Ranking methodology

Transparency Index

Answers one question: how openly does this company run its hiring process? It is not an employer rating and not a measure of speed — a fast, opaque process should not outrank a slower, candid one, which is why response time is deliberately excluded.

Components

Four rates, each computed by the same weighted-rate function the rest of the product uses. Weights are equal at 25% each — any other split would assert a relative importance that no evidence in this dataset supports.

• Salary disclosure · weight 0.25

  Share of candidate reports where the salary range was disclosed before the final stage — in the posting, before the first round, or before the final round.

• Outcome clarity · weight 0.25

  Share of candidate reports that reached a definite outcome — an offer or a rejection — rather than no response at all.

• No unpaid assignment · weight 0.25

  Share of candidate reports with no unpaid work request recorded. Higher means the process asked less unpaid effort of candidates.

• No silent drop-off · weight 0.25

  Share of candidate reports that did not end in silence — the last interaction gap was under 30 days, or the process reached a definite outcome.

How the score is calculated

Each component is a weighted share of eligible candidate reports. The index is the weighted average of the four, expressed out of 100. Only reports from candidates count: an employee or former-employee report never went through the hiring process, so it is excluded from eligibility rather than diluting the rate.

When a company is not ranked

A company is ranked only when all four components independently clear 5 effective reports. Averaging whichever components happen to be available would make the index mean something different for each company while presenting them in one ordered list — the most misleading thing this ranking could do. Companies that do not qualify are shown as “not enough data to rank”, never as a low score.

“Effective reports” is a weighted count, not a raw one: external evidence carries less weight than a first-party report, and no single source can dominate a company's evidence base.

Ties and updates

Ties break on effective report count, then alphabetically — so a company's position is identical between two identical page loads. Rankings are recomputed on every request from current moderated evidence; there is no cached or periodically-frozen table.

Algorithm version: transparency-index-v1

Improvement Index — “Most improved”

Answers: has this company's candidate-reported experience actually improved? It compares the Transparency Index across two 6-month windows — the same composite, measured twice — so there is one methodology rather than two.

A company cannot mark itself improved

This is the property the ranking lives or dies on. Improvement is computed only from what candidates reported afterwards. Marking an issue resolved, recording a process change, publishing a company response, or paying for anything cannot move the number — those values are not reachable from the calculation at all.

A recorded process change is still useful: it explains why a metric moved and gives a date to look either side of. But it is company-reported context shown next to a CandidateVoice-observed measurement — never an input to it.

When improvement is not reported

Both windows must independently qualify. A company with many recent reports and almost none before them has not demonstrated improvement — there is nothing to compare against, and treating a thin baseline as one is how a ranking invents progress. Movements smaller than 3 index points are reported as no meaningful change, because sampling noise alone moves a four-part composite by a point or two.

Regressions are measured the same way and are not hidden — but only genuinely improved companies appear on the Most Improved board, since a board with that name containing flat or worsening companies would be misleading.

Algorithm version: improvement-index-v1

What cannot influence a ranking

• —Paying CandidateVoice. Ranking modules cannot reach payment, sponsorship or billing data.
• —Claiming a company profile, or publishing a company response.
• —Marking an internal issue resolved, or recording a process change.
• —Adding employee or former-employee reports — they are excluded from hiring-process eligibility.
• —Verification tier. How a report's origin was verified is never a score input.
• —Synthetic, QA or demo data. Internal organisations are excluded from every ranking, and the B2B demo is a code fixture with no database rows at all.

Companies can pay for diagnostics, dashboards, benchmarks and workflow tools. They cannot pay for a number, and they cannot remove or suppress a legitimate report.