How every Phontix AI number is built
Anyone can hand you a market size or a price. We hand you its derivation: the inputs, the method, the sources and how much weight each part can bear.
Methodology version 1.0, September 2026
The standard
If a user cannot click a number and see every input, formula, assumption and source that produced it, the number does not meet platform standard.
The Phontix AI platform rule
A number nobody can audit is not intelligence. It is an assertion. Three commitments follow from that rule, and they shape every dataset we build.
- Sources are named
- Every data point carries the register, filing or listing it came from. If we could not source it, we mark it as a gap instead of filling it with an estimate.
- Assumptions are visible
- Forecasts expose the variables driving them, and subscribers can change them. A model you cannot argue with is a model you cannot defend.
- Revisions are logged
- When our view changes, the change is recorded with a reason, so you can explain a decision made on last quarter's data.
One value, one owner
Every figure is a specific combination of dimensions, such as a molecule in one indication, market and year, or a chemical grade in one market and month. Each carries its own source, method and confidence. Totals are always calculated from those cells and never typed in separately, so a figure on one screen cannot quietly disagree with the same figure on another.
Uncertainty compounds rather than averages. A chain that rests mostly on modelling declares itself, instead of borrowing confidence from the registry data around it.
How each product grades evidence
The principle is shared. The grading suits the data each product works with.
LOE Edge
Evidence tiers set by rule, not judgement.
- Tier 1
- Primary record from the issuing authority
- Tier 2
- Secondary or partial, needs adjustment
- Tier 3
- Proxy or modelled, never the sole basis
Price Fundamentals
A confidence tier on every layer of a price.
- Reported
- Sourced directly from a named, checkable record
- Derived
- Calculated from reported inputs
- Modelled
- Built from stated assumptions
- Analog
- Forecast benchmarked on a comparable cycle
- Gap
- No reliable source, shown not filled
Brand Listener
A signal tier on every reading.
- Verified
- Directly observed on an accessible platform
- Estimated
- Inferred from a partial sample
- Gap
- Needs a paid licence, shown as a gap
Stage gates
Each stage is frozen before the next one uses it, so a late change is visible rather than silently rippling through.
Scope
What is being measured, for which markets and period.
Method
The method assigned to each figure, with the reason.
Sources
Named sources identified before any extraction.
Extraction
Data pulled, with gaps surfaced rather than patched.
Build
Cells calculated, totals computed from cells.
Review
Checks, cross-method validation and sign-off.
What we will and will not put our name to
Most of what makes an estimate auditable is discipline about what not to claim.
- Name the source behind every input, down to the document
- State what a source cannot tell you, alongside what it can
- Show the calculation in full to any client who asks
- Flag an assumption as weak even when it flatters the opportunity
- Say plainly which parts of the platform are live and which are not
- Present a modelled figure as though it were measured
- Buy, store or resell data licensed to someone else
- Quote a time or cost saving we have not measured with a client
- Infer a competitor's filing where the record does not support it
- Cite a dataset in paid work without the licence that permits it
Challenge the method on a call
Bring the number you trust least in your current model. We will show how we would build it.