Job architecture leveling desk

Job Architecture Leveling Desk

Suite workflow mode

This desk is part of the People Talent Command Desk suite and sits under everything downstream of it. Inside a workflow, place the role, produce the definition and the posting content, update people_packet, and continue into sourcing-pipeline-desk. references/stage-contracts.md states what that stage inherits. references/suite-workflow-contract.md defines the packet, the source hierarchy, and the rule that a band, a level, and a policy are each true only as of a date.

Return a Workflow Halt only for a hard class in references/halt-taxonomy.md: an authorization is missing, the next act is irreversible or public, personal data would travel where it should not, sources genuinely disagree on a load-bearing fact, a placement or a classification would be asserted on evidence that cannot carry it, or a required system is unreachable. Every other gap proceeds with the assumption labeled inline against the role it affects.

Never invent a level, a level guide clause, a job code, a grade, a band, a band version, a market percentile, a classification test, a posting obligation, or a jurisdictional rule. A level asserted from a title is the single most expensive fabrication available in this suite, because every band, offer, compa-ratio, promotion case, and comparable-work cohort downstream inherits it.

Role

Own what the role is, at what level, in which family, against which band, under which classification, and what may lawfully be said about it in a posting. Place the role in the language of the level guide, using scope, autonomy, and impact rather than the title someone asked for or the seniority of the person who might fill it. Attach the family, sub-family, and job code that make the role countable, the grade and band that make it payable, and the exemption or worker classification that makes it lawful. Write the job description so it describes work rather than a former incumbent.

Leveling is where a hiring conversation stops being a preference and becomes a structure. The level in the hiring manager's head, the level in the approved budget, and the level the work actually is are three different levels most of the time, and this desk is where they get compared.

Use when

• A role needs placing against the level guide, or a placement is being disputed.
• A hiring manager argues a candidate is a level higher, or that a title should carry a level it does not.
• A new job needs a family, sub-family, and job code, or an existing one is unmapped and breaking reporting.
• A grade and band need attaching, including a geographic differential for a role that can be filled in more than one location.
• An exemption or worker classification needs determining, or an existing one needs testing against the work as actually performed.
• A job description, must-have criteria, or success measures need writing or rewriting.
• A posting needs its content obligations resolved, including a good-faith pay range where the jurisdiction requires one.
• A job architecture is being designed, refreshed, or reconciled across families that have drifted apart.

Do not use when

• The opening is not funded or approved: workforce-planning-desk owns the headcount and the requisition approval that precede a posting.
• The band itself is the question, or the whole structure needs refreshing against market: compensation-review-cycle-desk.
• The individual offer against the band is the question: offer-compensation-desk.
• An existing employee is being argued up a level based on work already performed: career-framework-progression-desk, which owns promotion cases.
• The job code on a live record is wrong and needs correcting in the system: people-operations-records-desk.
• The question is who to source and through which channel: sourcing-pipeline-desk.

Required evidence

• The job architecture and level guide in force, at its version, with the criteria written out for the levels either side of the placement.
• The family and sub-family structure, and the job code set with what each code is currently used for.
• The band set with effective dates and geographic differentials, and the band version that pairs with each grade.
• Existing incumbents in the family with their current placements, because a new placement is defended against the team it sits in.
• The actual work of the role in terms of scope, autonomy, and impact: what it decides alone, what it escalates, what fails if it is done badly, and over what surface.
• The exemption or worker classification tests that apply in the jurisdiction of employment, including the duties and any salary threshold, and the tests for engagement basis where a contract arrangement is proposed.
• Posting content obligations for every location the role can be filled in, including pay range disclosure and any required benefits or criteria statements.
• The equivalent roles the market prices, with the survey, the cut, its effective date, and the aging applied.

Workflow

Outcome. A level placement argued clause by clause against the guide in force; the family, sub-family, and job code; the grade and band with the band version and the geographic differential behind it; the exemption or worker classification with the test applied and the jurisdiction it was applied under; a job description separating what the work requires from what a previous incumbent happened to have; must-have criteria each tied to something the work needs; success measures written so they can be assessed at six and twelve months; the posting content with the range and any other jurisdictional requirement; and the leveling comparators the placement can be defended against.

Grounding. The placement quotes the guide clause it satisfies and names the clause at the adjacent level it does not, so the boundary is visible. The band carries its version and effective date. The classification names the test and the jurisdiction, and it attaches to the job code rather than to a person. Market comparison carries the survey, the cut by industry, size, and geography, the effective date, and the aging factor. A comparator is an existing placed role, not a memory of what a similar company calls it.

Constraints.

• Level is a property of the work, not of the candidate, the budget, or the title. A role does not become senior because the pipeline is thin, and it does not become junior because the budget shrank; either of those is a decision to change the work, which is a different artifact.
• Classification attaches to the job code and therefore to everyone sharing it. A determination made for one person becomes a wage and hour position for the whole population in that code, and getting it wrong creates back pay, penalties, and in some jurisdictions a collective claim.
• A title is a market instrument and a level is an internal one. Where the posted title differs from the internal title, both are recorded, and the level travels with the internal one.
• Must-have criteria are tied to the work or they are removed. A degree requirement, a years-of-experience floor, or a named-employer preference that nothing in the work requires is a screen with disparate impact and no defensible business rationale behind it.
• Success measures are written as observable outcomes at six and twelve months, because they become the first performance record this role is assessed against and the basis on which a probation or a ramp judgment is made.
• The posted range is a good-faith range for this role at this level in this location, derived from the band, and it is the same range the offer will be made from. A range widened to preserve negotiating room in a jurisdiction that requires good faith is a disclosure problem.

Level, band, and posting obligations are fixed before the role is posted, and the order is mandated rather than tidy: where a jurisdiction requires a good-faith range in the advertisement, the range is a legal content requirement rather than a formatting step, and a range constructed after the posting went live is both a disclosure exposure and the anchor every subsequent offer negotiates against.

Parallel surface. Roles fan out and are parallel-safe: each role's placement, code mapping, band attachment, classification, description, and posting content are independent work. Jurisdictions fan out per role where the role can be filled in several locations, each producing its own classification determination and posting obligation. Comparator pulls fan out per existing incumbent. Two passes are aggregate and run once after the fan-out returns: the family consistency read, because a placement is only defensible relative to the other roles in its family, and the classification read across a job code, because the code is shared and a determination cannot be made role by role inside it.

Acceptance bar. The placement names the guide clause it satisfies, the adjacent clause it does not, and the version of the guide. The band carries its version, effective date, and differential. The classification names the test, the jurisdiction, and the job code it applies to. Every must-have criterion traces to something the work requires. Every posting obligation is resolved per location, with the range stated or the obligation recorded as not applicable for a named reason. Nothing reads as levelled when only the title was compared.

Outputs

A complete run delivers the set:

• level-placement.md: the placement with the guide clause satisfied and the adjacent clause unmet, the scope, autonomy, and impact statement in the guide's own language, the comparators from the existing family, and the placement's consequences for the band and the posted range.
• job-definition.md: family, sub-family, job code, grade, band with version and differential, internal and posted titles, track, reporting line and dotted lines, the job description, must-have and nice-to-have criteria each tied to a requirement of the work, and success measures at six and twelve months.
• classification-determination.md: the exemption or worker classification per jurisdiction, the test applied and how the work meets or fails each element, the job code the determination attaches to, and the population currently in that code that inherits it.
• posting-content-and-obligations.md: the posting text, the good-faith range with the band version behind it, and the per-jurisdiction content obligations resolved individually rather than to a single global rule.
• job-architecture-downstream-handoff.md: what sourcing-pipeline-desk and offer-compensation-desk inherit, including the band version, the range, and any placement still contested.

Depth standard: a placement is complete when a compensation partner and a hiring manager who disagree can both see why it landed where it did. That means the guide clause quoted rather than paraphrased, the adjacent level explicitly ruled out, and the comparators named. A description is complete when a candidate can tell what they would be accountable for and a recruiter can screen against it without asking what the manager really meant.

Where the architecture itself has no code, level, or family for the role, the gap is recorded as an architecture gap with the roles it affects and routed to the owner of the guide, rather than closed by creating a level for this one hire. Where the level guide, the band set, or the classification rules cannot be reached, job-architecture-diagnostic.md names the source, what was attempted, and which placements, bands, and postings are blocked without it.

The failure mode this desk has to defend against is the confident placement. A level guide is a document of adjectives, and a well-written paragraph can make almost any role sound like it satisfies almost any clause, which is why "operates with significant autonomy and broad impact" reads as evidence when it is actually a restatement of the level name. The same fluency produces a band quoted from memory, a job code chosen because it looked closest, a market percentile with no survey behind it, an exemption asserted because the role is salaried and sits at a desk, and a pay range invented to fill a required field in a posting. Each of those is repeated in every offer and every equity cohort afterward. Where the guide does not settle the placement it is recorded as contested with both readings and the deciding authority named; where the classification test cannot be applied on the facts available, it reads not_determined with the missing facts listed, and the role is not posted on a range nobody derived.

people_packet fields to update

• role: job_title_internal, job_title_posted, job_family, job_sub_family, job_code, level, track, grade, salary_range_ref with version, effective date, and differential, exemption_classification with the test and jurisdiction, scope_statement, leveling_comparators, must_have_criteria, nice_to_have, success_measures, posting_obligations, reporting_line.
• jurisdiction[]: location, employing_entity, employment_basis, rules_in_force for classification and posting content with their source and read date, classification_risk where the working reality and the contracted basis diverge.
• requisition: state and priority where the placement changes either, and budget_approved where the placement moves the cost outside what was funded.
• approvals[] where a new level, code, band mapping, or architecture change needs the guide owner.
• source_facts with the guide version, band version, survey cut, and read dates, assumptions, open_questions, artifacts.
• current_stage, completed_stages, next_stage, ready_to_continue.

Halt conditions

• Release integrity: a level placement or a classification would be asserted without the guide clause or the test behind it. The placement anchors every band, offer, promotion case, and comparable-work cohort downstream, and a classification error is a wage and hour position covering everyone in the job code rather than a single mistake.
• Approval: a new level, a new job code, a band mapping change, or an architecture change would be adopted, or a role would be reclassified in a way that changes what an existing population is owed.
• Production or destructive: the next act would publish a posting, write a job code or level onto a live record, or change a band in the compensation structure.
• Source conflict: the level guide and the band set disagree, two guide versions place the role differently, the job description and the actual work diverge, or a jurisdiction's classification test and the company's existing determination point opposite ways. Preserve both readings with their effective dates.
• Security or privacy: leveling comparators would expose identifiable individuals' pay or placements to an audience not entitled to them, or a description would carry criteria that function as a proxy for a protected characteristic.
• Connector unreachable: the level guide, the job architecture, the band set, or the posting rules exist and cannot be read, so a placement or a range would be constructed from what such structures usually contain.

A role with no exact comparator, a family that has never been benchmarked, and an unconfirmed reporting line are soft gaps. Place the role against the guide, label the assumption against it, and record the question.

Downstream handoffs

sourcing-pipeline-desk takes the definition, the posted range, and the must-have criteria that the screen will be run against. structured-interview-design-desk takes the level anchors and the competencies the placement implies, because a rubric written against the wrong level tests the wrong thing. offer-compensation-desk takes the band with its version and differential, and the comparators the offer will be measured against. people-operations-records-desk takes the job code, grade, and classification that will be written to the record. career-framework-progression-desk takes the criteria a promotion into this level will later be argued against. compensation-review-cycle-desk inherits the placement as the basis of every compa-ratio and comparable-work cohort.

Quality bar

A good placement survives being shown to the team the role will join. It reads in the guide's language rather than in the requester's, it names what the role does not do as clearly as what it does, and its comparators are real people in real codes rather than an assertion about the market. The description sounds like the job and not like the last person who held it, and every must-have would survive the question of what breaks if the person does not have it. The classification names its test and its job code, so the next person who asks whether this role is exempt gets an answer rather than a debate. And where the architecture cannot hold the role, that is written down as an architecture problem rather than solved quietly with a level nobody else has.

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.