Discovery and Radar

Control Rankhog's independent proactive and reactive lanes and manage every result through Opportunities.

Each product has one strategy with two independent lanes.

Radar

Radar watches your verified communities and shared keywords. It evaluates new Reddit posts and comments continuously and surfaces timely, relevant work.

  • Radar is enabled by default after input setup.
  • Disabling it stops new evaluations, not existing conversations or history.
  • Radar is independent from Discovery and account warm-up.
  • Time-sensitive Radar work does not consume the Discovery daily limit.

New products show Setting up while Rankhog derives keywords and candidate communities from your product context.

Review Radar finds

Radar checks what a discussion is about against what your product does before showing it. Ordinary opportunities need a concrete product fit and a score of at least 50. Keyword overlap, unrelated launches, and already-solved needs are filtered out.

Genuine product and competitor mentions can appear at lower scores after their identity is verified in context. Incidental tool names do not qualify. During a classifier outage, new candidates wait internally. Existing cards keep their current lifecycle.

The inbox puts other decisions needing you first, then Radar finds by relevance. Choose Load more to reach lower-ranked finds. Copilot drafts a reply when you ask. Strong autopilot finds can prepare a draft after appearing; a passing audit and the existing intervention window still precede sending. Drafting failure leaves the discussion available.

list_opportunities returns nextCursor. Pass it as cursor with the same workspace and includeRecent option to continue. An older mention's classificationState: "classification_pending" means its numeric confidence is a storage placeholder, not a relevance judgment.

Shared read capacity

Radar creates watcher demand only for an active paid product or an unexpired trial with completed setup, verified communities, and an active Reddit account. When hosted collection is enabled, Rankhog supplies collection capacity and no desktop contribution is required. Products watching the same community share its scans. Normal refreshes target five minutes, with faster scans when activity requires them. A recent scan does not prove complete coverage; Stats shows freshness and unresolved catch-up separately.

When hosted collection is disabled, an organization must also make one read-ready Rankhog Desktop available. The desktop may use a signed-in or anonymous Reddit browser; availability counts even when Rankhog selects another node. Browser extensions do not count because they cannot collect Radar activity.

  • Before the first contribution, the Radar preference stays enabled but status reports pool_contribution_required. Open Rankhog Desktop to start automatically.
  • A desktop seen within seven days keeps normal freshness.
  • From day 7 through day 14, exclusive communities use idle capacity with no freshness promise.
  • At day 14, Rankhog switches Radar off. A returning desktop restores only the Radars Rankhog switched off, never one you paused.

If multiple products watch one community, one eligible recent contributor keeps the shared read at normal priority. Inactive or unpaid products are not evaluated just because another product caused the read.

Adding a community

Add up to 30 watched communities and 20 keywords per product, on the Radar tab or through the API.

A community Rankhog has a guide for starts watching right away. Any other one is saved to your list but does not become an active watcher yet: Rankhog tells the team, and it starts watching once its guide is ready. Nothing you add is ever discarded, and a community that is waiting is shown as waiting rather than silently doing nothing.

Discovery

Discovery is proactive and starts only when you ask it to. It runs daily at 09:00 in the organization's timezone.

Discovery:

  1. Searches ordinary web results, then Reddit-restricted results for relevant conversations.
  2. Verifies Reddit threads through Rankhog's read pool.
  3. Extracts a useful pattern without copying the source.
  4. Offers a useful comment and, after three days, a competing original post.
  5. Audits drafts against current evidence and community rules.

A result includes the actual query and search type. A Reddit-restricted result means a conversation was found, not that it ranks for the ordinary buyer query. Search placement does not prove Google traffic or inclusion in an AI answer. Older threads can qualify for comments when they remain open and relevant. Confirm reviewed text posts to send them through Rankhog Desktop. If Reddit needs a flair or another manual step, the action shows Human needed. Complete the step in Desktop, then choose Done, continue; Rankhog checks for an existing submission before resuming.

Discovery rotates through every enabled keyword, checking up to six per scan and analyzing up to three candidates. Temporary read or audit failures leave saved finds available for a later scan.

You can edit your direction while Discovery runs. Changes to product context, market, direction, or enabled keywords prevent older scans from surfacing new work under the previous inputs. Previously offered or dismissed actions are not recreated just because the strategy changed.

Manual runs use the same scan path and allow one accepted start per product per hour. Discovery can surface at most five Reddit-action opportunities per organization-local product day.

Reviewing a quiet strategy

After three completed empty daily scans with coverage of every enabled keyword, Discovery explains what it searched and offers up to three grounded changes. Manual runs, technical failures, incomplete coverage, and pending work do not count toward that threshold. If there are no enabled keywords, it asks you to add a buyer query.

Discovery keeps scanning. You receive one review per dry episode through your strategy-notification preferences. Review keyword or community proposals in Opportunities. Suggested direction text can be loaded into the direction editor, where you can edit it before saving. Configuration changes always need a human, including in autopilot.

Opportunity conversations

Every opportunity has one Chat SDK conversation and one executable action. A single search winner may create a separate comment opportunity and original-post opportunity so each can be revised or dismissed independently.

Opportunities can wait for an eligible Reddit account; sending still requires a non-warming account that passes the execution checks. If a system-selected account becomes unavailable, Rankhog can reassign it and records the change in the timeline. A user-selected account is pinned; Rankhog blocks and asks instead of switching silently.

Mark an opportunity done

Choose Mark as done beside Delete when you have handled an opportunity. For a Reddit post, comment, or reply, paste its permalink to track the score and replies, or leave the field empty and add it later. You can also tell the opportunity's chat, “I posted it at [your Reddit permalink].” Rankhog marks it done and starts checking that link.

Completed Reddit opportunities offer Add Reddit link or Edit Reddit link. You can add or correct the link in chat too. This updates tracking without restarting the work. If Rankhog is already submitting the action, wait for that execution to settle before marking it done.

Performance appears above the conversation and under Manually posted results on the Stats page. Checks use public Reddit metrics; views and visits are not available. A pending or unavailable check is not a zero. Manual results stay separate from Rankhog submissions, and duplicate links count once.

Public capabilities

REST, OpenAPI, MCP, and the CLI use the same capability layer:

get_workspace_strategy
get_radar_status
get_radar_coverage
start_discovery
stop_discovery
run_discovery_now
set_radar_enabled
update_strategy_inputs

get_workspace_strategy includes Discovery coverage, the current review, and typed search evidence. Reads require rankhog:read. Mutations require strategy:write, a human OAuth or CLI-device actor, an idempotency key, and rate limits. Organization API keys can read strategy state but cannot mutate it. Read priority is internal and cannot be changed through REST, MCP, or CLI.

get_radar_status can return the warning code results_delayed while its state remains live. This means Radar accepted one or more results but their cards are still being retried. The warning clears automatically after the last affected result surfaces, is rejected, or expires.

Its activity covers the last 14 days. activity.days lists one entry per day in your organization's time zone, oldest first, with the conversations Radar checked and the cards it surfaced that day. The totals are sums over those days; waitingCount counts every open card whatever its age.

update_strategy_inputs replaces the whole list. Communities Rankhog has a guide for come back verified; the rest come back pending and start watching once their guide ships. Read verificationStatus on each subreddit in the response to tell them apart.

CLI examples

rankhog --json strategy show
rankhog --json strategy radar coverage
rankhog --json strategy discovery start \
  --direction "Find comparison demand" \
  --idempotency-key "start-discovery-001"
rankhog --json strategy discovery run \
  --idempotency-key "run-discovery-001"
rankhog --json strategy discovery stop \
  --idempotency-key "stop-discovery-001"
rankhog --json strategy radar disable \
  --idempotency-key "radar-off-001"
rankhog --json strategy inputs \
  --keywords "Intercom alternative" "support software" \
  --subreddits SaaS CustomerSuccess \
  --idempotency-key "strategy-inputs-001"

Set RANKHOG_ORGANIZATION_ID and RANKHOG_WEBSITE_ID, or pass the matching flags to each command.

Next: review Reddit account protection or use the generated API reference.

Legacy unclassified mentions return confidenceScore: null and classificationState: "classification_pending" through the public API. Their storage placeholder is never a relevance score.

Radar activity

Radar shows the work it does for your product over the last 14 days:

  • Posts and comments screened: screening passes, including items that did not match and genuine rechecks after targeting changes. Retries do not count.
  • Opportunities found: relevant discussions brought to your inbox. The total covers the last 14 days. Today opens Radar opportunities found on the current reporting day, including completed and dismissed cards. Remove the filter at the top of the list to see other opportunities. Deleting a card does not erase the work from this total.
  • Opportunities handled: cards completed or dismissed, counted once. Expired cards do not count as handled.

The chart uses your workspace timezone and shows when the snapshot was updated. Days before screening measurement began are marked unavailable. Older opportunity history may be incomplete; the page labels it. Statistics refresh as work commits, with short caching, and retain a clearly marked saved snapshot if refresh fails.

Agents can read the same measurements with get_radar_activity.

On this page