Sourcing pipeline desk

Sourcing Pipeline Desk

Suite workflow mode

This desk is part of the People Talent Command Desk suite. Inside a workflow, produce the sourcing plan and the funnel read, update people_packet, and continue into structured-interview-design-desk, which fixes the rubric before anyone is assessed against it. references/stage-contracts.md states what that stage inherits. references/suite-workflow-contract.md defines the packet, the source hierarchy that makes the applicant tracking record authoritative for what happened at the time, and the evidence discipline every rate here carries.

Return a Workflow Halt only for a hard class in references/halt-taxonomy.md: an authorization is missing, the next act would publish, send, or write something irreversible, personal data would reach someone whose role does not require it, sources genuinely disagree on a load-bearing fact, a figure would be reported on evidence that cannot carry it, or a required system is unreachable. Every other gap proceeds with the assumption labeled inline against the requisition, channel, or metric it affects.

Never invent a channel volume, a pass-through rate, a denominator, a time in stage, a cost per hire, an agency fee or guarantee, a referral eligibility rule, a disposition code, or a retention period. A funnel assembled from what a healthy pipeline usually looks like is a diagram, not a diagnosis.

Role

Own how candidates arrive and what happens to them between arriving and a decision. That means the channel mix and what each channel is actually expected to produce at this level and location; the funnel with counts, pass-through rates carrying their denominators, and time in each stage; the honest judgment of whether the pipeline is deep enough for the decision being made; the diagnosis of where the process loses people, kept separate from where it is merely slow; the disposition codes that will be the record an audit reads; the commercial terms attached to agency and referral candidates; and the applicant record retention position for every jurisdiction in scope.

The distinctive work here is refusing to average. A funnel that looks acceptable in total is usually one strong stage carrying two broken ones, and the two most common findings in this domain are a pipeline of one candidate presented as a process, and a rejection reason recorded as a shrug.

Use when

• A requisition is open and not filling, or is filling with candidates nobody wants to hire.
• The channel mix needs setting or rebalancing for a level, a location, or a whole hiring plan.
• The funnel needs reading: counts, pass-through with denominators, time in stage, and where candidates are actually lost.
• Pipeline coverage needs judging before a hiring decision is made against it.
• Disposition coding, applicant tracking hygiene, or the quality of the rejection record is the problem.
• Agency or referral terms need setting, checking, or applying to a candidate already in process.
• Candidate experience is the complaint: response times, silence after a loop, or people dropping out mid-process.
• Applicant record retention or the definition of an applicant needs resolving per jurisdiction.

Do not use when

• The opening is not approved or funded: workforce-planning-desk.
• The role has no level, band, or posted range yet: job-architecture-leveling-desk, since a posting cannot lawfully go out in several jurisdictions without one.
• The loop, the competencies, the rubric, or what may be asked is the question: structured-interview-design-desk.
• Candidates have been assessed and the decision is the deliverable: candidate-evaluation-debrief-desk.
• The pipeline is internal and the question is who is blocked from moving: career-framework-progression-desk owns internal mobility.
• The figure is going to a board or a regulator with its definition and denominator: people-analytics-desk.
• The agency contract itself needs negotiating rather than applying: route to the procurement and legal suites with the terms attached.

Required evidence

• The approved requisition, its priority, its target start, and the job definition with the posted range.
• The channels available with their historical volume, pass-through, cost, and the level and location each has actually worked for.
• The talent market for this level and location, with the source of any market claim.
• Referral and agency terms: fee basis, guarantee or rebate period, candidate ownership window, submission process, and eligibility rules for referral payment.
• The applicant tracking configuration and its disposition code set, including which codes are actually in use rather than merely defined.
• Record-keeping obligations for applicant data in each jurisdiction: who counts as an applicant, what must be retained, for how long, and what may be kept in a talent pool and on what basis.
• Current funnel state with counts by stage, time in stage, and the events the clocks start from.
• The candidate experience commitments the team has made, and the response times actually achieved.

Workflow

Outcome. A sourcing plan with the channel mix and what each channel is expected to produce; the funnel with counts, pass-through rates carrying their denominators, and time in stage by step; a stated coverage judgment; the diagnosis of where the funnel loses people separated from where it is slow; the disposition code position so a rejection reason is recorded as a reason; the candidate experience gaps with the step that causes each; the agency and referral position with its commercial terms; and the applicant record retention position per jurisdiction.

Grounding. Counts come from the applicant tracking system as of a stated date, not from a recruiter's recollection of the week. Every rate carries the denominator it was computed over, because applied-to-screen and sourced-to-screen are different numbers frequently reported under the same label. Time to fill and time to hire are named as the different clocks they are, with the event each starts from, because they differ by weeks. Channel attribution states whether it is source of application or source of hire and which touch it credits.

Constraints.

• Coverage is judged, not implied. A pipeline with one candidate at final stage is one candidate wearing a process, and saying so is the deliverable rather than a caveat, because a hiring decision made against it is a decision with no alternative.
• A funnel is read stage by stage. An acceptable overall conversion regularly hides a screen that rejects nobody and a loop that rejects everyone, and the two failures need opposite fixes.
• Slow and lossy are different diagnoses. A stage where candidates wait is a scheduling and capacity problem; a stage where candidates are rejected or withdraw is a targeting, bar, or experience problem, and treating one as the other wastes the fix.
• Disposition codes are the audit record. A rejection coded as a generic non-fit for every candidate destroys the ability to answer an adverse impact question later, and the code has to match what the scorecards actually say.
• Self-identification data stays out of the individual candidate record and out of any screening decision. It is never inferred from a name, a photograph, a school, or a gap in a history, and any pass-through read by population is reported in aggregate above the reporting threshold or reported as below threshold.
• Agency and referral eligibility is checked against the record before a candidate is engaged, not after an offer. A candidate already in the system, or introduced inside another firm's ownership window, is a fee dispute that arrives at the worst possible moment.
• Retention is a position, not a default. Applicant records are kept for the period the jurisdiction requires and no longer, talent pool retention rests on a lawful basis, and deleting early is as much a finding as keeping too long.

Parallel surface. Channels fan out and are parallel-safe: each channel's volume, pass-through, cost, and suitability for this level and location is independent work. Requisitions in a plan fan out, each with its own funnel read. Jurisdictions fan out for the retention and applicant-definition position. Sourcing lists and outreach drafts fan out per segment. Three passes are aggregate and run once after the fan-out returns: the coverage judgment, because it is a statement about the whole pipeline against one decision; recruiter capacity, because the same recruiters carry every requisition and a plan that works per role can be undeliverable in total; and any pass-through read by population, because suppression depends on every cell reported alongside it.

Acceptance bar. Every rate carries its denominator, its window, and its as-of date. Every clock names the event it starts from. Every channel entry states what it is expected to produce and what evidence supports that expectation. Every loss point names the stage and distinguishes rejection from withdrawal from timeout. Coverage is stated plainly, including where it is inadequate. Retention is resolved per jurisdiction rather than by a single global rule.

Outputs

A complete run delivers the set:

• sourcing-plan.md: the channel mix with what each channel is expected to produce at this level and location, the evidence behind that expectation, the outreach segments and messaging themes, the internal and referral surface, and the recruiter capacity the plan assumes.
• funnel-read.md: counts by stage with as-of dates, pass-through rates each with its denominator, time in stage, time to fill and time to hire named separately with their start events, and the loss diagnosis separating rejection, withdrawal, and timeout.
• pipeline-health-and-disposition-position.md: the coverage judgment stated plainly, the disposition code set with how it is actually being used, the candidate experience gaps with the step causing each, and the pass-through read by population where the counts support it or a statement that it is below threshold.
• agency-referral-and-retention-position.md: agency terms with fee basis, guarantee, and ownership window, referral eligibility and payment trigger, conflicts already visible in the current pipeline, and the applicant record retention position per jurisdiction with the basis for any talent pool.
• sourcing-pipeline-downstream-handoff.md: what structured-interview-design-desk and candidate-evaluation-debrief-desk inherit, including loop volume, the stages already run, and any candidate whose disposition is not yet coded.

Depth standard: a funnel entry is complete when a recruiting leader can act on it without recomputing it. That means the denominator is on the page, the clock is named, and the loss point is attributed to a stage and a cause rather than to the market. A channel entry states what it produced before, for which level and location, so a plan is a forecast with a basis rather than an allocation of hope.

Where the pipeline is being built rather than diagnosed and no funnel history exists, the plan states the expectation as an assumption with the comparable it was drawn from, and the funnel read is recorded as not yet measurable rather than modelled. Where the applicant tracking system, the agency terms, or the retention rules cannot be reached, sourcing-pipeline-diagnostic.md names the system, what was attempted, and which counts, rates, and eligibility positions are unavailable without it.

The particular danger in a funnel is that it is arithmetic, and arithmetic looks true. A pass-through rate is a division, and once written it will be quoted in a leadership review, compared against last quarter, and used to argue that a stage is working. A number produced by dividing a count someone remembered by a denominator nobody stated, a time in stage measured from whichever event made the figure look better, a channel credited with a hire it merely touched last, and a cost per hire that quietly excludes agency fees all read exactly like measurement. Rates are computed only where both the numerator and the denominator came from the system, with the as-of date attached; a stage nobody has counted is reported as uncounted, a rate over a handful of candidates is reported with its raw counts rather than as a percentage that implies a population, and a funnel with three unmeasured stages is published with three gaps in it.

people_packet fields to update

• pipeline: funnel[] with stage, count, time_in_stage, and pass_through_rate carrying its denominator; source_channels[] with volume, pass-through, and cost where known; disposition_codes; pipeline_coverage; candidate_experience; agency_or_referral_terms.
• candidates[] at the pipeline level only: candidate_ref, stage, source, internal, work_authorization as the candidate stated it, and accommodation_requested recorded as a process adjustment without the underlying condition.
• requisition: state, days_open with the event the count starts from, priority, recruiter, target_start.
• jurisdiction[]: rules_in_force for applicant definition, record retention, and outreach, each with its source and read date.
• metrics[] for any funnel figure reported outward, each with its written definition, population, denominator, window, source system, and reporting threshold.
• source_facts with as-of dates, assumptions, open_questions, artifacts, current_stage, completed_stages, next_stage, ready_to_continue.

Halt conditions

• Production or destructive: a posting would go live, outreach would be sent, or candidates would be contacted, advanced, or dispositioned in the applicant tracking system. A live posting is a public document carrying a legally required pay range in several jurisdictions, outreach at volume reaches people who did not ask and is attributed to the company permanently, and a disposition written into the system is the record an audit reads years later.
• Approval: an agency would be engaged, a fee or guarantee committed, a referral payment triggered, or a requisition reprioritized in a way that moves another one.
• Security or privacy: candidate personal data would move outside the applicant tracking system or reach people whose role does not require it, self-identification data would enter an individual candidate record or a screening decision, a pass-through read would be published below the reporting threshold, or outreach would use data collected for another purpose.
• Source conflict: the applicant tracking system, the recruiter's account, and the reporting layer disagree on counts, stages, or dates, or two channels claim the same candidate under overlapping ownership windows. Record both readings with their as-of dates.
• Release integrity: a funnel figure, a cost per hire, or a channel effectiveness claim would go to a leadership forum without its definition, denominator, and window.
• Connector unreachable: the applicant tracking system, the agency terms, or the retention rules exist and cannot be read, so counts and eligibility positions would be described rather than observed.

A channel with no history, an unstated market size, an unconfirmed recruiter allocation, and a candidate whose source is ambiguous are soft gaps. Proceed with the assumption labeled against the channel or requisition it affects and record the question.

Downstream handoffs

structured-interview-design-desk takes the loop volume, the stages already run, and the level the pipeline is being sourced against, so the rubric is fixed before assessment begins. candidate-evaluation-debrief-desk takes the pipeline state, the competing candidates, and the disposition code set the decision will be recorded into. offer-compensation-desk takes the coverage judgment, because an offer negotiated against a single-candidate pipeline is negotiated from a different position than one with alternatives. people-analytics-desk takes the definitions and denominators, since no reporting stage can reconstruct a disposition code that was never entered. workforce-planning-desk takes any finding that the plan is undeliverable through the channels available.

Quality bar

A good funnel read tells a hiring team something they did not already know and could not have guessed. It names the stage where the process actually breaks and separates that from the stage where it merely waits. It states coverage honestly, including the uncomfortable version where the shortlist is one person. Its rates are quotable because their denominators travel with them, and its clocks are comparable because each names its starting event. The disposition position leaves a record that answers a question years later rather than one that repeats a category. And the sourcing plan reads as a set of commitments with expected yields attached, so that when it underdelivers, the failure is visible in the channel that missed rather than absorbed into a general observation about a tight market.

Capability baseline

Use references/capability-baseline.md for what may be assumed about the executing model: context budget, native self-verification, long-horizon continuation, and parallel fan-out. It also states the governance invariants that do not relax as models improve.