The engineering IC promotions decision record: why the promotions model you chose determines your career growth ceiling and your calibration consistency surface

A 19-person developer tools startup had no formal promotion criteria. The founding CTO had promoted two engineers during the company's first year by instinct: both had been doing work that was clearly beyond their current title, both had asked for a promotion conversation, and the CTO had agreed in both cases based on three or four months of close observation. At fifteen engineers the process was still informal — a manager would mention to the CTO that someone seemed ready, the CTO would have a thirty-minute conversation with the engineer, and a decision would follow within the week. The process felt personal, fast, and fair to the people inside it.

By the time the company reached nineteen engineers, three senior engineers had been on the team for twelve to eighteen months and were waiting for conversations about promotion to staff. None of them had received a clear signal about what "staff" would require. One of them had asked her manager directly, and the manager had said "keep doing what you're doing and make sure the CTO sees it." The other two were waiting without actively asking, unsure whether asking would seem presumptuous. Meanwhile, the one engineer who did receive a staff promotion that quarter had been working on a public API launch that received eight hundred GitHub stars and generated significant Slack discussion among the founders. The CTO mentioned the engineer in a board update. The promotion followed naturally from that visibility, and it was genuine — the engineer had done excellent work. But the criteria for the promotion, to the extent they could be reconstructed, were: this person's work was highly visible to the organization's leadership in the most recent quarter.

Six months after the promotion, the newly-minted staff engineer was continuing to do work at the senior engineer scope — executing well-defined tasks, shipping features in her domain, producing high-quality code. She was working harder than ever. But no one had defined what staff-level work looked like at this company: which cross-team technical decisions she was now accountable for influencing, what it meant to set technical direction rather than execute against it, how a staff engineer should navigate the gap between engineering and product strategy. The promotion document, to the extent one existed, said "congratulations, promoted to Staff Engineer, effective Q3." The engineer's manager had no coaching model for the new level because no description of the level had been written. Two of the three senior engineers who had been waiting for promotion conversations accepted external offers from companies that had made them explicit staff-level offers with written role descriptions during the same period. One of them, in her exit interview, said she had believed her performance was strong but had no way to know whether the company agreed, because there was no criteria document against which to self-assess. The third remained on the team and continued waiting.

The failure was not a failure of the CTO's judgment about the promoted engineer. It was a failure of the promotion process to separate the criteria question — what behaviors and outcomes constitute staff-level performance at this company — from the visibility question — whose work was most salient to leadership in a given quarter. A promotion process without documented criteria makes visibility the de facto standard, because visibility is the only observable variable available when no explicit standard exists. The engineers who departed were not necessarily stronger than the one who was promoted; they may have been weaker. But neither they nor the organization had any mechanism to determine the answer to that question, and in the absence of a mechanism, the default outcome is that engineers who have external alternatives use them rather than wait.

A 45-person B2B SaaS had a detailed career ladder document specifying the expected behaviors and outcomes for promotion from mid-level to senior and from senior to staff. The ladder had been written by the VP of Engineering during the company's Series A year and updated once since then. It described each level along five dimensions — technical scope, technical depth, collaboration and communication, product and business impact, and leadership and mentorship — with two-to-three-sentence descriptions per dimension per level. By the standards of most Series A engineering organizations, the career ladder was unusually explicit. The VP was proud of it and had shared it in several engineering management forums as an example of what a usable career ladder looked like at this stage.

Eighteen months after the career ladder was finalized, the VP ran a promotion-rate analysis as part of a retention audit triggered by three senior engineer departures in a single quarter. The analysis showed that when she controlled for performance distribution across the three managers who reported to her, the promotion rate varied by 2.8x: Manager A was promoting engineers in approximately fourteen months at the senior-to-staff level; Manager B was promoting engineers in approximately twenty-two months; Manager C was promoting engineers in approximately thirty-nine months. The VP had not seen this pattern before because she had been reviewing individual promotion decisions — was this specific engineer ready? — rather than the promotion rate distribution across managers. No individual decision had seemed wrong; it was the pattern that was wrong.

When she investigated, she found that the three managers had developed materially different interpretations of what the promotion process required. Manager A required an engineer to be demonstrably performing at the next level — delivering outcomes at the expected scope, receiving strong peer feedback, visible to cross-functional partners — and would submit a promotion packet when she believed the engineer had been doing so consistently for two quarters. Manager B submitted promotion packets when an engineer "seemed ready," a judgment she made based on a gut assessment that she could not fully articulate but that roughly tracked to six-to-nine months of strong performance combined with the engineer's expressed interest in growth. Manager C required engineers to write their own promotion self-assessments and would only submit a packet to the VP if he "believed in it" after reading the self-assessment — a bar whose height depended on the quality of the self-assessment as a document, not only on the underlying performance. An engineer on Manager C's team had been performing at the next level by any observable standard for eight months but had never been asked to write a self-assessment and did not know the process required one. She had not been promoted. An engineer on Manager B's team had been promoted after five months with significant on-the-job support still needed at the new level; the support was genuine and Manager B was providing it, but the promotion decision had been made before the engineer had demonstrated the next-level scope rather than after.

The annual calibration session the VP ran reviewed final promotion decisions: were the promoted engineers performing appropriately at their new levels? It consistently found that the answer was "mostly yes," because Manager A's promotions were reliably strong, Manager B's promotions were mostly strong with occasional gaps, and Manager C's promotions were almost all very strong. The calibration session had been designed as a quality gate on promotion outcomes, and it was performing that function accurately. It was not designed to detect variance in the nomination process — how managers decided who to nominate — and it was not detecting it. The three engineers who had departed in the quarter that triggered the audit had all been on Manager C's or Manager B's teams. Their exit interviews cited feeling stuck and uncertain about growth paths. The career ladder they had read described the same standard the VP intended. The promotion process they had experienced was shaped by their specific manager's interpretation of an evidence requirement the ladder document had never addressed. Connect this to the engineering career ladder decision record: a career ladder that defines the level criteria without specifying the evidence model — what types of behavioral observations, project outcomes, and cross-functional signals are required in a promotion packet, and at what volume — produces a criteria document that is consistent in intent and inconsistent in application, because the gap between "performs at the next level" and "has submitted a packet that demonstrates performance at the next level" is filled by each manager's independent interpretation of what constitutes demonstration.

A 72-person enterprise SaaS with fifteen engineering managers ran a quarterly promotion calibration process that the VP of Engineering had built over two years into something she was genuinely proud of. The process had four components: a promotion packet template with seven required sections, submitted by the nominating manager two weeks before the calibration session; a written peer calibration form completed by two engineers on adjacent teams who had direct visibility into the nominee's cross-team work; a calibration session where all fifteen managers reviewed every active nomination together, with each manager expected to speak to their team's performance distribution relative to the nominee's; and a quarterly report showing promotion rates by manager and by level, which the VP reviewed for outliers. The process had been designed specifically to address the calibration variance that the VP had experienced at her prior company, and it was working — the promotion rates were consistent within normal variance bands across all fifteen managers, the calibration sessions were substantive, and post-promotion performance tracking showed strong alignment between the nomination evidence and six-month post-promotion outcomes.

Fourteen months into running this process, an engineer on one of the teams came to the VP with a question. The engineer had been at the company for twenty-one months in a senior role. She had submitted her promotion packet eleven months ago, had been told at the time that she was "close but not quite there," and had been given a list of areas to develop. She had worked on all of them. Her most recent quarterly performance review had been the strongest in her tenure on the team. A peer on an adjacent team had been promoted to staff after eighteen months in a senior role. She wanted to understand what the difference was.

The VP reviewed the engineer's file. The performance was unambiguously strong. The promotion packet from eleven months ago had been borderline at the time, but the subsequent eleven months of development had closed the gap identified in the feedback. The VP asked the engineer's manager directly: why hadn't a new nomination been submitted? The manager's answer was that he had been waiting until the engineer reached her two-year mark in the role, which was coming up in three months. He had always believed that staff promotions required at least two years of demonstrated performance at the senior level — this was how it had been done at his previous company, and it seemed like the right bar to him. He had never stated this requirement to the engineer because he assumed it was standard. He had never mentioned it to the VP because he assumed everyone operated the same way.

The VP's investigation found that two other managers had the same informal policy, each having arrived at it through a different path: one from the same prior-company convention, one from a comment the CEO had made at an all-hands the previous year about "being thoughtful about how quickly we advance people." None of the three managers had asked whether this interpretation was consistent with the company's official criteria. None of the three had documented it in any written form. The three engineers affected by the informal policy were at months nineteen, twenty, and twenty-one in their roles respectively. All three had cleared the official promotion criteria; one had cleared it more than a year ago. The quarterly promotion rate reports showed no anomaly because all three managers' teams had comparable tenure distributions — their promotion rates looked correct relative to their teams' tenure profiles. The inconsistency was invisible in the aggregate and only visible at the individual level, where three engineers were accumulating months against a criterion that had no official existence.

The calibration architecture had been designed to catch the outcome failures: promotions made prematurely or against insufficiently strong evidence. It was not designed to catch the nomination failures: promotions not made because a manager was applying a criterion outside the official documentation. The quarterly report's denominator — engineers who had been nominated — was correct. The numerator — engineers who had cleared the criteria and had not been nominated — had no reporting surface, because the process had no mechanism for a manager to signal that an engineer was criteria-eligible but had not yet been nominated, and no mechanism for the calibration sessions to ask whether all eligible engineers had been nominated rather than only whether all nominated engineers were eligible. Connect this to the engineering performance management decision record: the calibration failure here is the promotions-process version of the performance management calibration failure — just as performance calibration sessions that review only final rating decisions catch the overrating failures and miss the underrating failures, promotion calibration sessions that review only nominations catch the over-nomination failures and miss the under-nomination failures; both failures have the same structural root cause, which is a calibration architecture designed to catch one direction of error and structurally blind to the other.

Structural properties set by the engineering IC promotions decision

Two structural properties are determined by the founding decisions about criteria specificity, evidence requirements, and calibration architecture. Both are invisible when the engineering team is small enough for the head of engineering to hold the complete performance picture, and visible when the team grows past that point and the gap between the official criteria and the manager-interpreted criteria becomes large enough to produce material variance in engineer outcomes.

Property 1: Career growth ceiling. The career growth ceiling is the maximum rate at which engineers can advance through the organization — the ceiling on how quickly demonstrated next-level performance is recognized and rewarded at scale, under the constraint that recognition depends on manager advocacy and calibration architecture rather than on the engineer's performance alone. The ceiling is bounded by three factors. First, criteria clarity: whether the promotion criteria specify observable behaviors and outcomes at each level transition with enough specificity that an engineer and their manager can jointly determine whether the criteria have been met without relying on interpretive latitude. A career ladder that describes "leads cross-team technical work" is a criterion that can be interpreted as requiring one completed cross-team project or five, a project that the engineer initiated or one that they executed after being asked, a project whose scope involved two teams or ten — and each manager will fill the gap with their own interpretation, producing a criteria clarity floor that is bounded by the least specific level description in the ladder rather than by its most specific. The criteria clarity requirement is not that the criteria be exhaustive or mechanical; it is that they be specific enough that a manager reviewing an engineer's performance history can determine whether the criteria have been met without needing to decide what the criteria mean. Second, calibration consistency: whether similar performance produces similar promotion decisions across different managers and teams. The calibration consistency requirement is not that every engineer with identical performance be promoted on an identical timeline; it is that the variance in promotion timing across managers be explained by variance in engineer performance rather than by variance in manager interpretation of the evidence requirements. A 2.8x promotion-rate variance when controlled for performance distribution is a calibration consistency failure, regardless of whether the individual decisions within each manager's set look appropriate when reviewed in isolation. Third, promotion cadence architecture: how frequently the organization formally reviews promotion eligibility. A fixed-cadence model — all promotions are evaluated quarterly, nominations are submitted in the second month of each quarter, decisions are made at the calibration session in the third month — makes the review timeline predictable for engineers. A rolling model — promotions happen when a manager decides the engineer is ready and submits a packet — makes the timeline dependent on manager initiative, which introduces a timing variance that is indistinguishable from criteria variance from the engineer's perspective: an engineer who waits six months longer than a peer for a promotion conversation that a rolling model leaves to manager timing has no mechanism to determine whether the delay reflects a criteria gap or a nomination timing gap.

Property 2: Calibration consistency surface. The calibration consistency surface is the degree to which similar performance produces similar promotion decisions across different managers and teams — the inverse of the promotion-rate variance that remains when performance distribution variance is controlled for. The surface is determined by three factors. First, the evidence requirements model: whether the promotion packet specifies what types of evidence are required — specific behavioral examples from the review period, documented cross-team feedback from named partners, specific project outcomes with measurable results — or whether the evidence requirement is left to manager discretion. A promotion packet template that requires seven sections covering the right topics but leaves the evidence requirement implicit produces manager-to-manager variance in what counts as sufficient evidence for a nomination: Manager A believes two strong behavioral examples per section are sufficient; Manager B believes the examples must be at the same scale as the largest projects the team has shipped this year; Manager C believes the packet must be strong enough to be compelling to the VP rather than just sufficient against the criteria. Each of these is a different evidence bar, and none of them is written down. Second, the calibration architecture: whether calibration sessions review the evidence that produced nominations, not just the final outcomes. An outcome-only calibration can detect when a promoted engineer's performance evidence looks weak relative to other promoted engineers' evidence. It cannot detect when a non-promoted engineer's performance evidence looks strong relative to the official criteria, because non-promoted engineers produce no nomination and no packet, and thus produce nothing for the calibration session to review. The under-nomination failure is structurally invisible in an outcome-only calibration, and it is precisely the failure mode that produces the departure pattern the 45-person company experienced: engineers who are not being nominated when they should be, for criteria-interpretation reasons that the calibration session was designed to catch but is architecturally unable to see. Third, the informal policy surface: the set of unstated criteria that have accumulated outside the official documentation. Informal policies are not always the result of managers acting in bad faith; they are more often the result of managers filling gaps in the official criteria with their own interpretive frameworks, often imported from prior companies or derived from directional statements by leadership that were never intended as policies. The informal policy surface has a natural accumulation dynamic: each time a manager makes a promotion decision that is influenced by an unstated criterion, the criterion is reinforced in that manager's mental model without any corresponding update to the official documentation, and the gap between the official criteria and the manager-interpreted criteria grows wider with each decision cycle.

The engineering IC promotions ADR: five sections

Section 1: Promotion criteria model and evidence requirements. The promotion criteria model specifies what behaviors and outcomes constitute promotion eligibility at each level transition, derived from the career ladder's level definitions and operationalized as observable evidence types. The most consequential decision in this section is whether to specify the evidence model — what types of observations count as evidence for each criterion — alongside the criteria themselves, or to leave the evidence model implicit and rely on manager judgment to determine what constitutes sufficient demonstration. A criteria document without an evidence model produces a criteria floor that is bounded by the least specific criterion in the document: any criterion that can be interpreted as satisfied by a range of evidence types will be interpreted differently by different managers, and the variance in those interpretations will produce calibration inconsistency that is invisible in outcome-review calibration sessions. The evidence requirements should specify, for each major dimension in the criteria: (1) the minimum evidence types required — behavioral examples from the review period, cross-team feedback from named partners, project outcome documentation with scope and impact described — rather than leaving the evidence type to manager judgment; (2) the recency requirement — evidence from the current review period versus evidence from the trailing twelve months — because a promotion decision made on evidence from eighteen months ago, when the engineer's scope was different, is not the same decision as one made on evidence from the current period; and (3) the co-observation requirement — whether the evidence must be observable to people other than the engineer's direct manager, which catches the confirmation-bias failure mode where a manager promotes an engineer based on strong performance they have observed directly without checking whether the performance is visible to and valued by adjacent teams. The evidence requirements model should be co-designed with the career ladder update process: when the career ladder's level descriptions change, the evidence requirements for the affected criteria should be reviewed in the same session, because criteria and evidence requirements that are designed at different times will produce a gap between what the criteria say and what the evidence model allows a manager to demonstrate. Connect this to the engineering feedback culture decision record: the evidence base for a promotion packet is built from the performance period's feedback documentation; a feedback culture that produces specific behavioral observations with named outcomes provides the promotion packet's evidence base at no additional administrative cost; a feedback culture that produces directional summary statements ("great work this quarter") provides no evidence base and forces the manager to reconstruct behavioral evidence retroactively, which produces evidence of lower specificity and lower calibration reliability than evidence documented at the time of observation.

Section 2: Promotion nomination model and advocacy architecture. The promotion nomination model specifies how promotion eligibility is identified — who initiates the promotion conversation, under what conditions, and with what process for an engineer to understand their own status. The most consequential decision in this section is whether to design the nomination model as manager-initiated or criteria-triggered. A purely manager-initiated model — the manager decides when to submit a packet — produces timing variance that depends on manager initiative, manager confidence in the engineer's readiness, and manager calculation about whether the current quarter is the right time to make the case. A criteria-triggered model — any engineer who has been at the current level for a minimum period and whose most recent performance review meets a specified threshold is automatically flagged for promotion eligibility review — reduces timing variance by making the trigger a function of documented performance rather than manager initiative. The criteria-triggered model requires more administrative infrastructure than the manager-initiated model, but it produces a calibration consistency benefit that is worth the cost: engineers whose managers have not submitted a packet are visible in the system as criteria-eligible-but-not-nominated, which is the population the outcome-only calibration architecture cannot see. The nomination advocacy architecture should also specify the engineer's visibility into their own promotion status: whether the engineer has access to the promotion criteria documentation (they should), whether the manager is required to have an explicit conversation about promotion timeline at least annually with any engineer who has been at the current level for twelve months or more (they should), and whether the engineer can self-nominate for a promotion eligibility review if they believe they have met the criteria (they should, with a lightweight review process that routes to the manager and the manager's manager if the manager does not submit a packet within thirty days of the engineer's self-assessment). The self-nomination mechanism does not bypass the manager's authority to nominate; it creates a forcing function for the manager to articulate in writing why the criteria have not been met, which produces documentation that prevents the informal-policy failure mode the 72-person company experienced: a manager who is operating an undocumented two-year minimum cannot respond to a self-nomination by saying "wait three more months" without producing a written explanation that makes the undocumented criterion visible to the manager's manager and to the calibration session. Connect this to the engineering technical interview design decision record: the interview loop and the promotions process are jointly responsible for the level distribution of the engineering organization; the criteria the interview loop uses to evaluate "would this candidate perform at the level we're hiring them at?" and the criteria the promotions process uses to evaluate "has this engineer demonstrated the level we're considering advancing them to?" should be derived from the same career ladder level definitions and described in consistent language; an organization whose interview rubric describes one model of staff-level performance and whose promotion criteria describe a different model will hire engineers who meet one definition and promote engineers who meet the other, producing a level distribution whose internal consistency cannot be evaluated because the two models are not comparable.

Section 3: Calibration architecture and cross-manager evidence review. The calibration architecture specifies how promotion decisions are reviewed across managers to ensure that similar performance produces similar decisions. The most consequential decision in this section is whether to review the evidence that produced nominations alongside the final promotion decisions, or to review only the final decisions. An outcome-only calibration detects over-nomination failures: cases where a manager nominates an engineer whose evidence, when reviewed by all managers, appears weaker than the evidence of recently promoted engineers at the same level transition. It does not detect under-nomination failures: cases where an engineer who has met the criteria has not been nominated, because those engineers produce no packet and thus produce nothing for the calibration session to review. The under-nomination failure is the failure mode that produces the departure pattern — engineers who are performing strongly, believe they are eligible for promotion, and receive no signal that the organization agrees — and it is structurally invisible in an outcome-only architecture. The calibration architecture should include three components that are not present in a standard outcome-only calibration. First, an eligibility roster: a quarterly report of all engineers at each level who have been at that level for at least the minimum period specified in the promotion cadence model, segmented by manager, with their most recent performance review summary attached. This makes the not-yet-nominated eligible population visible, and it is the denominator against which the nomination rate should be measured rather than the total headcount. Second, a nomination coverage review: for any engineer who appears on the eligibility roster and does not have an active promotion packet, the nominating manager is required to submit a brief written explanation of why the engineer has not been nominated — not as a punitive exercise, but as a mechanism that makes the criteria-interpretation and informal-policy surface visible before it produces a departure. Third, a process-consistency review: a quarterly review of the evidence types and evidence volumes in recent promotion packets, to detect whether manager-to-manager variance in evidence standards is emerging. A calibration session that reviews only the quality of the nominated engineers produces excellent calibration within the nominated population and zero visibility into the criteria-interpretation variance that determines who enters the nominated population. Connect this to the engineering performance management decision record: the calibration architecture for promotions and the calibration architecture for performance reviews are solving the same problem — ensuring that similar performance produces similar outcomes across different managers — and they should be designed as a coherent system rather than as separate processes; in particular, the performance review calibration sessions should be designed to detect the same under-rating failures that the promotion calibration sessions are designed to detect, because under-rated performance reviews produce under-nominated promotions with a one-to-two-quarter lag.

Section 4: Promotion timeline and cadence model. The promotion timeline model specifies how frequently the organization formally reviews promotion eligibility, what the nomination submission timeline looks like, and what the minimum-in-role requirements are and how they are documented. The most consequential decision in this section is whether the cadence model is explicit and written or implicit and manager-determined. A written cadence model specifies: the promotion cycles per year (quarterly, biannual, or annual, with explicit rationale for the choice and its implications for engineer waiting time); the submission deadline for nominations (the date by which a manager must submit a packet for the engineer to be considered in the current cycle); the decision timeline (how quickly after the calibration session the decisions are communicated to engineers); and the feedback timeline (how quickly after a no-promotion outcome the manager is required to have a development conversation with the engineer that specifies what needs to be demonstrated before the next cycle). The minimum-in-role requirement deserves specific treatment because it is the most common source of informal policy accumulation: if the official criteria document does not specify a minimum time at the current level before promotion eligibility, managers will fill the gap with their own minimum requirements imported from their prior companies, producing the variance the 72-person company experienced. The minimum-in-role requirement, if the organization chooses to have one, must be documented explicitly in the official criteria alongside the rationale — not because the specific number matters more than any other policy choice, but because an undocumented minimum is the most direct path to an informal policy that three managers implement differently based on where they worked before. The cadence model should also specify the treatment of off-cycle promotions — promotions that happen outside the regular cycle for engineers who have demonstrated exceptional performance — including the evidence standard for off-cycle nominations (typically higher than the standard cycle threshold), the approval model (typically requires the manager's manager's approval in addition to the VP), and the frequency constraint (typically no more than one per manager per year) to prevent the off-cycle mechanism from becoming an alternative to the standard process rather than a supplemental track for truly exceptional situations.

Section 5: Promotion communication, feedback, and re-evaluation model. The promotion communication model specifies how promotion decisions are communicated — both to the engineers who are promoted and to the engineers who are not — and what the feedback and re-evaluation architecture looks like for engineers who are not promoted in a given cycle. The most consequential decision in this section is whether the feedback provided to engineers who are not promoted is specific enough to constitute a genuine development plan rather than a general directional observation. The failure mode is the same one that appears in the performance management decision record: a manager who provides a no-promotion outcome accompanied by "keep doing what you're doing and continue developing your cross-team influence" has not provided a development plan; they have provided a directional label that the engineer cannot use to prioritize their work, calibrate their effort, or assess their progress. A specific development plan for a no-promotion outcome specifies: the specific criteria that were not met (not "cross-team influence" but "demonstrated technical leadership on a project that required coordinating dependencies across at least two teams beyond your immediate team"), the specific evidence types that would constitute demonstration of that criterion in the next cycle (a project that required the engineer to drive a technical decision that was contested between two teams, with the engineer's reasoning documented in writing and the outcome attributable to the engineer's intervention), and the feedback cadence during the next cycle (a check-in conversation at month two and month four to assess whether the engineer's work is producing evidence of the required type, rather than a check-in at month six when the next cycle submission deadline is two weeks away). The re-evaluation architecture should specify what happens when an engineer receives a no-promotion outcome in two consecutive cycles despite following the development plan: whether this triggers a manager-level review (is the manager providing sufficient coaching?), a criteria review (are the criteria appropriately calibrated?), or a scope review (does the engineer have access to work that would allow them to demonstrate the required behaviors?). An engineer who cannot demonstrate a criterion because they are not assigned to work that requires it is an org design problem, not a performance problem, and the re-evaluation architecture should produce that distinction explicitly. Connect this to the engineering compensation philosophy decision record: promotion decisions and compensation decisions are coupled, and the promotions process ADR and the compensation philosophy ADR should be designed as a coherent system; specifically, the promotions process should specify what happens to compensation when a level change is approved (the mechanism, the timing, the magnitude), and the compensation philosophy should specify whether compensation advances ahead of level (merit increases within level) or only at level transitions, because an engineer who has met the promotion criteria but is waiting for the next cycle while a peer hired at the same time at the same level received a merit increase has experienced a compensation inequality that is indistinguishable from the compression failures described in the compensation philosophy decision record, regardless of whether the source was a level decision or a merit decision. Connect this to the open-source extractor: the reasoning behind IC promotions process decisions — why the calibration architecture reviews nomination evidence rather than only final outcomes, why the criteria model specifies evidence types rather than leaving evidence to manager discretion, what the informal policy audit is designed to detect and how frequently it runs, why the self-nomination mechanism was added and what failure mode it prevents — is reasoning conducted in AI chat sessions across multiple promotion cycles and leadership transitions and rarely preserved anywhere; the promotions process decision record that captures that reasoning is the institutional memory that prevents the next head of engineering from inheriting a process whose design rationale has been lost, and re-discovering at the cost of three departures what the previous iteration learned about what happens when calibration architecture is designed to catch only one direction of error.

FAQ

Further reading

• Engineering career ladder decision record — the promotion criteria are derived from the career ladder's level definitions; the promotions ADR operationalizes how those criteria are applied and what evidence is required to demonstrate them; a career ladder that defines level criteria without specifying the evidence model produces a criteria floor bounded by the least specific criterion in the document, because any criterion that can be interpreted as satisfied by a range of evidence types will be interpreted differently by different managers in ways the calibration architecture cannot detect.
• Engineering feedback culture decision record — the evidence base for a promotion packet is built from the performance period's feedback documentation; a feedback culture that produces specific behavioral observations with named outcomes provides the promotion packet's evidence base at no additional administrative cost; a feedback culture that produces directional summary statements provides no evidence base and forces the manager to reconstruct behavioral evidence retroactively, which produces evidence of lower specificity and lower calibration reliability than evidence documented at the time of observation.
• Engineering performance management decision record — the calibration architecture for promotions and the calibration architecture for performance reviews are solving the same problem — ensuring that similar performance produces similar outcomes across different managers — and they should be designed as a coherent system; in particular, the performance review calibration sessions should be designed to detect the same under-rating failures that the promotion calibration sessions are designed to detect, because under-rated performance reviews produce under-nominated promotions with a one-to-two-quarter lag.
• Engineering technical interview design decision record — the interview loop and the promotions process are jointly responsible for the level distribution of the engineering organization; the criteria the interview loop uses to evaluate candidates at a target level and the criteria the promotions process uses to evaluate internal advancement to that level should be derived from the same career ladder level definitions and described in consistent language; an organization whose interview rubric describes one model of staff-level performance and whose promotion criteria describe a different model will produce a level distribution whose internal consistency cannot be evaluated.
• Engineering compensation philosophy decision record — promotion decisions and compensation decisions are coupled and should be designed as a coherent system; the promotions process should specify what happens to compensation when a level change is approved, and the compensation philosophy should specify whether compensation advances ahead of level or only at level transitions, because an engineer who has met the promotion criteria but is waiting for the next cycle while a peer at the same level received a merit increase has experienced a compensation inequality indistinguishable from the compression failures the compensation philosophy decision record is designed to prevent.
• Open-source extractor — the reasoning behind IC promotions process decisions — why the calibration architecture reviews nomination evidence rather than only outcomes, why the criteria model specifies evidence types rather than leaving evidence to manager discretion, what the informal policy audit is designed to detect, why the self-nomination mechanism prevents the undocumented-criteria failure mode — is reasoning conducted in AI chat sessions across multiple promotion cycles and rarely preserved anywhere; the promotions process decision record that captures that reasoning is the institutional memory that prevents the next head of engineering from re-discovering at the cost of departures what the previous iteration learned about what happens when calibration architecture is designed to catch only one direction of error.