The Two-Times Rule: What the Width of a Delivery Estimate Admits

Ninety-seven floors, ninety-seven ceilings

VIT MARKET, the services arm of the VITON13 platform, attaches a delivery estimate to every one of its 100 listings. Ninety-seven of those estimates name two numbers — a floor and a ceiling. Three name a single leading quantity and hide the rest of the story in a trailing clause. Not one listing is sold without a stated time.

That completeness makes a measurement possible that a scattered market normally refuses. VJOURNAL took every duration string, stripped its stopwatch prefix, read the first quantity and its unit, and divided the ceiling by the floor. The result is a width: the factor by which the slowest promised outcome exceeds the fastest one. To make sure the figure was not an artefact of the parser, the same 100 strings were re-read a second time straight out of the raw catalogue file with an independent regular expression. Both passes return the same numbers, and those numbers are the subject of this article.

The width matters because it is the only thing in the catalogue that discloses uncertainty about time. A price band discloses uncertainty about money, and a description discloses scope. The gap between one and four hours, or between fourteen and twenty-five working days, is the seller telling you how much of the schedule they are unwilling to promise.

Forty-three listings land on exactly double

Across all 100 services the median ceiling sits at exactly 2.00 times the floor, and the mean at 2.05. Those two statistics agreeing that closely is already a hint. The distribution confirms it: 43 listings set their ceiling at precisely twice their floor — 4 to 8 hours, 3 to 6 hours, 2 to 4 days, 15 to 30 hours, on and on. Sixty-seven listings admit at least a doubling. Ten admit a tripling or worse. The single widest band in the catalogue is a factor of four.

A quantity that lands on a clean multiple in 43 of 100 independent cases is not being measured. It is being drafted. VJOURNAL's reading, offered as interpretation rather than fact, is that the doubling is a house habit — the shape an estimator reaches for when asked to express ordinary uncertainty, in the same way a builder says 'two, maybe three weeks'. That does not make it dishonest. It makes it a convention, and conventions should be read as conventions rather than as calculations.

The width table, read aloud

Here is every width figure in the catalogue, stated in prose. Across all 100 listings: lower quartile 1.80, median 2.00, upper quartile 2.00, minimum 1.00, maximum 4.00. By service family, the median ceiling-to-floor ratio runs 2.00 for web development, 2.00 for SEO and content, 2.00 for automation, and 1.83 for AI implementation. Four families, three identical medians and one that is barely lower.

The catalogue's own difficulty labels — written in Russian as 'Новичок', 'Нужен опыт' and 'Дорого и сложно', roughly beginner, experience needed, and expensive and complex — tell the same flat story. Beginner work, 21 listings, median width 2.00. Experience-needed work, 50 listings, median width 2.00. Expensive-and-complex work, 29 listings, median width 1.83. Difficulty moves the number by less than a fifth of a step, and it moves it downwards.

That is the first genuinely counter-intuitive result in the data. If the width of an estimate expressed how risky a job was, the hardest tier would carry the widest bands. It carries the narrowest.

Relative width is flat; absolute slack triples

The flatness disappears the moment the ratio is replaced with a subtraction. Measuring slack as ceiling minus floor, converted to hours at the dataset's declared eight-hour day, the three difficulty tiers separate cleanly: a median of 3 hours of slack for beginner work, 15 hours for experience-needed work, and 32 hours for expensive-and-complex work. The same split by family gives 8 hours for SEO and content, 8 for automation, 13 for web development, and 40 for AI implementation, against a catalogue-wide median of 16.

The AI family is the sharpest illustration. Its median width, 1.83, is the narrowest of the four — and its narrowest individual band still contains 24 hours of unpromised time, three full working days. A ratio of 1.83 sounds tighter than a ratio of 2.00 right up to the moment you notice what it is 1.83 times.

So the two ways of measuring a range answer two different questions. The ratio asks how confident the estimate sounds. The subtraction asks how much of your calendar is unaccounted for. Only the second one can be planned against.

The widest bands belong to the shortest jobs

Ten listings quote a ceiling at least three times their floor. Every single one of them has a floor of exactly 1 — one hour, one day. There is no eleventh case, and there is no wide band anywhere in the catalogue that starts from a larger number. Sorted by their floor, listings that start at 1 have a median width of 2.00 and a maximum of 4.00; listings that start at 5 or more have a median of 1.83 and never exceed 2.67. The correlation between the floor value and the width is negative, at -0.25.

The mechanism is arithmetic, not risk. When the smallest sensible unit is one, the next value up is two, and the ratio has already doubled before anyone has thought about the work. Nine of those ten widest rows are the same idiom — 'one to three days' — and the widest row in the entire catalogue is an emergency layout repair quoted at one to four hours for $50 to $250. Nothing about that listing is riskier than a fifty-hour build. It is simply short enough that integers are coarse.

The practical consequence is blunt: a buyer comparing two suppliers on the tightness of their percentage ranges is mostly measuring how long their jobs are. Ratios are not comparable across scales, and any procurement rule built on them will systematically punish whoever quotes the smallest tasks.

The money band is always the wider one

Now put the time band next to the price band on the same listing. In 99 of the 100 services the price range is wider than the time range. In zero services is the time range wider. One service ties, and only because both of its published bands collapse to a single parsed value. The median price ratio is 3.33 against a median time ratio of 2.00.

The mean price ratio is 8.80, but that figure should be ignored and is quoted here only so that nobody quotes it elsewhere: it is dragged there almost single-handedly by one listing, a template sold on a marketplace at $19 to $10,000, a spread of 526 times. The median is the honest statistic for this distribution.

A seller who is three times less certain about the money than about the clock is telling you something about where the variance in their business actually lives. VJOURNAL's reading is that duration is bounded by the process — the steps exist, they take what they take — while price is bounded by the buyer, the scope they eventually ask for, and what the work turns out to be worth to them. If that reading is right, then the timeline is the more reliable half of a quote, and the number most likely to move after signature is the invoice.

What a range cannot tell you

This is a catalogue of intentions, not a delivery log. There is no record anywhere in the repository of where inside a band a real engagement landed, how often a ceiling was breached, or by how much. Every figure above describes what was promised, and none of them describes what happened. An estimate cannot be audited by the document that contains it.

The bands are also silent about their own interior. A floor of 15 hours and a ceiling of 30 says nothing about whether the typical job finishes at 17 or at 29, and there is no weighting, no most-likely value, and no distribution published anywhere. Readers should resist the temptation to treat the midpoint as an expectation; this article uses midpoints only where the arithmetic requires a single representative value, and says so each time.

Three further limits belong in plain sight. The three single-number listings are not fixed promises — only the leading quantity is parsed, and their real range lives in a trailing clause such as 'one day of diagnosis plus one to three days of implementation'. The eight-hour day used to put day-quoted work on the same axis as hour-quoted work is a declared modelling constant, chosen by the analysis and not observed in the catalogue; the 25 AI listings quote no hours at all, so every hour figure for that family passes through it. And nothing here establishes cause: the data cannot say whether an estimator widened a band because a job was genuinely volatile or because doubling is simply how the sentence gets written.

Reading the next estimate you are handed

Four habits follow from the measurements, and each of them is a direct consequence of a figure above rather than general advice. Budget against the ceiling: 67 of 100 published bands admit at least a doubling, so the floor is the number that gets quoted and the ceiling is the number that gets scheduled. Ask what moves the job from one end to the other, because in this catalogue the answer is frequently written down already — the qualifier clauses name dependencies like the number of fields, the size of the database, or whether a parser already exists.

Compare slack in hours rather than in percentages when you are choosing between suppliers, since a 2.00 width on a two-day task and a 2.00 width on a five-week project are different promises wearing the same number. And treat a clean doubling as a drafting convention until the seller can name the variable behind it; in 43 cases out of 100 here, the ceiling is exactly twice the floor, which is a very large coincidence for a hundred independently estimated jobs.

A range is not a weaker version of a date. It is the most informative sentence in a quote, provided it is read for its width and not just for its lower number.

Decision framework: how to read a project timeline range

A timeline range should expose uncertainty in decisions, dependencies and review, not merely offer an optimistic and pessimistic date. Its width becomes actionable only when each boundary has stated assumptions.

Schedule uncertainty is also a cost driver. The earliest date usually assumes available content, prompt approvals and stable integrations; the latest should name the events that consume contingency.

Request a three-point schedule with earliest, expected and protected dates, then attach owner, dependency and exit evidence to every milestone on the critical path.