook registered once with the partner key through POST /api/v1/webhooks (data carries connectLink.externalUserId and account.id/displayName), the browser redirect to your success URL with externalUserId and connectLinkId appended (only expired or revoked links use the failure URL), and GET .../connect-links/{linkId} polled every 3 seconds until status is not pending (connected, expired or revoked are final), giving up after 15 minutes. The three signals have no ordering guarantee; treat each as idempotent on connectLink.id. Manage identities with GET and DELETE .../linkedin/accounts.
4. Run outreach for a client over REST (enterprise, enterprise.outreach) by adding X-PumpGTM-User: <externalUserId> to any /api/v1 request made with the partner key, never together with X-PumpGTM-Client (400 invalid_request); without the header the key means your own partner workspace. The user must already exist and be active (404 user_not_found, 403 user_disabled; 403 enterprise_required when the capability is not enabled). Manual/CSV/Sales Navigator imports need the client's LinkedIn connected first (409 linkedin_not_connected); contacts can add email/X-only people without LinkedIn. Over MCP, point your agent at the client's mcpUrl with the same bearer.
5. You pay for what your clients do; they never see a card prompt. People search: 50 free leads per partner, then 403 billing_required with a checkout link in message; with Energy on your account each person found costs 2 Energy from your pool instead of the per-lead meter (402 energy_exhausted when empty). Cap any client with PATCH /api/partner/v1/users/{externalUserId} { monthlyLeadCap } (403 user_lead_cap_reached at the cap). GET /api/partner/v1/users lists every client with the same fields as GET .../users/{externalUserId} plus usage; DELETE (or PATCH { status: "disabled" }) disables a client: active sequences pause, requests answer 403 user_disabled, data is kept and nothing more is billed; PATCH { status: "active" } reactivates.
# Create a client workspace (idempotent: 201 created, 200 existing)
curl -X POST https://app.pumpgtm.com/api/partner/v1/users \
-H "Authorization: Bearer $PARTNER_KEY" -H "content-type: application/json" \
-d '{"externalUserId": "acme", "name": "Acme Robotics"}'# Mint a hosted LinkedIn connect link, then open url in a popup (never an iframe)
curl -X POST https://app.pumpgtm.com/api/partner/v1/users/acme/linkedin/connect-links \
-H "Authorization: Bearer $PARTNER_KEY" -H "content-type: application/json" -d '{}'
# window.open(url, "connect-linkedin", "popup,width=520,height=720");# Register once, with the partner key, to learn when a client's LinkedIn is connected
curl -X POST https://app.pumpgtm.com/api/v1/webhooks \
-H "Authorization: Bearer $PARTNER_KEY" -H "content-type: application/json" \
-d '{"url": "https://your-app.example/hooks/pumpgtm", "events": ["linkedin.connected"]}'X: check whether someone follows you
1. Discover the workspace and connected sender with GET /workspaces and GET /x/accounts. Read GET /x/readiness for saved account, billing and Energy state. This snapshot does not contact X or prove delivery is possible.
2. If your lead has a known X handle, call POST /x/eligibility directly. If you also need the numeric ID for import, call POST /x/users/resolve. Email-only leads require your own verified email-to-X enrichment provider; Pump does not currently offer complete email-to-X matching.
3. Inspect followsAccount: true means the lead follows your selected account; false is a verified negative from a direct handle relationship check; null means unknown. Missing IDs in a public follower snapshot remain unknown. Check checkedAt for freshness. A lead following you is different from your account following them.
4. Offer X through Pump only when your workflow's channel policy allows it. The eligibility call never enrolls or sends. A positive follower result is not a guarantee that the recipient can receive a DM.
5. When you intend outreach, add them to a Play with POST /plays/{id}/leads.
6. Track lead status and activity, receive signed webhooks, and stop the lead when needed. Reply handling already freezes automated follow-ups. X reply sending and sequence/Play authoring are outside this core release; use the product and X to manage them.
Advanced: manual review
Most integrations never need this. It is for Plays where a person approves each lead before anyone is messaged. To message people, use POST /plays/{id}/leads.
POST /plays/{id}/import stages 1–100 unique numeric X identities as review candidates. requestId is required (8–120 letters, digits, periods, underscores, colons or hyphens; first character alphanumeric). Scope it to one logical import. Retry an identical request after a timeout; changing its payload or revision with the same requestId returns 409. The transaction commits the whole import or nothing.
Use the returned candidate IDs, or POST /plays/{id}/candidates/lookup, to avoid duplicates. Existing candidates keep their review state. Import names are caller-provided labels; import never asserts that a lead follows you. Resolve a public handle first if you do not have the numeric ID.
Read GET /plays/{id}/candidates to get the latest Play revision and selected sequence revision. POST /plays/{id}/enroll requires both plus candidateIds. Repeated completed good-fit reviews return already_reviewed. Existing enrolled, stopped or replied lead records are preserved; they are not moved to another sequence or reactivated.
Only Plays explicitly enabled for the review API support these operations. Check reviewApiEnabled on /x/plays. Imports require manual review, not autopilot. Sequence state and approvals stay unchanged. A paused sequence stays paused; enrollment alone does not guarantee a send.