A number is only publishable if the method that produced it is publishable.
This page is the standing definition of how a NoxOrigin research measure has to be written, collected, and cleared. It is published now, while there is nothing to excuse. A method written after the result exists is a rationalisation, not a methodology.
No measure on this programme has been collected. Every definition below is marked unresolved where it is genuinely unresolved, because pretending a definition is settled is the first and easiest way to produce a number that means nothing.
What has to be fixed before collection starts
A measure is not a metric name. It is a definition with five parts, all settled in advance and none of them revisited once numbers exist.
The unit of observation
The thing being counted. An invoice, a payment allocation, an enquiry, a completed project, a stock item. If two people defining the measure would pick different units, the measure is not ready.
The population and its boundaries
Which businesses and which of their records are in scope, in what geography and business type, and what the population deliberately does not include. A measure without a stated boundary describes everyone and therefore describes nothing.
The exclusions and the lost data
What is removed, and what is known to be missing. Silent exclusion is the most common way a real dataset produces a flattering number, so exclusions are stated as part of the definition rather than left to the analyst.
The statistic and the shape of its spread
Median, distribution, rate, or share — chosen because it suits the measure, not because it is the friendliest headline. Averages over skewed operational data are usually the wrong instrument and are never the only one reported.
The window and the sample floor
The minimum number of contributing businesses and the minimum period of data, both decided from what the measure actually needs rather than from what is convenient to publish. Where a measure is seasonal or cycle-dependent, the window has to cover the cycle, not a convenient slice of it.
The candidate measures, and what is still open
Each measure is written as it currently stands, with the unresolved question named rather than smoothed over. None has been collected, and none carries a value.
Invoice payment delay
The elapsed time between an issued invoice and the payment that settles it in full, expressed against the invoice's own credit terms rather than as a raw calendar gap.
Requires: Invoice issue date, payment allocations, credit terms, cancellation and write-off states.
Unresolved: Whether the clock stops at first payment or at settlement, and how invoices that never settle are represented instead of quietly dropped.
Partial payments per invoice
The count of settlement events against a single invoice, where each payment is an allocation to that invoice rather than an amount stored against it.
Requires: Payment records, allocation history, invoice balances over time.
Unresolved: Whether an advance before invoicing counts, and whether a single transaction split across two invoices is one partial payment or two.
Lead response distribution
The full distribution of first-response latency per enquiry, reported as a distribution rather than a single central value, with the clock starting at recorded enquiry time.
Requires: Enquiry timestamps, first outbound contact on any channel, channel and direction, working-hours definition.
Unresolved: Whether an automated acknowledgement is a response, and how enquiries received outside working hours enter the distribution.
Quotation-to-invoice conversion
The share of issued quotations that reach an issued invoice within the quotation's own validity period, with time-to-conversion reported alongside the share.
Requires: Quotation issue and revision history, validity dates, invoice origin link, lost and expired states.
Unresolved: How a repeatedly revised quotation is counted, and whether expiry is a conversion failure or a separate outcome.
Frequency of scope change
The rate at which agreed project scope is amended after delivery starts, counted from recorded change records rather than from recollection.
Requires: Baseline scope, change requests, approval records, delivery and rework entries.
Unresolved: The hardest part is informal extra work that never becomes a change record. The definition has to decide whether unrecorded work is absent or merely unobserved, and say so.
Reconciliation discrepancy
A reconciliation attempt that does not balance on first pass, classified by recorded reason code where one exists and by uncategorised where one does not.
Requires: Reconciliation sessions, expected versus counted or matched values, variance reasons, resolution time.
Unresolved: Whether the measure counts reasons, sessions, or resolution effort, and what a missing reason code does to the distribution.
Stockout frequency
The frequency and duration of periods in which an item was unavailable to sell against a level the business had reason to expect to be in stock.
Requires: Stock ledger, reorder points, availability state, lost-sale or unfulfilled-order records.
Unresolved: How a stockout is distinguished from a deliberate decision not to reorder, which is a judgement the data cannot make on its own.
Project margin distribution
The distribution of realised margin across completed projects, where realised is measured after approved change value and recorded direct and internal cost.
Requires: Quoted value, approved change orders, billed value, direct cost, recorded internal time, rework cost.
Unresolved: How internal time and rework are valued. That single decision moves the figure more than any other, so the method has to state the rate basis rather than absorb it.
Where data may come from
Three routes are legitimate. None has been run. Fixing the list in advance is the point — a fourth, looser route appears under deadline pressure, and it is always the same one.
Opt-in aggregation from live deployments
Operational records from businesses that have granted specific, written, revocable permission to contribute aggregates. Aggregates only, never per-customer figures, and never any of the business's own customers, contacts, or staff as data subjects.
Structured opt-in survey of operators
Self-reported operational practice, collected with a fixed instrument and reported as self-reported wherever it appears. Useful for things the software cannot observe, and weaker for everything it can, so the two are never blended into one figure.
Documented experiment on public datasets
Work on published official statistics, carried out as an explicit experiment with its own stated method and limitations. A public dataset can exercise and test the measurement pipeline before any customer data is involved, and its result is about public data, never about NoxOrigin's customers.
On public datasets: named at publication, not guessed at now
Public official statistics are a reasonable source for a documented experiment — national statistics releases, central bank banking and payments data, tax filing volume reporting, and similar. Each figure taken from a public source cites the publishing body, the release, and the period, with a link to the official publisher added at the point of publication. No public-dataset figure is published on this programme today, and no source is cited here that has not been opened and read.
Privacy and the consent basis
This is the part that decides whether a measurement is possible at all. A customer’s invoices, payments, customers, and staff are their data, not ours, and a product relationship is not permission to publish from it.
Permission is specific, written, and revocable
Contribution is agreed in writing, naming the measures involved, and can be withdrawn at any time. The permission record itself is kept, dated, and stateable. Consent to use NoxOrigin is not consent to research use, and general permission to publish a story is not permission to publish their figures.
The permission model already exists here
Customer stories on this site are gated by a granular permission model — separate approval for identity, logo, quote, screenshots, metrics, and links, each defaulting to withheld, with an explicit publication state per record. Research is the same problem with a narrower surface, and the same model is the right basis for it. Any research-specific grant has to be written and recorded before collection starts.
Aggregates only, never per-customer figures
No published figure identifies a business, and no figure can be reverse-engineered into one. Where a contribution is large enough to be identifying, it is not published in that form. Small samples are held back rather than rounded into usefulness.
The customers’ own customers and staff are out of scope entirely
No personal data about a participating business’s customers, contacts, or employees enters the dataset. That means no names, phone numbers, email addresses, tax identifiers, bank details, free-text fields, or message content. A measure that requires free-text analysis is a measure we do not run.
Withdrawal is a real right, not a formality
A contributing business can ask what was contributed, ask for it to stop, and is answered under the same rights that apply to any data we hold. If withdrawal would make a published figure unreproducible, that consequence is disclosed on the page rather than discovered quietly.
It sits inside the published privacy position
Research collection follows the same commitments described in our privacy policy — what is collected, why, how long it is kept, and the rights over it. If a study cannot be described consistently with that policy, it does not run.
What disqualifies a result
A disqualifier means the number is not published at all, on this programme or anywhere else. It is not a caveat, a footnote, or a chart labelled “illustrative”.
Records the business did not permit
Data contributed without specific written permission for the named measures, or contributed under a general permission that does not cover research. Consent to run a product is not consent to be measured.
A period when the behaviour was not yet happening
Measuring payment delay or scope change over a window in which most participants had not yet adopted the workflow produces a figure about software adoption, not about the business practice the measure claims to describe.
A sample or window below the stated floor
Publishing below the minimum sample or minimum collection window stated in the method, even when the result is more useful to us. A small result may be described as a small result on an internal page; it is not an industry figure.
One business presented as a population
A single deployment, however large, is that business. It becomes a case study with the customer's permission, or it is aggregated, but it is never described as typical.
Self-report presented as measurement
Figures recalled by an operator and presented in the same voice as figures read from records. Self-reported values are published as self-reported, in a separate series.
A number carried over from someone else
Vendor marketing figures, cited industry averages, and figures that cannot be traced to a named source with a stated method. Restating an unsourced number is how a site ends up confidently wrong.
A method that cannot be printed in full
If the definitions, exclusions, or corrections cannot be described on the page in full, the result stays internal. There is no abbreviated version for public release.
The publication gate
The same conditions apply to every measure, and all of them must be visible on the page carrying the number.
Method stated in full on the page
Definition, population, exclusions, statistic, and collection route, readable next to the result rather than summarised in a caption.
Sample size stated
The number of contributing businesses and the number of underlying records, both at or above the floor written into the method beforehand.
Collection window stated
The period covered, and confirmation that it spans the cycle the measure describes.
Consent basis stated
How permission was obtained, from whom, on what terms, and how it can be withdrawn. Where participants chose not to be identified, that is said rather than left implied.
Disqualifiers and limits stated
What was excluded, what is known to be missing, and the cases where the measure does not apply at all.
Independently reproduced from the extract
A second person recomputes the figure from the underlying extract before release. If the review is internal, the page says it is internal. We do not claim an external audit we have not had.
Every prior finding is rechecked, not carried forward
An earlier figure is restated when the method, population, or consent basis changes, with the revision stated. A number that is quietly restated without a note is worse than one that was never published.
How a finding would be reviewed
The honest description of the process, including what it is not. There is no independent research board, no external ethics approval, and no third-party audit behind these pages, and nothing on this site implies otherwise.
Definition review
Before collection, the measure’s definition, exclusions, and window are written down and fixed. Later changes to a definition are treated as a new measure version, not as a correction.
Consent check
Every contributing business’s written permission is confirmed on file for the specific measures before any extract is taken. One missing permission removes that business from the dataset, not a caveat from the page.
Recomputation
A second person runs the calculation from the extract without reusing the first person’s intermediate figures, so the arithmetic is checked rather than trusted.
Disclosure review
The draft page is checked for anything that could identify a participant, anything unsupported by the extract, and any language stronger than the evidence. If it cannot survive that reading, it does not ship.
Where this programme actually stands
Stage zero. No collection has run, no measure has cleared the gate, and no finding has been published. The next honest step is a documented experiment on a public dataset, which tests the measurement pipeline without touching a customer’s records. The step after that — opt-in aggregation from live deployments — depends on written permission that does not yet exist.
We are not publishing a date for the first finding, because no honest research plan can. The gate is the commitment. When it is met, the page changes; until then, this is the whole of it.