Weekly shortlist of past Lever candidates for every open role

Every Monday at 8am, go through every live job posting in Lever and rebuild a shortlist of strong past candidates who are already sitting in our database but were archived without being hired, so we stop paying to source people we have already talked to. This runs unattended on a cron schedule and pushes its results out to us; nobody opens a screen. It uses Lever, Slack, and Google Sheets.

Load the guardrails first, before touching any candidate. Call Lever List Archive Reasons and build the set of reasons that disqualify someone from being resurfaced. Note that in Lever, Hired is itself an archive reason, so people we hired sit inside the archived list and have to be excluded explicitly. Alongside Hired, rule out anything meaning do not hire, withdrew, offer declined, failed reference, or asked not to be contacted. Keep that excluded list configurable rather than hardcoded, seeded from the reasons the account actually has. Then read the rediscovery log in Google Sheets with Get Values. It holds one row per past suggestion with the candidate, the posting, and the date. Anyone suggested in the last 90 days is off limits for this run.

Get the roles with Lever List Postings, filtered to published, live postings. For each posting keep the title, team, location, commitment, and the requirements written in the posting content, since that text is what you judge fit against.

Build the candidate pool with Lever List Opportunities, with the archived filter set to true and a time window covering the last 24 months. Page with the cursor: pass the next value as the offset parameter on the following request and keep going until hasNext is false. Push filters down to the API wherever it supports them instead of pulling everything and filtering in memory. Expand contact, applications, and stageChanges so each record arrives with the person's name, email, the postings they applied to, and their stage history.

For each plausible candidate, pull the evidence before judging anyone. Use Lever List Applications for an Opportunity for what they originally applied to, Lever List Feedback on an Opportunity for what the interview panel actually wrote and scored, and Lever List Notes on an Opportunity for recruiter context. Judge genuine fit against the posting requirements from that evidence. This is the entire point of the workflow, so do not keyword match on job titles or resume text. Weight people who reached onsite or offer above people who dropped at phone screen, and read the furthest stage they ever reached from their stage history rather than their current stage, since an archived candidate simply sits in whatever stage they were in at the moment they were archived. Lever List Stages gives you the account's pipeline order to rank stages against.

Post one Slack message per posting to the recruiting channel with Send a Message. Each message names the role and lists at most five rediscovered candidates, each with a one line reason grounded in something specific, such as a requirement they clearly meet or something the panel said about them, plus the furthest stage they previously reached. Keep the shortlist short and high confidence: a recruiter should be able to action all five before lunch. If a posting has no genuinely good match, say so plainly in the message instead of padding the list out to five. Fewer names, or none at all, is the correct output when that is the truth.

For every candidate you suggest, write the reasoning back where recruiters will find it. Use Lever Create Note on an Opportunity to put the same reason on their profile, naming the new posting and the date, so whoever opens that profile understands why they resurfaced. Use Lever Update Opportunity Tags to add a rediscovery tag so recruiters can filter on it inside Lever. Then append one row per suggestion to the tracker with Google Sheets Append Values, recording the candidate, the posting, the furthest stage they reached, the reason, and the date. That tracker is the same sheet you read at the start of the run, so writing these rows is what mutes these people for the next 90 days. Write the rows even when a recruiter never acts on the suggestion.

A few Lever specifics to get right. Every response wraps its payload in a data field, so read from there rather than assuming the resource sits at the top level. Timestamps are Unix milliseconds rather than ISO strings, and they drive both the 24 month lookback and the 90 day mute, so convert carefully in both directions. The current model is the Opportunity, one per candidate per posting, not the legacy Candidate. Confidential postings and opportunities are excluded unless the credential holds confidential access and the request opts in, which is the right default here. Lever rate limits to roughly 10 requests per second and this sweep touches a lot of records, so page in order, batch the detail fetches, stay under that ceiling, and back off on a 429.