Methodology
How the numbers are made.
Two claims here matter enough to explain: whether you are ready for a role, and whether an employer has sponsored before. Both are simpler than they look, and the parts that fail are parts we chose.
None of this is intelligence.
Employer matching is one string-normalising function and a hash lookup.
Readiness is a keyword parser feeding an ordered list of rules. No model produces a requirement, an evidence match, or a verdict. Saying so plainly is more useful than the alternative, because it tells you exactly how much weight each answer can carry.
Where the data turns out not to support a claim, we delete the claim rather than soften it, and we have removed a shipped feature on exactly those grounds. Where we could not determine something from our own code, this page says so instead of filling the gap.
Role readiness
Readiness is a label from a closed list of seven, never a score.
There is no percentage, and none is derived anywhere: ready now, ready with minor tailoring, ready with interview preparation recommended, achievable stretch, not ready yet, hard constraint conflict, and insufficient requirements.
Where the requirements come from
A posting’s requirements are found by matching patterns against its text, plus a fixed list of 71 tool names. That is a keyword parser, so what you see is always the requirements a parser could extract from the text we happened to have, never the employer’s real hiring bar. Postings can state requirements we could not parse. At most 36 requirements are kept per posting, and anything past that cap is dropped without being shown to you.
What counts as your evidence
Evidence comes from your structured Persona records and from capabilities you have confirmed in your own words. Anything whose provenance names a model is dropped before matching, not down-weighted. Matched evidence is scored by how directly it was demonstrated, where it appeared, and how recent it is, with older work counting for less: full weight within two years, decaying to just over half beyond eight.
The word we are not using
Our matcher has a mode the code calls “semantic”. It is not. It means two or more shared words after a hand-written table of eight verb families. Nothing is embedded, learned, or inferred by a model. We are naming this because “semantic” on a page like this would be read as something far stronger than what runs.
What unconfirmed evidence can reach
Text-matched evidence you have not confirmed is capped below the threshold for a moderate rating, so it can never rise above weak. That is arithmetic, not a tendency. Structured facts you have explicitly confirmed — work authorisation and visa status especially — are held to a stricter standard again: they need an exact, current, fill-authorised record, not a legacy alias or a confirmation attached to an older value.
What it refuses
A requirement our parser flags as needing interpretation is returned as missing and never matched at all, so it fails toward “not ready” rather than assuming a match. An unrecognised verdict falls back to “not ready yet”. And the match model states its own limit directly: it does not produce a pseudo-scientific percentage.
What you should not conclude
A readiness label is about the evidence we could match against the text we could parse. It is not a prediction that you will be interviewed or hired, and we do not claim it correlates with either. If your essential requirements rest on capabilities you have described in conversation rather than demonstrated in your record, the highest label reachable is “ready, interview preparation recommended” — “ready now” is structurally out of reach, by design.
Matching an employer to filing records
A sponsorship badge means one thing: the employer name on the posting normalised to the exact same key as an employer in the government data.
Nothing looser earns a badge.
How a name becomes a key
We lowercase, expand & to “and”, drop punctuation, and strip legal suffixes such as Inc, LLC, Ltd and Corp. We deliberately do not strip descriptive words, because doing so collapses genuinely different companies into one another. A short curated alias table maps a handful of brand names to legal names; an alias is a fact we assert, not a guess.
Exact match, or nothing
Only an exact key match produces a badge. Near matches are offered separately as candidates you can inspect, and are never used to assert sponsorship on your behalf. Searching a partial name shows you possibilities; it does not claim any of them filed anything.
What we cannot see
A company that does sponsor can show no record, because it files under a legal name or a subsidiary that does not normalise to the name on its job posting. An acquisition can leave history under the acquired entity. A badge is scoped to the exact legal entity we matched, not to the corporate family — so a figure on one arm of a large company will not reflect another arm. Absence of a record is not evidence that a company does not sponsor, and we say so wherever a record is missing.
The cost of that choice, in the open
Roughly half the companies whose boards we track produce no sponsorship page at all, because no exact substantive record exists under the name we hold. They appear as plain text in our company directory. We would rather show nothing than attach one company’s filing record to another company’s name.
The posting always outranks the history
When a posting states that sponsorship is not offered, that governs the role, regardless of how much filing history the employer has. History is evidence about a company, never a promise about a job.
Where the data age is, and is not, shown
Every public company page prints its sources, the government data release, and the date we compiled it. The sponsorship badge on a job card does not: it carries the figures and the caveat that history is not a guarantee, but no data age. That is a gap in the surface most people see, and it is ours to close rather than yours to infer.
Sources
H-1B petition outcomes come from the USCIS H-1B Employer Data Hub.
Green-card certifications come from the U.S. Department of Labor OFLC PERM disclosure data. Both are public U.S. government records. Every public company page states which release it is showing and when we compiled it, so a figure can always be dated.
Where every other answer comes from, and the whole path from a resume to a sent application, is on How it works.