Most CV tools show you a number and ask you to trust it. MyChapter takes the opposite position: a score you cannot inspect is a score you should not act on. Every score in the product arrives with an Evidence Line — a short statement of what was measured, how much of it there was, and what method produced it.
This page is the methodology behind that claim. It is written for users, not engineers, and it describes the system as it actually behaves.
The three basis labels
Every score is labelled with one of three methods:
- Deterministic — computed by code from your document. Same input, same number, every time. No AI model involved.
- Heuristic — computed by code from patterns that correlate with quality. Directionally useful, never precise.
- Data-informed — grounded in real outcomes, shown only when the sample is large enough to mean something.
If a score cannot be honestly labelled, it is not shown.
The five audit passes
The audit (/audit, or the Re-audit card inside the CV editor) runs five passes over your master CV.
1. Readability — deterministic
Flesch Reading Ease and Flesch–Kincaid grade level, average sentence length, passive-voice rate, and jargon density, computed from the text of your CV. The score rewards writing a recruiter can absorb on a first skim: shorter sentences, active voice, concrete words.
What it cannot tell you: whether your content is impressive. A CV can be perfectly readable and still thin. That is what the other passes are for.
2. ATS rules — deterministic
A checklist of the parsing rules applicant tracking systems actually enforce: file type and size, structure markers, contact-field presence, date formats, and similar. Where the binary layout cannot be inspected directly, the check says so and falls back to a text heuristic — the card tells you which.
3. ATS round-trip — deterministic
The strongest check we run. MyChapter exports your CV, re-extracts the text the way a parser would, and diffs it against what you intended. The fidelity score is the percentage that survives the round trip; character loss is reported separately. If a template scrambles your content for machines, this is where you find out.
4. Language & spelling — deterministic
An en-GB dictionary spellcheck plus code checks for repeated words, spacing slips and lowercase sentence starts. Each finding quotes the exact excerpt and offers a dictionary suggestion — never an invented one. Names and unusual terms can be false positives; check the excerpt before changing anything. This pass is deliberately not part of the overall blend, so scores from before it existed stay comparable.
5. AI-tell — heuristic, guidance only
Burstiness (sentence-length variance), cliché frequency, specificity, and paragraph rhythm. If you have added a voice profile, a voice-match estimate is computed against your own writing samples.
Two hard rules apply here. This is guidance, not a verdict — heuristic signals over a short document are noisy. And MyChapter will never help you “beat” an AI detector: the answer to robotic writing is writing that sounds like you, which is what the voice profile is for.
The overall score
The headline number is a fixed blend of readability, ATS rules + round-trip, and AI-tell. The weights are constants, shown in the codebase, and the blend excludes the language pass (see above). It is a quality indicator, not a success estimate — it tells you the document is well-formed, not that it will get you hired.
Fit, relevance and tailoring
When you tailor to a job description, the JD is parsed into must-haves, nice-to-haves and keywords, and your evidence is matched against it. Matched requirements and gaps are shown side by side. Gaps are never filled by invention — the tailor can acquire (you do the work), reframe truthfully, or accept the gap. Every generated line is checked against your source facts by a locked guardrail before it reaches you.
Success estimates and peer benchmarks
Your personal rates (applied → interview → offer) are computed from your own logged outcomes, labelled data-informed, and shown with the sample size. Fit-band and role-family rates appear only once a bucket has at least 3 applied roles. Anonymised peer comparisons appear only when the cohort clears a minimum size. Below those thresholds, MyChapter shows nothing — a small honest answer beats a large confident one.
What no score can tell you
Whether you will get the job. Hiring depends on timing, competition, and judgement calls no document controls. MyChapter's promise is narrower and more durable: your CV is truthful, machine-readable, and written in your voice — and every claim the product makes about it shows its working.