The engineering career ladder decision record: why the promotion criteria model you chose determines your talent retention surface and your compensation equity ceiling

A 35-person B2B SaaS company built project management software for professional services firms. The engineering organization had grown from 4 to 35 over three years, adding engineers as product needs expanded. The company had no formal engineering career ladder. Promotions happened via manager conversations — an engineer would have a 1:1 with their manager, express interest in growing, and the manager would evaluate whether they were "ready for senior" based on their judgment of the engineer's contributions over the past year. Three managers ran the three engineering teams: the product team, the platform team, and the integrations team. Each had been promoted to manager from senior IC roles at different companies with different cultures and different implicit definitions of what senior engineering looked like.

Three engineers — call them A, B, and C — joined the company in the same quarter, all as mid-level software engineers at the same base salary. All three were strong individual contributors. Over the next 18 months, all three were promoted to senior engineer by their respective managers in separate promotion conversations with no cross-manager calibration. Engineer A, on the product team, was promoted after demonstrating that she could own the entire checkout flow — design, build, and maintain it with minimal guidance — across two major feature releases. Her manager's implicit criterion for senior was scope of independent ownership: could you take a significant surface area and be trusted to run it end-to-end? Engineer B, on the platform team, was promoted after delivering a particularly complex database migration and being the go-to reviewer for the team's most technically demanding PRs. His manager's implicit criterion for senior was technical depth: could you write the hardest code on the team and identify problems others missed in review? Engineer C, on the integrations team, was promoted after successfully coordinating a three-team API integration with an enterprise customer — managing the technical requirements across two vendor systems and the company's own API, handling the customer communication, and delivering on a tight deadline. Her manager's implicit criterion for senior was coordination ability: could you get complex, cross-functional work done by influencing people you didn't manage?

None of these criteria were wrong. All three engineers were genuinely strong at what they were evaluated on. The problem became visible 18 months after the three promotions, at the company's first formal compensation audit, which was triggered by a departing engineer sharing his offer letter salary with a colleague before he left. The audit revealed that the three "senior" engineers had compensation that differed by $45k: Engineer B — whose technical depth was evaluated against the platform team's pay range, adjusted by his manager's strong advocacy — was at the top of what the company informally thought of as the senior band. Engineer C — whose coordination skills were evaluated by a manager who weighted compensation conservatively and did not advocate for the high end of the range — was at the bottom. Engineer A was in the middle. All three had the same title, the same tenure at the company, and had been "promoted at the same time" in the sense that all three promotions had happened within the same 18-month window. Connect this pattern to the technical interview design decision record: the interview process and the career ladder are two ends of the same evaluation system; the three managers had also conducted different interviews, hired for different implicit criteria, and their compensation offer instincts at hire reflected the same divergent standards that later produced divergent promotion placements; the career ladder equity problem is partially inherited from the interview process, and a compensation audit that reveals level divergence should trigger an audit of the interview process for the same implicit criteria divergence.

The equalization process took three months and involved the engineering manager, the VP of Engineering, and the HR partner. The outcome was three compensation adjustments: Engineer C received a $28k adjustment to bring her to the midpoint of a newly-defined senior band; Engineer A received a $9k adjustment; Engineer B received no adjustment (he was at the top of the new band, which was set to include his salary). The conversation with Engineer B was the most difficult: his manager had to explain that the new band definition did not mean he had been overpromoted, but that other engineers at his nominal level had been undercompensated. Engineer B interpreted this, reasonably, as a signal that the organization had been paying him at a rate it now considered above the standard, and that future raises would be constrained by his position at the top of the band. He left two months later for a company that offered him a salary above the old top of his range. The post-departure analysis produced a written career ladder with explicit criteria. The written criteria had not existed when any of the three original senior promotions were made.

A 48-person developer tools company built a static analysis and code quality platform. The engineering organization had grown to 8 teams with 38 engineers. Two years before the events described here, the VP of Engineering had invested three months in writing a detailed six-level engineering ladder: levels 1–4 as individual contributor stages, level 5 as "senior engineer," level 6 as "staff engineer." The criteria at each level were documented in Notion, shared with all engineers at their annual review, and referenced in every promotion conversation. The VP had been proud of the ladder: it was explicit, it was written, it was accessible. The criteria at level 5 required that an engineer own a significant service or codebase area with minimal guidance, identify and drive technical improvements proactively, provide effective code review across the team, and estimate and deliver work reliably. The criteria at level 6 required that an engineer design systems that span multiple services with org-wide technical implications, author RFCs that are adopted across the engineering organization, and provide technical leadership to at least two engineers at lower levels.

Two years after the ladder was published, the VP reviewed the promotions made under it. Three engineers had been promoted to staff level. Their profiles were strikingly similar: all three had high individual code output, all three had authored at least two RFCs that were adopted organization-wide, all three had done significant individual technical design work. There was nothing wrong with their promotions — they met the criteria. The VP's concern came from looking at who had not been promoted to staff. One engineer in particular had been evaluated twice for staff-level promotion and declined both times. Her name was Maria. Maria had mentored four junior engineers over the previous two years; all four had been promoted to senior engineer, and two of them credited her specifically in their promotion packets as the reason they were ready. She had led a complete redesign of the company's on-call rotation structure when the original 12-person rotation was producing burnout at a rate that was damaging the team — the redesign reduced the average weekly on-call load per engineer by 35% and reduced the informal escalation rate by 60%. She had authored the incident postmortem process that the entire company now used, and the adoption of that process had measurably reduced P1 incident recurrence: in the 12 months after the process was adopted, P1 recurrence (same root cause within 90 days) dropped from 23% to 8%. She had been the first call for production incidents across three teams for 18 months because engineers on those teams knew she would unblock them faster than any other senior engineer available. Connect this pattern to the incident postmortem culture decision record: the postmortem process Maria authored was a systems improvement that reduced the organization's recurrence rate — a direct, measurable organizational outcome; a career ladder that does not credit the authorship and adoption of organization-wide operational processes as evidence of staff-level organizational leadership is a ladder that attributes organizational reliability improvement to no one, because the improvement is produced by process work rather than by architectural design work, and process work does not produce the RFC artifact the ladder specifies.

The two promotion declines for Maria had the same rationale: she had not authored any RFCs that were "adopted across the engineering org," and her system design artifacts were "not clearly at the staff level." Her manager had submitted her postmortem process documentation and on-call rotation redesign as evidence of system design at the staff level. The calibration panel had reviewed these and concluded that they were process documents, not system design documents, and that the ladder criteria specified architectural design — not operational process design. Maria's manager escalated to the VP of Engineering. The VP reviewed the calibration notes and discovered that the distinction between "architectural design document" and "operational process design document" was nowhere in the written ladder criteria — the criteria said "author RFCs that are adopted across the engineering organization," and the panel had interpreted "RFC" to mean technical architecture documents specifically. The panel's interpretation was plausible but not written. The ladder criteria had an implicit constraint that was not explicit in the document: it was a ladder for a specific archetype of staff engineer (the architect) and it was invisible to a different archetype of staff engineer (the operational leader) who was producing org-level impact through a different set of artifacts.

Maria left three weeks after the second decline. Her manager's exit conversation confirmed that the primary reason was the promotion decision. The VP of Engineering revised the ladder criteria the following month: the level 6 organizational contribution criteria were expanded to include "leads or authors organization-wide operational processes whose adoption produces measurable improvements in engineering reliability or team effectiveness" alongside the RFC requirement, and the definition of "technical leadership" was extended to include on-call and operational mentorship contributions alongside architectural design review. The revision produced a different problem: two of the three existing staff engineers raised concerns with their respective managers. Their argument was that the original criteria had been set at a specific technical bar, and they had met that bar, and if the bar was now lower — if Maria would have been promoted at the new bar — then the implicit message was that they had been held to a higher standard than was now required. Both conversations required 2–3 weeks of management time. Neither engineer left, but both described the revision as "confusing" in the next engagement survey. Connect this outcome to the developer experience measurement decision record: the engagement survey signal — "confusing" criteria revision — is a lagging indicator of a criteria design failure that was visible 18 months earlier when the two promotion declines were made; a developer experience measurement model that tracked "clarity of career growth path" as a real-time signal rather than an annual survey metric would have surfaced the criteria confusion earlier, before the departure that prompted the revision.

A 42-person fintech startup built a payment orchestration platform for marketplace operators. The engineering organization had grown from 8 to 42 engineers over 18 months during a period of rapid customer acquisition. During the growth phase, the company had faced a retention challenge: several strong mid-level engineers had received competing offers from larger companies with more established career ladders and higher base salaries. The engineering director had made a decision, informally discussed with the CEO but not written down: during the growth phase, the bar for senior engineer would be effectively "solid IC, no significant incidents in the past 6 months, strong performance reviews." The decision was practical — the company couldn't match FAANG compensation, but it could offer faster promotions, and a senior title carried weight in the market. Six engineers were promoted to senior engineer in an 18-month window under this informal bar. The engineering director understood internally that the bar had been lowered. The written ladder criteria, which had been created before the growth phase, still specified the original standard: ownership of a significant service area, independent technical judgment, proactive identification of improvements. The six promotions were made with written justifications that cited the formal criteria, but the calibration was against the informal lower bar.

The growth phase ended when two large contracts were delayed and the company reduced hiring. To fill capability gaps without growing headcount, the company began selectively hiring externally at the senior and staff levels. External candidates were benchmarked against market rates: the recruiting team used Levels.fyi data and the data from the most recent Radford survey, which the company had purchased for the first time as part of the hiring process. Two external hires were brought in at the senior level with salaries 18–22% above the six internal engineers promoted during the growth phase. The discrepancy surfaced when one internal engineer — promoted during the growth phase at the lower bar — was assigned to work with one of the two external hires on a new payments feature. The internal engineer, who had 3 years of tenure at the company, asked her manager directly in a 1:1 whether her salary was in line with market for her level. Her manager, who knew about the discrepancy and had expected this conversation, confirmed that there was a gap. Connect this pattern to the on-call rotation design decision record: the internal engineers promoted during the growth phase had, in many cases, earned their promotions partly through sustained on-call reliability — being available, managing incidents competently, carrying rotation load during the company's fastest growth — and on-call performance had been an implicit factor in the informal lower-bar promotions; when external hires were benchmarked against market rates, their on-call experience at other companies was not the same as on-call experience at this specific company's critical growth period; the gap between the informal bar (which credited operational reliability as a promotion factor) and the market benchmark (which did not specifically credit that specific company's operational history) was part of the equity problem that the market-benchmarked external hiring revealed.

The engineering director escalated to the CEO and the VP of People. The HR review confirmed the discrepancy: four of the six growth-phase promotions were compensated below market rate for the senior level as defined by the Radford benchmark. The remaining two were at or near market (one had negotiated more aggressively at promotion time; one had received a retention offer 6 months prior for a different reason). The remediation had two options. Option A: retroactively level the four engineers down to mid-level, adjust their compensation to mid-level market rates (a pay cut), and re-evaluate them for senior under the original formal criteria. This was rejected immediately as politically untenable and legally risky — telling four engineers that their promotions were made in error and adjusting their pay downward would produce immediate departures and potential legal exposure. Option B: bring the four engineers' compensation to market for their current level (an average increase of $29k per engineer), accept that the informal bar had been applied and that the promotions stand, and use the current external hiring exercise to reset the formal bar to the market benchmark going forward. This was the option chosen. It cost $116k in additional annual compensation for four engineers.

The implementation of Option B required a different conversation for each engineer. Three of the four responded positively — the compensation adjustment was welcomed, the explanation was accepted, and the conversation produced increased rather than decreased trust. The fourth engineer interpreted the explanation differently: if her promotion had been at a lower bar than the company's formal criteria, and the company was now restoring her compensation to match the formal criteria level, the implication she drew was that the company had promoted her before she was fully ready and was now paying her to reach the level she nominally held. She resigned. Her manager's post-departure notes described the conversation as "the adjustment framing landed as 'you were promoted early' even though we intended it as 'the bar was informal and we're correcting for market.' The distinction was too subtle under the circumstances."

Structural properties set by the engineering career ladder decision

Three structural properties are determined by the decisions — often made informally, under time pressure, or by default as an organization grows — about what behaviors and contributions to reward at each level, how promotions are calibrated across managers, and how compensation bands are aligned to the criteria and the market. None of these properties are visible in the sprint velocity metric, the deployment frequency dashboard, or the quarterly engagement survey. They are established by whether the criteria make organizational contribution as legible as technical contribution, whether the calibration model requires cross-manager consistency or permits individual manager divergence, and whether the band alignment is refreshed against market rates at a cadence that prevents the equity gap from compounding before a hiring cycle forces reconciliation.

Property 1: The talent retention surface. The talent retention surface is the set of engineers for whom the current career ladder provides a credible path to growth. Engineers who see a clear, credible path — whose current contributions match the criteria for the next level, whose trajectory is confirmed in regular 1:1s with specific artifact examples, whose manager is calibrated to the same standard as peer managers — stay. Engineers who are doing the work that matters most to the organization but that the ladder does not explicitly credit leave when a competing offer makes the path clearer elsewhere. The talent retention surface of an engineering organization is invisible until a departure makes it visible: the first departure of a high-performing engineer whose contributions were not ladder-aligned is a revelation about which work the ladder is crediting and which it is not. The second departure with the same profile is a system finding. The second failure pattern produced both: Maria's departure was the revelation, and the engineering director's post-departure review revealed two additional engineers whose primary contribution mode was operational rather than architectural, who were at or near the staff evaluation gate, and who had begun having "what does my path look like" conversations with their managers that had previously been absent. Both cited Maria's departure as having made the path less credible, not more: if the engineer whose organizational impact was most visible couldn't be promoted to staff, the path to staff for engineers with similar contribution modes was not legible. Connect this property to the technical interview design decision record: the talent retention surface is partially determined by the interview process because the interview process is the intake gate for the engineering organization; a ladder that retains engineers whose primary contribution mode is individual architectural output and loses engineers whose primary contribution mode is operational leadership and organizational improvement will select, through the interview process, for the archetype it retains; over time, the engineering organization's capability mix converges toward the archetype the ladder rewards and diverges from the archetypes it does not reward; when the organization needs the non-rewarded archetype (a production reliability crisis, a team culture degradation, a cross-team coordination failure), it has fewer engineers of that type because the ladder's retention surface has been selecting against them since the criteria were first written.

Property 2: The compensation equity ceiling. The compensation equity ceiling is the largest gap the ladder's current criteria and calibration process permit between two engineers at the same nominal level who are both performing at a strong standard. In an organization with explicit criteria, calibrated bands, and explicit within-band placement policy, the ceiling is bounded by the band width and the precision of the placement policy. In an organization with implicit criteria, the ceiling is bounded by the divergence between individual managers' implicit standards — which diverge in proportion to the number of managers, the variety of career trajectories represented among the manager cohort, and the time elapsed since the last cross-manager calibration event. The first failure pattern's $45k ceiling was produced by three managers over 18 months with no calibration — a moderate-size gap in a moderate-size organization over a moderate time horizon. A larger organization with more managers, over a longer horizon without calibration, produces proportionally larger gaps. The equity ceiling is a legal and morale risk that is invisible until a measurement event forces quantification. The measurement event itself does not cause the gap — it reveals the gap that has been accumulating under implicit criteria. The remediation cost is not the cost of the measurement event; it is the cost of all the divergent decisions made under implicit criteria since the last calibration, plus the management time required to equalize them, plus the departure risk of engineers who interpret the equalization in ways the organization does not intend. Connect this property to the developer experience measurement decision record: the compensation equity ceiling is a measurable property of the compensation system that requires deliberate measurement to stay visible — like the developer productivity signal in the DX measurement decision record, it is invisible without a principal owner who is accountable for tracking it against a defined benchmark; the compensation equity ceiling should be measured quarterly (maximum gap between same-level engineers at each level) and reported to the VP of Engineering alongside the engineering velocity metrics; a ceiling that is widening indicates implicit criteria divergence that is compounding, and a ceiling that is widening faster than the compensation band width indicates that within-band placement decisions are not using explicit criteria.

Property 3: The calibration consistency overhead. The calibration consistency overhead is the total engineering management time consumed per quarter by the resolution of inconsistent leveling decisions across managers — conversations that occur after individual managers have made promotion decisions independently and then discover that the decisions are inconsistent when compared across teams. In an organization with explicit criteria and a defined calibration process, the overhead is bounded: the calibration panel review (a bounded time cost, typically 2–3 hours per candidate) ensures consistency in advance, and the decisions produced by the panel require minimal post-hoc revision. In an organization with implicit criteria, the overhead is unbounded: each promotion decision that a peer manager disagrees with produces a criteria conversation, which produces a criteria revision discussion, which produces a political conversation about what the revision implies for engineers previously promoted under the old implicit standard, which produces escalation. The second failure pattern's criteria revision produced 3+ weeks of management time across multiple manager conversations — time that was not available for product decisions, hiring calibration, or team support. The calibration consistency overhead is not a line item in any engineering management dashboard; it appears in "management availability" and "responsiveness" metrics but is not attributed to the career ladder. The overhead compounds because implicit criteria revisions are not stable: when managers agree on an implicit revision, they agree on the revision as they understand it at the moment of agreement, and their understandings diverge again over the following promotion cycles. A written, explicit criteria framework that is agreed before individual decisions are made does not eliminate calibration discussion — it channels the discussion to the framework rather than to ad-hoc post-hoc negotiation. Connect this property to the incident postmortem culture decision record: the calibration process is the career ladder's structural equivalent of the postmortem process — a structured event where individual judgment is compared across multiple evaluators against a shared standard; a postmortem process without a shared root cause category framework produces inconsistent findings that each investigating engineer attributes differently, requiring a post-hoc reconciliation that generates its own overhead; a calibration process without shared explicit criteria produces inconsistent promotion decisions that each manager justifies differently, requiring the same post-hoc reconciliation; both processes benefit from the same design principle: the shared standard must be agreed in advance, not reconstructed after individual assessments have been made independently.

The engineering career ladder ADR: five sections

Section 1: The promotion criteria framework. Begin the engineering career ladder decision record by specifying what behaviors and contributions are required for each level, in explicit, verifiable terms. The framework must cover four dimensions. Scope and ownership: what the engineer is independently responsible for at this level — specified as observable ownership with verifiable scope, not aspirational responsibility. A level 5 criterion of "owns a significant service with minimal guidance" requires a definition of "significant" (team-critical service, cross-team dependency, customer-facing surface) and "minimal guidance" (no design review required, or quarterly architecture sync required, or pair-review required for major changes — these are materially different levels of independence). Technical contribution: what technical work is produced at this level and what quality bar it meets — with examples of what counts as evidence, not just descriptions of the work category. A criterion of "authors architecture documents" requires specification of what an architecture document is (a written design before implementation, reviewed by at least one engineer outside the team) and what adoption means (at minimum, the design was used as the basis for the implementation; preferably, it was referenced by at least one engineer who did not author it). Organizational contribution: how the engineer improves the organization's output beyond their own individual work — mentorship, process improvement, on-call leadership, postmortem ownership, cross-team coordination — specified with the same artifact-example precision as technical contribution. This dimension is the most commonly underspecified in engineering ladders and the most commonly responsible for criteria-archetype mismatches at the senior and staff levels. A criterion of "demonstrates technical leadership" without specifying whether operational process improvement counts as technical leadership is an implicit criterion: each calibration reviewer will answer the question differently, and the ladder will systematically exclude the archetype whose primary leadership mode is process and operational rather than architectural. The criterion should be specified as outcomes with artifact examples: "evidence of organizational leadership with engineering-wide impact — examples include: RFCs adopted across teams, operational processes authored and adopted that produce measurable reliability improvements, mentorship of engineers who credit the mentorship as enabling their promotion; the artifact type is less important than demonstrating the organizational impact." Communication and influence: how the engineer makes decisions visible, shares knowledge, and influences technical direction at the scope appropriate to their level. Connect this section to the technical interview design decision record: the interview evaluation dimensions and the ladder evaluation dimensions should be consistent, because the interview is the intake gate and the ladder is the growth gate of the same system; if the interview evaluates systems design, cross-team communication, debugging methodology, and self-directed learning, the ladder should credit all four dimensions at the corresponding levels; a mismatch between interview criteria and promotion criteria means engineers are hired for capabilities the organization does not reward growing, and engineers who grow in the directions the ladder rewards but were not evaluated for those dimensions in the interview will face inconsistent evaluation at both the hire and promotion stage.

Section 2: The calibration model. Specify how promotions are decided and who is involved. The calibration model has three elements. The promotion packet: what the engineer and manager must submit as evidence for each evaluation dimension, with a format that requires specific artifacts rather than holistic assessments. The packet should require dated examples with measurable outcomes where possible — not "improved the team's code review culture" but "introduced the two-reviewer requirement for changes to the payments service; in the following quarter, the change-related production incident rate for payments dropped from 3 per month to 1 per month." Specific, dated, measurable artifacts prevent the most common calibration failure: the strong manager failure, where an advocate-manager produces a promotion by advocacy force rather than by evidence quality. A well-written advocate-manager can describe any engineer's contributions in terms of the ladder criteria, and a calibration panel without artifact requirements has no basis for disagreeing with the description other than a competing holistic assessment. Artifact requirements shift the calibration from "do we believe the manager's description?" to "do the artifacts demonstrate the criteria?" The calibration panel: the direct manager (who has daily context), one manager from a different team (who has direct context on cross-team contributions and is not subject to the direct manager's advocacy bias), and an engineering manager at the level above the promotion level (to calibrate against the organization's overall standard, not just the team's standard). The cross-team panel member is specifically important for surfacing on-call and operational contributions that are visible to the on-call rotation peers but invisible to a cross-team reviewer who did not share the rotation: the panel member should be asked specifically about on-call performance observations if the engineer is in a shared rotation. The decision protocol: panel consensus is required; individual manager advocacy is not sufficient without supporting artifacts; calibration disagreements are resolved by identifying specifically which criterion the engineer does or does not meet, with which evidence — not by negotiating until one party concedes. Connect this section to the on-call rotation design decision record: the on-call rotation is a significant source of promotion evidence for the senior and staff levels, where operational ownership and incident response quality are key differentiators between levels; the calibration panel should explicitly review on-call performance artifacts as part of the promotion evaluation — postmortem authorship quality, incident response time, rotation design contributions — because on-call work is visible to the on-call rotation peers but invisible to cross-team panel reviewers who did not share the rotation; a calibration panel without a structured process for surfacing on-call contributions will systematically under-credit engineers whose highest-impact work is operational, producing the same exclusion failure as a ladder criteria framework that does not specify operational contribution as a creditable dimension.

Section 3: The compensation band alignment. Specify how levels map to compensation ranges and how those ranges are calibrated against the market. The band alignment has three components. The market data source: which benchmark the organization uses (Levels.fyi for tech companies with strong market visibility; Radford or Option Impact for a more comprehensive market survey including pre-IPO compensation and equity data), how often the benchmark is refreshed (annually is the minimum; for organizations in markets with high compensation volatility — AI/ML engineering, fintech, infrastructure — semi-annually is more appropriate), and who owns the refresh process (an HR partner, the VP of Engineering, or an engineering manager who is accountable for maintaining the alignment). The within-band placement policy: where within the band a new hire or promoted engineer is placed, and what criteria justify higher or lower placement. Without an explicit within-band placement policy, within-band placement is effectively determined by manager advocacy and negotiation, producing the first failure pattern's equity ceiling. An explicit within-band placement policy specifies that promotions land at the midpoint of the new level's band by default, that placement above the midpoint requires a written justification citing the specific criteria that differentiate this engineer from the standard, and that new hires are placed based on years of relevant experience relative to the band's experience distribution, not on negotiation outcome alone. The overlap policy: whether adjacent-level bands overlap, and by how much. Overlapping bands (the top 20% of Level N overlaps with the bottom 20% of Level N+1) allow within-band growth without requiring a promotion decision when an engineer reaches the top of their current band, and allow promotions without requiring a large compensation step-change when the engineer is already in the overlap range. Non-overlapping bands create a pressure to promote at the top of the band (the engineer is capped without a promotion) or to delay compensation increases until the promotion is confirmed (which produces a perception of compensation being withheld pending a promotion decision that may or may not materialize). The third failure pattern's informal bar lowering was a compensation management decision as much as a promotion criteria decision: the engineering director was using promotion as a compensation tool, and the ladder's informal lower bar was the mechanism. Specifying an explicit overlap policy and within-band placement policy would have provided an alternative — a new-hire landing at the top of the mid-level band rather than at the bottom of the senior band — that would have retained the engineer without miscalibrating the senior-level standard.

Section 4: The promotion cadence and pipeline management. Specify when promotions can happen, what the pipeline looks like, and how timing is communicated to engineers. The cadence policy has three elements. The promotion windows: when in the year promotions are processed. Twice per year (mid-year and end-of-year) is the most common cadence; it is frequent enough that engineers who are ready do not wait more than 6 months, and infrequent enough that the calibration panel can invest the time to do it well rather than rushing through ad-hoc requests. Quarterly cycles are high-overhead for management calibration and produce pressure to promote at each cycle rather than when the criteria are genuinely met. Ad-hoc promotions are equitable in principle (the engineer gets promoted when ready, regardless of cycle timing) and equity-risky in practice (ad-hoc timing is influenced by manager advocacy intensity and compensation budget availability in ways that are invisible to the engineer, and produce the same equity ceiling as implicit criteria). The pipeline visibility policy: whether engineers know they are being considered for promotion in an upcoming cycle and what the expected timeline is. The "in the pipeline" failure is the most common cadence management failure: a manager tells an engineer they are being considered without specifying a cycle, leaving the engineer uncertain for 6–12 months while the manager waits for a cycle they judge to be optimal. The engineer experiences this as a promise that is not kept. Specifying a pipeline visibility policy means that managers must commit to a specific cycle ("I'm targeting the June review for your promotion based on what I'm seeing") or explicitly not commit ("you're close but I need more evidence on X before I can advocate — here's what I'd need to see before the December cycle"). The decline process: when an engineer is evaluated for promotion and not promoted, what they are told, and what happens next. A promotion decline must include: which specific criteria were not met, which specific artifacts were evaluated and found insufficient, and what the path to demonstrating the missing criteria looks like before the next cycle. A decline that says "the panel felt you weren't quite ready" without criteria specificity is not actionable — it tells the engineer to "be better" without specifying in what dimension. Connect this section to the incident postmortem culture decision record: the promotion decline process requires the same communication design as the postmortem finding — both require the person delivering a difficult assessment to separate the systemic finding from individual attribution and to provide a specific, actionable path forward rather than a holistic judgment; a postmortem that says "the engineer should have caught this" is as unhelpful as a promotion decline that says "you're not ready yet" — both omit the specific criteria gap that would make the finding actionable and allow the person receiving the assessment to change the outcome; the postmortem process's move to systems findings ("the monitoring configuration permitted this failure mode") is the analogue of the promotion decline's move to criteria findings ("the staff-level organizational contribution criterion requires evidence of org-wide process impact — your current artifacts demonstrate team-level process impact and we need one example at the org level").

Section 5: The ladder transparency policy. Specify what information is visible to engineers about the career ladder and their standing within it. Transparency policy has three dimensions. Criteria visibility: whether the criteria at each level are visible to all engineers or only to managers and engineers currently being evaluated. The argument for full criteria visibility is that implicit criteria produce divergent evaluation, and the argument for restricted visibility is that explicit criteria produce criteria-gaming ("I will author 3 RFCs this quarter to meet the staff requirement" rather than authoring RFCs because there is an organizational need). The criteria-gaming concern is real but overstated: an engineer who authors high-quality RFCs because they want a promotion is producing work the organization values, even if the motivation is the promotion rather than the architectural interest. The real risk of explicit criteria is rigidity: criteria specified as artifacts ("authored 3 RFCs") rather than outcomes ("technical decisions with org-wide implications are made with input from this engineer") reward the artifact over the outcome. Criteria stated as outcomes with artifact examples — the specification recommended in section 1 — are less gameable than artifact requirements because the calibration panel evaluates whether the outcome was produced, not whether the artifact format was met. The default should be full criteria visibility, because the second failure pattern's exclusion — engineers who did not know whether their operational contributions counted toward staff-level criteria — would have been partially preventable if the criteria had been visible and Maria's manager had been able to say clearly that the current criteria did not credit her work. Band visibility: whether compensation bands are visible to all engineers. Full band visibility reduces the equity ceiling by making the within-band placement decision visible to the engineer and enabling them to ask whether their placement is consistent with the policy; it also reduces the likelihood that a manager will make an advocacy-based placement above the policy band, because the above-policy placement would be visible to the engineer as above-band. The primary argument against band visibility is that it produces compensation conversations that managers do not feel equipped to have; this is an argument for manager training, not for band opacity. Distribution visibility: whether the current headcount at each level is shared with engineers. Distribution visibility provides organizational context: an engineer evaluating whether staff engineer is a realistic target at a 42-person company benefits from knowing that the company currently has 2 staff engineers out of 38 ICs, because this indicates that the company's definition of staff is calibrated to a scarcity model and the promotion criteria will reflect that scarcity. Connect this section to the open-source extractor: the career ladder decisions — what criteria to use at each level, whether to make bands visible, how to handle the retroactive calibration problem when historical promotions were made under a different bar, whether organizational contribution should be a first-class criterion or a secondary credit — are exactly the kind of multi-variable trade-off discussions that happen in AI chat conversations; a CTO at a 35-person company deciding whether to make compensation bands visible to all engineers, or how to handle the scenario where an engineer gaming explicit criteria produces the artifact without the outcome, or whether to set overlapping bands at the cost of promotion-step-change predictability — those conversations happen with ChatGPT or Claude because they are the kind of decisions where a thinking partner who surfaces second-order effects changes the outcome; the written output is the career ladder document, which records the criteria and the policies but not the reasoning behind the transparency choice, the artifact-vs.-outcome decision, or the overlap policy selection; when the ladder needs to be revised 18 months later because the original criteria are producing the wrong archetype or the bands have drifted from market, the engineer or manager revising it cannot recover the original rationale from the policy document; the reasoning is in the AI chat history, and the extractor surfaces it.

FAQ

Further reading

• Technical interview design decision record — the interview process and the career ladder are two ends of the same evaluation system; the interview is the intake gate and the ladder is the growth gate; a mismatch between interview criteria and promotion criteria means engineers are hired for capabilities the organization does not reward growing, and the engineering organization's archetype mix converges over time toward the archetype the ladder retains; if the interview evaluates systems design, operational thinking, cross-team communication, and self-directed learning, the ladder must credit all four dimensions at the corresponding levels or the interview selection will produce engineers whose careers stall at the stage where the unmatched criteria become the promotion gate.
• Developer experience measurement decision record — the compensation equity ceiling is a measurable property of the compensation system that requires deliberate measurement to stay visible, just as the developer productivity signal requires a defined metric and a principal owner to remain actionable; both degrade invisibly without measurement ownership — the equity ceiling widens as implicit calibration diverges, and the developer productivity signal drifts as team structure and tooling change — and both require a refresh cadence calibrated against an external benchmark (market compensation data, industry DX benchmarks) rather than against internal history alone; the engagement survey's 'clarity of career growth path' signal and the DX measurement model's engineering productivity signal face the same organizational incentive structure: neither generates short-term product output, both degrade invisibly, and both require a principal owner who is accountable for maintenance rather than only activation.
• On-call rotation design decision record — the on-call rotation is a significant site of level-differentiation at the senior and staff levels, where operational ownership and incident response quality are key criteria for promotion; the calibration panel must have a structured process for surfacing on-call contributions because on-call work is visible to the rotation peers but invisible to cross-team reviewers who did not share the rotation; an engineer who redesigns the rotation, reduces the informal escalation rate, and mentors the engineers who are new to on-call is doing organizational leadership work that is directly relevant to the staff-level criteria — but only if the calibration process explicitly asks for on-call performance artifacts, and only if the on-call leadership criterion is specified in the promotion framework rather than treated as a subcategory of 'technical leadership' that defaults to architectural design work.
• Incident postmortem culture decision record — the postmortem process creates promotion artifacts that the ladder may or may not credit: engineers who author high-quality postmortems demonstrating systems thinking, who drive the action items to completion, who improve the postmortem process itself are producing organizational output that is directly relevant to senior and staff promotion criteria — but only if the calibration panel treats postmortem quality as a creditable artifact and the ladder criteria specify what postmortem ownership demonstrates at each level; a blameless postmortem culture that produces high-quality findings and system improvements is a talent retention mechanism for engineers who value operational work, and a career ladder that does not credit postmortem authorship at the appropriate level undermines the retention benefit by making the operational work invisible in the promotion evaluation.
• Open-source extractor — the engineering career ladder decisions — what criteria to use at each level, whether to weight operational contribution equally with architectural contribution, how to handle the retroactive calibration problem when historical promotions were made under an informal bar, whether to make bands visible at the risk of producing compensation conversations managers are not equipped to have — are exactly the kind of multi-variable trade-off decisions that happen in AI chat conversations with a thinking partner; a CTO at a 35-person company thinking through whether overlapping salary bands are worth the within-band equity complexity, or whether outcome-based criteria are less gameable than artifact-based criteria, or how to communicate a retroactive bar correction without implying that previous promotions were errors — those conversations happen in ChatGPT or Claude; the written output is the career ladder document, which records the criteria and the policies but not the reasoning behind the transparency decision, the band overlap choice, or the organizational contribution weighting; when the ladder needs revision 18 months later, the engineer or manager revising it cannot recover the original rationale from the document alone.