The evidence standard
Every number on the page names the source it came from.
Three kinds of evidence, never mixed, each labelled on the panel where it appears, including the panels that have nothing to report and say so.
Every panel names its lens before you ask, and whether or not the answer helps us
01The three lenses
Three questions. Three different kinds of answer.
Each lens is authoritative for one thing and worthless for the others, so the panel always says which one it used.
What the market asks for.
Which skills are employers hiring for right now?
Live job postings, description analysis output, extracted skills, validated role archetypes. 345,595 postings, 274,051 analysed.
Authoritative for current demand. Never for how a particular institution performs.
What the literature says.
Where is this sector heading over five years?
Secondary research carried with page-level citations, never paraphrased loose and never restated as a live figure.
Authoritative for context and forecasts. Never treated as live market demand.
What this university does.
What do our students actually learn, and become?
Your curriculum files, employer and alumni survey responses, and the career footprint of your own graduates.
Authoritative for your programmes and your graduates. Never generalised to a sector.
A dean's first question is always “says who?” Nexus answers it by design, not by exception.
02One question, three lenses
The same question, asked three ways, gives three correct and different answers.
Should our MBA add a data analytics specialisation? Here is what each lens says.
| Lens | The answer it gives | What it cannot tell you |
|---|---|---|
| Market demand | Data Analysis is named in 22,174 postings and SQL in 18,602. In the top ten most-requested skills across the corpus, two are analytical and seven are competencies. | Whether your graduates are already getting these roles without the specialisation. |
| Research evidence | Sector forecasts on where analytical work is heading over a five-year horizon, cited to the page. | What employers are hiring for this quarter. A forecast is not a vacancy. |
| Institution evidence | Which of your courses already evidence these skills, what your alumni went on to certify in afterwards, and what your employer survey respondents said. | Anything about the wider market. Your graduates are not a sample of it. |
A platform that returns one number here has decided on your behalf which question you meant. The decision it makes will be the one that produces the most confident-looking dashboard.
03The labelling rule
Four labels. Every panel wears one, including the unflattering ones.
This is what makes the doctrine checkable from inside the product, whether or not the answer helps us.
Read from the corpus now
Drawn directly from the governed corpus at the timestamp on the panel. This is the only label that supports a demand claim.
Real, but thin
Live data with known coverage gaps for the slice you are looking at. Shown and labelled, rather than quietly padded out to look complete.
A sample, and it says so
Demonstration data used while a tenant is being set up. It never represents your institution, and the label stays until your own data replaces it.
The empty panel
Where there is nothing to report, the panel says nothing to report. This is the label that costs us the most, and the reason the other three can be trusted.
Every partner dashboard opens with a plain-language disclosure that sources are mixed and that each section names its own. A second control lets an administrator mask compensation and named-alumni detail, so a screen can be shown in an open meeting without exposing an individual graduate.
04Who approves
The analyser drafts. A person decides.
In a procurement review, that is the difference between a recommendation that can be approved and one that cannot.
| The question an assessor asks | What a self-deciding system answers | What Nexus answers |
|---|---|---|
| Who approved this curriculum change? | The system recommended it. | A named person, on a dated record. |
| Can you reproduce last year’s figure? | The model has been updated since. | Yes - model and prompt version are stamped on all 578,750 runs. |
| What happened to the uncertain cases? | It made a best guess. | They queued. 88,276 are waiting for a person right now. |
| Where did this number come from? | A blended score. | One lens, named on the panel, with its source. |
Two-thirds of what the analyser sees does not clear the confidence bar for automatic use, and we publish that. The alternative - a system that always returns an answer - is not more accurate. It has moved the uncertainty somewhere the buyer cannot inspect.
05The second lens, in full
The second lens is not a gesture. It is a read library.
Research evidence is the easiest lens to fake, because nobody checks a citation in a slide.
Published sector research, ingested
Reports are loaded as documents and processed, not summarised from memory. Each one stays attached to the claims it supports.
Named publisher types
Global consultancies, multilateral institutions, national and regional skills councils, and job platforms.
Region-tagged, like everything else
Global, India, UAE and GCC. A global workforce forecast and a Gulf sector skills framework are not interchangeable, so they are not stored as if they were.
We could name every report in the library and it would read impressively. These are third-party published documents, and listing which ones we hold raises a licensing question that deserves a real answer before it becomes marketing copy. The corpus is described by publisher type and region until that is settled. Ask us in a walkthrough and we will show you the library.
06The audit trail
What survives the meeting: a record, not a recollection.
- 01
Per claim
The lens, the source and the timestamp
Carried with the figure wherever it goes, including into a deck.
- 02
Per run
Model and prompt version
Recorded on all 578,750 analyser runs, so any output can be reproduced or its drift explained.
- 03
Per decision
The named human
Who confirmed, reassigned or dismissed an item, and when.
Explore Nexus
Five capabilities, one evidence base.
Each one is a different question about the same corpus. They share a vocabulary, so a finding in one holds up inside another.
07Before you ask
The questions this page usually provokes.
Why not just combine the three into one score?
Because they answer different questions and a blended score answers none of them. "Is this skill in demand now?", "where is this sector heading?" and "what do our graduates actually do?" have three different right answers, and a single index hides which one you were given. Every platform that blends them does so because it makes a cleaner dashboard, not a truer one.
What if the market data disagrees with the research?
Then you are shown both, labelled, and the disagreement is the finding. It usually means one of two things: the sector is turning, or the corpus is thin for that slice. Either way it is more useful than a number that has quietly split the difference.
What does "Data not available" actually mean?
Exactly what it says - we have nothing worth showing for that slice. It is the label that costs us the most in a demo and it is the reason the other three can be trusted. A dashboard that is full at every zoom level is telling you something about its padding, not about your institution.
Can we see the underlying records?
Yes, down to the postings behind a demand figure and the records behind an alumni figure, subject to the masking controls your administrator sets. A claim you cannot open is a claim you cannot defend in front of an assessor.
Does this apply to the consulting work too?
It applies to it more strictly, because a consulting recommendation carries further than a dashboard panel. Every slide in an engagement deliverable carries the lens it was built on - which is the specific thing an opinion-led strategy deck cannot do.
Bring the number you cannot currently source.
Forty-five minutes. We will show you which lens answers it, and what it would take.
We reply within 48 hours. +91 88796 54341