Context Theory
Questions a business actually has to decide.
One question per page. The answer comes in the first forty words, the substantiation after it, and every figure carries the source it came from and the date it was last checked.
A page publishes when it holds something that would be wrong on its sibling — a figure scoped to this subject, or a constraint an operator inside the trade could correct you on. That is the whole editorial standard, and it is enforced by the build rather than by a reviewer, because a gate a person applies is a gate that stops being applied somewhere around page sixty.
Where to start.
Not the newest, and not a shortlist anybody typed. Each of these is the most cited question in its part of the corpus — cited by the other pages that had to reference it to make their own argument — with one taken per topic, so the set is not six versions of the same problem.
- How fast should you respond to an online lead? The honest answer is a number, not a range, and almost nobody hits it.
- What data should you never put into an AI tool? Anything you hold on someone else's behalf, and anything you could not explain having sent to a third party.
- How do you check whether an AI answer is correct? Check the specifics against a source, not the reasoning against your intuition. Fluent reasoning is the least informative part.
- How should you handle a customer complaint? Fast, in one conversation, with the decision made by whoever is speaking to them. Delay is what turns complaints into reviews.
- What should a small business track about its enquiries? Five fields, captured at arrival. Everything a business needs to fix its intake comes out of them.
- How should an AI agent be given a job? State the finished condition, the material, the boundary and the report. Describing the steps instead is what produces drift.
The whole set is listed in full, every question with the date its figures were last checked.
Ways in.
The same body of work, entered from different directions. Answers are the layer everything else is built out of; the rest are the questions asked with a constraint attached — an industry, a market, a system, a number you need for yourself.
| Surface | What is in it |
|---|---|
| Answers | The general case, with nothing attached — the question as any business meets it. |
| Markets | What changes when the same trade runs in a named metro, and which local record decides it. |
| Industries | How acquisition actually works in a trade — the channels, the response window, the stack. |
| Solutions | Where AI and automation change the work — follow-up, qualification, recovery — each defined by what it will not do. |
| Calculators | Your own number, before you have spoken to anyone, with the arithmetic shown. |
| Benchmarks | Published figures reconciled against each other — definition, population, basis, sample. |
| Provenance | One circulating claim, followed back to a primary source or to a stated dead end. |
| Research | The standard a study here is held to. No study has published yet, and the index says so. |
What a source line is for.
A number with a citation beside it is not the same as a number somebody checked. The provenance pages exist to show the difference: each takes a figure that circulates everywhere, follows it back through whoever repeated it, and stops either at the study that produced it or at the point the trail dies. One is reproduced whole here rather than summarised, because the argument this site makes about evidence is not worth making in the abstract.
The claim, as it circulates: “Multiply a base wage by the standard loaded-labour multiplier and you have what the employee actually costs your business”
It comes from a monthly advice column for entrepreneurs, written by Joseph G. Hadzima Jr., a senior lecturer at the MIT Sloan School of Management and reprinted on an MIT server under his own copyright. The column states a range covering basic salary, employment taxes and benefits, names no dataset, no sample and no measurement window, and treats the figure as a planning heuristic. The single point figure in general circulation does not appear in it: it is the midpoint of a range whose upper bound is higher than the one the column states.
| Step | Publisher | Attributes it to | Status |
|---|---|---|---|
| 01 | TrueTools, a free employee-cost calculator for the US | the Internal Revenue Service's employer tax guide and the U.S. Bureau of Labor Statistics, in a source line at the foot of the page | |
| 02 | Sunbytes, an offshore engineering-team provider | the US Bureau of Labor Statistics Employer Costs for Employee Compensation series, with a sample size attached to it | |
| 03 | U.S. Bureau of Labor Statistics | its own quarterly survey of employer establishments. It publishes benefit cost as a share of total compensation, and no multiplier on a wage anywhere in the release | |
| 04 | U.S. Small Business Administration | nobody | |
| 05 | BeeBole, a timesheet product, in a real-cost-of-an-employee guide | Joe Hadzima, a Senior Lecturer at MIT — the only document in this trail that names anybody who can be walked to | |
| 06 | Joseph G. Hadzima Jr., MIT Sloan School of Management, in the column Starting Up: Practical Advice for Entrepreneurs | its own worked example. No dataset, survey, sample, population or measurement window is named anywhere in the column | Primary source |
The trail in full, with every source linked and the drift between them recorded: Where the loaded-labour multiplier actually comes from. Also traced: Where the five-minute rule actually comes from . Where two published figures disagree about the same thing, the reconciliation is in the benchmarks.
Where the answer changes.
Most of the corpus is general on purpose — an enquiry that waits two days was handled badly whatever the business sells. These are the places it is not general. Each row is the question that industry changes the answer to, picked because that hub is the only one the answer references.
| Industry | Where the answer changes | Markets |
|---|---|---|
| Healthcare | How fast should a medical practice reply to a patient enquiry? | 2 |
| Home services | Should you answer calls outside working hours? | 1 |
| Legal | What does a legal lead actually cost? | 1 |
| Manufacturing | Why do manufacturing enquiries take so long to close? | 1 |
| Real estate | How long do real estate agents take to reply to a lead? | 4 |
The hubs themselves — the operating environment, the stack, where each one leaks — are at industries, and a market page sits inside the hub it belongs to. The systems those hubs point at are at solutions, and the calculators that give you your own version of the number are at calculators.
What this is not.
Limits
-
will not tell you what to do in your business. A reference does not know your numbers, and a page that pretends to is guessing in a confident font.
will give you the measurement that would tell you, and say what it costs to collect.
-
will not stay true forever. Sources move, figures are restated, and a page written this month is a page written this month.
will print the source and the date beside every figure, on a re-check cadence, and mark it stale on the page when it passes.
Every page names who wrote it, who reviewed it and when. That record is published rather than asserted — it is the editorial log, and every entry in it says what the review actually changed. Context Theory is one person, Siddharth Sharma, and the standard everything here is held to is how I work.
Context Theory also takes client work. There is one thing to buy — a growth audit that measures your own response times against three competitors — and it is priced, bounded and described on the audit page. None of the pages above were written to sell it.