Overview

The Performance Dashboard gives product, engineering, and data teams a single view of how fast the core recruiting workflow pages are loading for recruiters, and how healthy the analytics ETL pipeline is behind the scenes.

The Performance Dashboard appears as the Performance Insights tab on the Instance Health page in the Admin Console. The dashboard also shows how quickly the backend APIs behind Talent Acquisition, Talent Management, and your career site respond. When users report that the product feels slow, you can check the numbers yourself.

This guide walks through every metric on the dashboard — its definition, how it's calculated, and anything you should know when interpreting the numbers.

Metrics

P50 Latency

Definition: P50 Latency measures the median (50th percentile) page-load latency across every recognized product page tracked in Eightfold, for the selected date range. It's the dashboard's headline "typical experience" number — half of all qualifying page loads complete faster than this value, and half complete slower.

How It Is Calculated: P50 Latency is the 50th percentile of load latency across qualifying page-load events, restricted to production traffic (dev, staging, QA, and CI hostnames excluded), non-bot interfaces, successful responses (response code under 400), and real full-page navigations — an initial visit or a navigation from another page, not a background component refresh. No page-type filter is applied, so every recognized page (Positions Home, Pipeline Leads, Pipeline Applicants, Full Profile, Minimal Profile, Community Home, and others) is blended into one number.

Note:

• Because this card blends every page type together, a slowdown on a single page can be masked if that page represents a small share of overall traffic. Use the P50 Latency by Endpoints chart to see page-by-page performance.
• Events that don't resolve to one of the dashboard's recognized page types are excluded from this metric entirely, not just hidden from view.
• Filtered by event date, via the dashboard's Date Range selector (default: last 30 days).

P75 Latency

Definition: P75 Latency measures the 75th-percentile page-load latency across every recognized product page, for the selected date range. Three out of every four page loads complete at or faster than this number — it's the "somewhat slower than typical" checkpoint teams watch alongside the median to catch experience degradation before it shows up in the tail (P90/P99).

How It Is Calculated: Same calculation and filters as P50 Latency, using the 75th percentile instead of the 50th.

Note:

• Blends all page types into one number — see the by-endpoints breakdown for page-level detail.
• Filtered by event date, via the Date Range selector.

P90 Latency

Definition: P90 Latency measures the 90th-percentile page-load latency across every recognized product page, for the selected date range. This is the slow-tail number for the overall product — nine out of ten page loads complete faster than this, so a rising P90 with a flat median usually signals that a specific page type or user segment is degrading even while most traffic still feels fine.

How It Is Calculated: Same calculation and filters as P50 Latency, using the 90th percentile instead of the 50th.

Note:

• A rising P90 alongside a flat P50/P75 is the signal the "by endpoints" charts further down are meant to help isolate.
• Filtered by event date, via the Date Range selector.

P99 Latency

Definition: P99 Latency measures the 99th-percentile page-load latency across every recognized product page, for the selected date range. This is the extreme-tail number — only one in a hundred page loads is slower than this — making it the most sensitive of the four percentile cards to rare-but-severe incidents (an overloaded shard, a customer with an unusually large candidate pool, a transient network blip) that would otherwise be invisible in the median or even P90.

How It Is Calculated: Same calculation and filters as P50 Latency, using the 99th percentile instead of the 50th.

Note:

• Because it captures rare events, this number can be noisier day-to-day than P50/P75/P90 — look for a sustained trend rather than reacting to a single day's spike.
• Filtered by event date, via the Date Range selector.

P50 Latency by Endpoints

Definition: This chart breaks the blended P50 Latency card down by individual page/endpoint, further split by navigation type (a fresh, first-time visit vs. navigating to the page from elsewhere within the product). It lets you see at a glance which specific pages are pulling the overall median up or down, and whether initial loads are systematically slower than in-app navigations.

How It Is Calculated: The 50th percentile of load latency, grouped by page/endpoint and by navigation type, using the same base filters (production traffic, non-bot, successful response, real page loads only) as the aggregated latency cards.

Note:

• Only page types the dashboard recognizes appear as bars — unrecognized page events are excluded from the underlying dataset.
• Filtered by event date, via the Date Range selector.

P90 Latency by Endpoints

Definition: This chart breaks the blended P90 Latency card down by individual page/endpoint, further split by navigation type. It surfaces which specific pages sit furthest into the slow tail, complementing the P50 breakdown by showing where a page can look fine on the median but still produce a rough experience for a meaningful share of users.

How It Is Calculated: The 90th percentile of load latency, grouped by page/endpoint and by navigation type, using the same base filters as the aggregated latency cards.

Note:

• Use alongside the P50 breakdown — a page that looks healthy at P50 but poor at P90 has a smaller group of users having a materially worse experience.
• Filtered by event date, via the Date Range selector.

Positions Home Page P75 Latency

Definition: Tracks the daily 75th-percentile load time for the Positions Home page — the main landing page recruiters see when opening the positions/pipeline area of the product.

How It Is Calculated: The 75th percentile of load latency for events where the page resolves to the Positions Home page, trended by day, using the same base filters as the aggregated latency cards (production traffic, non-bot, successful response, real page loads only).

Note:

• Trended daily and filtered by event date, via the Date Range selector.
• Compare against the P90 trend for this page to see whether slowdowns are affecting most recruiters or a smaller slow-tail group.

Positions Home Page P90 Latency

Definition: Tracks the daily 90th-percentile load time for the Positions Home page. Tracking the slow tail here, rather than the median, helps catch cases where recruiters on accounts with large position lists or heavier data loads experience meaningfully worse load times than the typical user — even when the overall median for this page looks healthy.

How It Is Calculated: The 90th percentile of load latency for Positions Home page events, trended by day, using the same base filters as the P75 trend above.

Note:

• Trended daily and filtered by event date, via the Date Range selector.

Pipeline Applicant Tab P75 Latency

Definition: Tracks the daily 75th-percentile load time for the Applicants tab of the candidate pipeline view — the tab recruiters use to review candidates who formally applied to a position. Because applicant pools can run large for popular postings, this metric flags when even a moderately loaded applicant list is starting to feel slow for a meaningful share of recruiters, before it becomes a widespread problem.

How It Is Calculated: The 75th percentile of load latency for events resolving to the Applicants tab, trended by day, using the same base filters as the other page-level metrics.

Note:

• Page classification for this tab relies in part on matching a pattern in the page's referring URL. If the underlying page's URL structure changes upstream, this classification could stop matching correctly — worth a periodic spot-check if this metric's volume unexpectedly drops.
• Trended daily and filtered by event date, via the Date Range selector.

Pipeline Applicant Tab P90 Latency

Definition: Tracks the daily 90th-percentile load time for the Applicants tab. This is the slow-tail signal for the applicant-review workflow specifically — a rising P90 here, even with a stable P75, points to a subset of applicant lists (likely the largest pools, from high-volume postings) that are degrading disproportionately.

How It Is Calculated: The 90th percentile of load latency for events resolving to the Applicants tab, trended by day, using the same base filters and page-classification logic as the P75 trend above.

Note:

• Same referrer-based page classification caveat as the P75 version of this metric applies here.
• Trended daily and filtered by event date, via the Date Range selector.

Pipeline Lead Tab P75 Latency

Definition: Tracks the daily 75th-percentile load time for the Leads tab of the candidate pipeline view — the tab recruiters use to review sourced or passive leads before they've formally applied. Tracked separately from the Applicants tab because lead lists and applicant lists can differ substantially in size and data shape, so a slowdown in one doesn't necessarily mean the other is affected.

How It Is Calculated: The 75th percentile of load latency for events resolving to the Leads tab, trended by day, using the same base filters as the other page-level metrics.

Note:

• Page classification for this tab also relies on matching a pattern in the page's referring URL — the same caveat as the Applicant Tab metrics applies.
• Trended daily and filtered by event date, via the Date Range selector.

Pipeline Lead Tab P90 Latency

Definition: Tracks the daily 90th-percentile load time for the Leads tab — the slow-tail counterpart to the P75 Leads Tab metric, isolating whether a subset of lead-heavy pipelines is experiencing materially worse load times than the typical lead-review session.

How It Is Calculated: The 90th percentile of load latency for events resolving to the Leads tab, trended by day, using the same base filters and page-classification logic as the P75 trend above.

Note:

• Same referrer-based page classification caveat as the P75 version of this metric applies here.
• Trended daily and filtered by event date, via the Date Range selector.

Full Profile P75 Latency

Definition: Tracks the daily 75th-percentile load time for a candidate's Full Profile page — the complete profile view, distinct from the lighter "minimal profile" panel shown inline within the pipeline view. Full profile loads tend to pull more data (work history, insights, matched positions, etc.), so this metric tracks whether that heavier page is holding up under normal load.

How It Is Calculated: The 75th percentile of load latency for events resolving to the Full Profile page, trended by day, using the same base filters as the other page-level metrics.

Note:

• Trended daily and filtered by event date, via the Date Range selector.

Full Profile P90 Latency

Definition: Tracks the daily 90th-percentile load time for a candidate's Full Profile page — the slow-tail view of the heaviest, most data-rich page tracked on this dashboard, useful for spotting when profiles with unusually large activity histories or match data start pulling load times into problematic territory.

How It Is Calculated: The 90th percentile of load latency for events resolving to the Full Profile page, trended by day, using the same base filters as the P75 trend above.

Note:

• Trended daily and filtered by event date, via the Date Range selector.

Talent Acquisition

Definition: This section shows how long key recruiter pages take to load and become usable, called time to interactive. It has Request Count and Time to Interactive cards, and a day-wise Talent Acquisition Time to Interactive Latency (seconds) chart. Below them, one row of cards per page shows the page name, request count, P50 Latency, and P90 Latency. The pages are Positions Home, the page recruiters land on to browse open positions, and Applicants Base and Applicants Full. Applicants Base measures the time for the Applicants tab to become usable. Applicants Full measures the same time once all applicant fields have fully loaded.

Note: Filtered by the Date Range selector.

Talent Acquisition Endpoints

Definition: This section shows how quickly the backend APIs behind Talent Acquisition pages respond. It covers the Positions Table API (position search and listing), the Applicants Data API (pipeline applicant data), and the Profile Fields API (candidate profile fields shown in the pipeline). Each API has a row of four cards: API Endpoint, Request Count, P50 Latency, and P90 Latency. The TA API P50 Latency by Endpoint (Seconds) chart trends the median response time of each API by day.

How It Is Calculated: The 50th and 90th percentiles of response time for each API, counting only successful requests that finish in under 30 seconds. Requests from bots and from development or staging sites are excluded.

Note: Use this section with the Talent Acquisition section to see whether a slow page comes from a slow API.

PCSX Endpoints

Definition: This section shows how quickly the backend APIs behind your candidate-facing career site respond. It covers the Search API (job search) and the Position Details API (job posting detail pages). Each API has a row of four cards: API Endpoint, Request Count, P50 Latency, and P90 Latency. The PCSX API P50 Latency by Endpoint (Seconds) chart trends the median response time of each API by day.

How It Is Calculated: Same calculation and filters as Talent Acquisition Endpoints.

Note: Shows only when your organization uses Career Site Experience IX.

Talent Management

Definition: This section shows how long Career Hub pages take to load and become usable. It has Request Count and Time to Interactive cards, and a day-wise Talent Management (Career Hub) Time to Interactive Latency (seconds) chart.

Talent Management Endpoints

Definition: This section shows how quickly the backend APIs behind Career Hub respond, for position search, profile details, and the home feed. Each API, such as the Position Search API, has a row of four cards: API Endpoint, Request Count, P50 Latency, and P90 Latency.

How It Is Calculated: Same calculation and filters as Talent Acquisition Endpoints.

Note: Shows only when your organization uses Talent Management.

Analytics ETL Completion Rate by Table

Definition: Shows the daily completion rate of Eightfold's analytics ETL pipeline for each customer-facing table — letting the data team spot a specific table silently under-loading records on a given day, which would otherwise show up as quietly wrong numbers on every downstream reporting dashboard.

How It Is Calculated: Completion rate, precomputed upstream as a percentage, plotted per ingestion date with one line per table.

Note:

• This chart reflects a fixed 14-day window built into its underlying dataset and does not respond to the dashboard's Date Range filter above — changing the date range will not change what this chart shows.
• No additional filters are applied at the widget level.

Analytics ETL Data Freshness (P90) by Table

Definition: Shows the daily trend of P90 data-freshness latency, in minutes, for the analytics ETL pipeline, per table — measuring how long it takes, at the 90th percentile, for a data change to land in the analytics tables that power dashboards like this one. Rising values mean the reporting data elsewhere is stale by roughly that many minutes for that specific table.

How It Is Calculated: P90 data-freshness latency in minutes, precomputed upstream, plotted per ingestion date with one line per table.

Note:

• Like the Completion Rate chart above, this reflects a fixed 14-day window and does not respond to the dashboard's Date Range filter.
• Read alongside the Completion Rate chart to tell apart "data didn't fully load" from "data loaded, but later than expected."