Every calibration room I have sat in has a moment where somebody stops arguing about a person and starts arguing about a number. The tell is always the same: the sentence gets longer. Nobody says "she is a four." They say "she is doing four-level work and honestly with what we are paying her." The level has become a price, and from that point on the room cannot say a true thing without costing somebody money.
The fix is not discipline. It is a second document, written after the first one closes, that takes a level as its input and produces a number as its output, and is not allowed to look back.
A band is a promise about the range of the fair. A number is a fact about one person. Publish the promise. Protect the fact.
That used to be a design choice. It is now mostly law. The EU pay transparency directive took effect in June 2026: the range goes in the job ad or before the first interview, nobody may ask a candidate what they earned before, and an unexplained gap of five percent or more triggers a joint assessment and a six-month fix. Eighteen US states and Washington DC now require the range in the posting, Colorado since 2021. A company that still treats its bands as confidential is a year or two behind its own legal department.
Three jobs in one meeting
A review cycle has to do three things. Rank each person against a published standard. Decide what each of them is paid. Decide what each of them should work on next. They have different inputs, different cadences, and different people leading them, and when a company runs them through one document the failure is not that one of the three suffers. All three do, in a specific and predictable way.
One document, three jobs
What each job needs, and what it becomes when it has to share.
The third row is the one companies notice least. Everybody remembers their raise. Almost nobody remembers their development plan, and the reason is that they heard it in the same breath.
Level in, money out
The compensation document has one rule that matters more than the rest: it never decides a level. It prices one. The level arrives from calibration, confirmed and written, with the six marks beside it, and the compensation document's whole job is to turn that record into a number using a table anybody could read.
The table is a band per level. Six levels, six bands, in every lane. Adjacent bands overlap, so a promotion from the top of one lands in the lower-middle of the next without a cliff, and a strong level 3 can out-earn a new level 4 without anyone being wrong.
Bands, and where they overlap
A starting table, not a finding. Width is the ratio of top to bottom, widening with level because tenures at level lengthen; the conventional range is twenty to fifty percent, narrower at the bottom. Overlap between adjacent bands is deliberate and sits between ten and twenty-five percent: none makes every promotion a large raise, and past thirty the levels stop meaning anything. Midpoints step twenty to thirty percent per level, growing with level, larger than the textbook figure because there are six broad bands rather than fifteen narrow grades.
One band per level, and no sub-bands. Every half-step you add to give a raise without a promotion is a level the leveling document does not know about, and it will be used as one within a year.
The market pays engineers more than designers at the same level. It does, and the answer is a single differential per lane applied to the midpoint, not a separate band structure per lane. One number for design, one for research, one for product, engineering as the reference. Not a per-level adjustment; if you find yourself wanting a different differential at level 5 than at level 2, what you have found is that the market data at one of those levels is thin, and the correct response is to trust the structure.
Most companies reading this have no bands at all, only a spreadsheet of numbers that each made sense on the day. The first year is not a repricing. Publish the bands, hold every existing number where it is, and let the zones absorb the corrections: a person above their band is held, not cut, and drifts back over two cycles; a person below is moved to the entering zone at the first comp cycle. Nobody’s number goes down because a table appeared. That rule is what makes the table survivable.
The four zones
Inside a band, where a person sits is the only compensation decision that is actually about the individual, so it is the one to be most disciplined about. Four zones, and a rule for what moves a person between them — a rule that is applied from the calibration record, not argued in the comp meeting, because the comp meeting has no arguments in it.
Position in band
What moves a person up a zone, and — the more useful column — what does not.
The shaded zones are where most people spend most of their careers, and where every raise-without-promotion happens. The top zone is a list: everyone in it for two cycles is a promotion case somebody is avoiding.
Why it runs after
The year I learned why the order matters was 2022. I had joined Pitch to grow an organization, the SaaS correction arrived between my signing and the first board meeting, and the plan I was hired on was replaced by an envelope. That is the year a structure is for. Bands do not move because the plan did; the zone rules apply as written; nobody’s number becomes a conversation about the market, because there is a table and the table did not change. A structure is cheap in a good year and it is the only thing you have in a bad one.
The comp cycle starts two to four weeks after calibration closes. The gap is not for the comp work, which is mostly a table applied to a record and takes an afternoon. The gap exists to make it impossible for a level conversation and a money conversation to be the same conversation.
During the gap, the calibration record is closed and nobody has a number. The record gets to be true before it gets to be expensive. Then the band check against the market, once, for everyone; then the zone table applied; then four checks across the whole set — same level and zone, demographic spread, tenure drift, and the top-of-band list — and anything they find is fixed at the level of the structure, never person by person. Then exceptions, written and named and fewer than one in twenty. Then the letters, all in one week, so nobody hears their number from a colleague.
If the four checks find nothing, the structure is doing its job. If they find something, fix the structure. If you find yourself fixing a person, stop.
The conversation is not a negotiation
The manager has the number, the band, the zone and the record. The conversation restates the level, which the person already heard three weeks ago; names the band, which they can read for themselves; gives the number and the zone and the reason in one sentence; and then points forward at the leveling axes, which are the thing the person actually controls.
Three things not to say. "I fought for you," because there was no fight, the table was applied, and saying there was one teaches them that next year they need a manager who fights harder. "This is the best I could do," because it is what the structure produced and the place to disagree with the structure is the annual band check. And anything conditional about next year's money, because a conditional promise about money is a promise.
The answer to "I think I should be paid more" is "Which zone do you think you are in, and which axes put you there?" It returns the conversation to the record, which is the only place it can be settled.
The money, separately
The level is settled. Now the number.
The essay makes the case that pay has to be its own document. The workbook is that document: one band per level across every lane, four zones inside a band and what moves a person between them, one market check a year, three defensible answers on geography, equity that follows the level rather than the negotiation, a cycle that runs after calibration closes, and the four queries to run before a single letter goes out.
Written to be opened during a cycle rather than read once: on the page, as a PDF set to print, and with the templates as a spreadsheet.
Or read section two, free.
v1. The bands and the cycle timing are the ones I have run at companies of 25, 250 and 2500 people; the mapping onto the six-level shared core is new. One payment, every revision by email.
A companion to the four essays on leveling — design, engineering, product and research — and to what a performance review is actually for, which makes the three-documents argument first. The band widths and zone structure are conventional; hanging them off a lane-neutral six-level spine with a single differential per lane is the part that departs from convention.
© 2026 Renato Valdés-Olmos