Model wildcard-v2.0
The wildcard carries no score. That is the point.
Ranks one to ten answer a single question: who scores highest against the public rubric? The eleventh entry answers a different one, so putting a number on it would be dishonest.
The problem
A scored wildcard is just an eleventh entry that scored slightly lower.
We checked our own corpus before changing the policy. Across 270 published rankings, 268 wildcards sat within half a point of the provider at number ten. The wildcard was supposed to be the pick the rubric could not see. Scoring it had quietly turned it back into the bottom of the same list.
So the score is gone. From wildcard-v2.0 onward the wildcard is unrated on every list we publish, and every list already published has been migrated.
What replaces the score
A reading, not a number.
Each wildcard is read against 20 signals. Four to six of them fire on any given provider, each with one line of specific evidence, plus a plain statement of who the bet is right for and who it is wrong for. Nothing is averaged into a composite, because averaging is what produced the problem.
None of these signals appear on any other rankings site. That is not a claim about our cleverness. Most of them simply were not computable until a machine could read an entire market's documentation, changelogs, review corpora, job ads, status pages and community threads in a single pass, and hold all of it in mind at once.
The 20 signals
Money
What it really costs, in money and in work.
Impact density
What does one unit of the outcome I actually want cost here, versus everywhere else?
- Reads
- Pricing pages, contract and overage terms, and customer accounts of what they got for what they paid, normalised into one comparable unit of outcome across seat, usage, project and retainer pricing.
- Why now
- Vendors in the same category price on incompatible units on purpose. Reconstructing a shared unit of outcome means reading the whole market's commercial language at once, not comparing two price tables.
Quality-per-price percentile
Is this the best thing available at its price, regardless of where it ranks overall?
- Reads
- Every provider's measured quality signals plotted against its real price band, then the provider's position on the quality curve within that band alone.
- Why now
- Ranking sites rank absolutely. The buyer with a fixed budget needs the leader of their price band, which is a different question and never gets published.
Price integrity under growth
What happens to my bill when my usage doubles?
- Reads
- Published tiers, overage clauses, minimum commitments and renewal accounts from customers who scaled, rebuilt into a cost curve rather than a starting price.
- Why now
- The punishing part of a contract is usually in clause language and renewal anecdotes, not the pricing table. Reading both together across a market is new.
Lock-in cost
What will it cost me to leave, in work rather than in fees?
- Reads
- Export fidelity in the docs, data portability, notice and termination terms, plus migration accounts from customers who actually left.
- Why now
- Exit cost is spread across API docs, legal terms and departure stories. Nobody assembles it because it takes reading all three for every vendor in a category.
Service
What happens after the sale, once you are not a new logo.
Time-to-first-value drift
How far is the promised onboarding time from the one customers describe?
- Reads
- Vendor onboarding claims against dated customer accounts of when the thing first did something useful.
- Why now
- The gap only appears when marketing claims and dated review language are read as one corpus, per vendor, at scale.
Support half-life
How fast does the quality of help fall off once I am no longer a new customer?
- Reads
- Support experiences in dated reviews and community threads, split by how long the customer had been paying.
- Why now
- Published SLAs describe intent. Half-life describes behaviour, and it is only visible by ageing thousands of accounts by customer tenure.
Effort transfer
How much of the work does this actually take off me, and how much does it hand back?
- Reads
- Implementation and maintenance burden in the docs, plus the job ads customers post to operate the tool after they buy it.
- Why now
- The clearest evidence that a product creates work is that its customers hire people to run it. Nobody has read the job market as product evidence before.
Founder attention proximity
How close do I sit to the people who actually build this?
- Reads
- Who answers in public across forums, issue trackers, community channels and social: the founder, an engineer, or a support tier.
- Why now
- Attribution of public answers to role and seniority across every channel a vendor operates is a language problem, not a metrics problem.
Incident candour
How does this vendor behave on its worst day?
- Reads
- Status page history and post-mortems: how fast they admit, how specific they are about cause, and whether the fix was named.
- Why now
- Uptime percentages are published. Honesty during failure is prose, and prose is now readable at scale.
Truth
Whether the story the vendor tells matches the record.
Promise-to-practice gap
Which marketing claims does the documentation actually support?
- Reads
- Every capability claim on the marketing site, matched against the documentation, API reference and release notes that would have to exist for it to be true.
- Why now
- This is claim-by-claim semantic reconciliation between two large bodies of text. It was not possible to do it honestly, per vendor, before.
Documentation truth ratio
How much of the documentation is current and correct rather than aspirational?
- Reads
- Doc freshness dates, deprecated endpoints, examples that no longer match the current API, and gaps where a documented feature has no reference.
- Why now
- Stale documentation is the most reliable early signal of a product losing attention, and reading a full doc set for currency is new.
Review authenticity weighting
What do the reviews say once the incentivised and templated ones stop counting?
- Reads
- Review corpora weighted by how much each review reads as lived, specific experience rather than prompted, reciprocated or templated text.
- Why now
- Star averages treat every review as equal. Weighting by the texture of the writing itself is only possible with language models.
Churn language signature
Why do people leave, as opposed to how many complain?
- Reads
- The structure of leaving accounts: what broke the relationship, at what stage, and whether the same fracture repeats across customers.
- Why now
- Negative review counts are noise. The recurring shape of a departure is signal, and it only emerges from reading the accounts as narratives.
Buyer regret window
When in the relationship does dissatisfaction usually show up?
- Reads
- Dated customer accounts mapped to tenure, to find whether pain clusters in week one, month six, or at renewal.
- Why now
- Requires ageing a review corpus by customer lifecycle stage rather than by publication date.
Trajectory
Where the product is actually going, not where the roadmap says.
Roadmap honesty index
Of what they publicly promised, how much shipped?
- Reads
- Two years of public commitments against changelogs, release notes and shipped documentation.
- Why now
- It means holding every past promise in memory and checking each one against the record. No human maintains that ledger for a whole category.
Hiring signal trajectory
Where is this product actually going in the next twelve months?
- Reads
- The vendor's own job ads: which functions are being staffed, which are being quietly wound down, and what that implies about the roadmap.
- Why now
- A company tells the truth about its priorities in the roles it pays for, long before it says anything in public. Reading that as a product signal is new.
Sustained attention
Is the specific thing I need still being worked on, or just the headline product?
- Reads
- Release, commit and documentation activity scoped to the individual capability the buyer cares about, not to the product overall.
- Why now
- Overall activity hides abandoned corners. Scoping activity to a single capability across a market requires reading what each change actually touched.
Fit
Whether the difference is an advantage, and for whom.
Ceiling distance
How long before I outgrow this?
- Reads
- The shape of complaints at each customer-size band, to find the size at which the product starts to strain.
- Why now
- The ceiling is described anecdotally by customers who hit it, never by the vendor. Locating it means clustering complaints by customer size.
Category fit anomaly
How different is this from the category's dominant architecture, and is the difference an advantage?
- Reads
- The structural pattern every other provider shares, and where this one departs from it, then whether that departure serves a definable buyer.
- Why now
- Identifying a category's implicit shared architecture, then measuring distance from it, is only tractable when the whole category can be read at once.
Under-the-radar coefficient
How much better is this than its share of voice suggests?
- Reads
- Measured quality signals against market awareness: search demand, mentions, review volume and category presence.
- Why now
- This is the wildcard's defining metric. It requires a quality read that is independent of popularity, which is precisely what a scored rubric cannot give you.
What this model is, and is not
It is an editorial judgement, made by machine. The same engine that writes the rankings reads the wildcard against these signals and states what it found. It is labelled as such on every page. It is not a measurement taken from an instrument.
Only supported signals are published. If the evidence for a signal is not there, the signal does not appear. A wildcard read with three signals is a wildcard read with three signals, not one padded to six.
A wildcard can fail to qualify. When the model concludes a provider is simply a slightly weaker version of the ten rather than a genuinely different bet, that list is flagged for a new wildcard rather than dressed up with a signal read.
No provider can buy this slot. The wildcard is the single most valuable position on any list for an unknown provider, which is exactly why it has never been and will never be for sale.
Anyone may use this. The model is published as the Unrated Wildcard Standard (UW-1.0), free to adopt under CC BY 4.0. Name it, link to the spec, and your eleventh slot stops being a lie too.