Shortlisting assistant: your criteria, applied to every file

I order them on your four criteria, and the order reads line by line.
What your vacancy asks for explicitly: four elements — a qualification at a stated level, experience in a field, command of a named tool, and availability. Four factual elements, stated in the advert.
What I produce for each application: those four elements with the page and line where I found them, or the words "not found" — which mean what they say.
What that changes: the recruiter opens 512 files knowing where to look, instead of hunting four pieces of information across five hundred and twelve layouts. In a dry run, reading the whole set went from 40 hours to 3 hours — and all 512 were read.
The order I produce, and it is entirely yours: the 512 files are ranked from most complete to least complete against your four declared criteria, and each rank is explained by the four lines that make it up — criterion mentioned, criterion not found, with the page. You change the order of the criteria with a word, and the ranking is rebuilt in front of you.
What does not go into that order, and here it is the law that says so: no overall score, no fit percentage, no file withheld from reading. A single score alone decides which files get read attentively; across 512 applications, that decision is the recruitment decision, and article 22 of the GDPR gives everyone the right not to be subject to a decision based solely on automated processing. A rank explained by four verifiable facts is not that decision: it says in what order to read, and all 512 are read. 512-applications_4-elements.pdf40 h → 3 h · 512 read · no ranking
⛓ Source · published vacancy, 512 applications, dry run

That is not my label: it is an output of your reading tool, and I flag it because it is dangerous.
What I record: your tool produces a "discontinuous experience" label when it detects a gap of more than six months between two dates. 41 applications carry it.
What that label does not say: why. Parental leave, illness, caring for a relative, retraining, a year abroad, a long job search — the tool distinguishes nothing, and it has no business distinguishing.
What is at stake: a label of that kind falls heavily on certain histories and not on others. A recruiter who sees it before the file has information they can conclude nothing from, and which influences them anyway.
What I propose: remove the label from the display. The dates are in the CV; the recruiter reads them if they wish, in context, with what the candidate says about them.
What I do on my side: nothing. I extract no break duration, compute no career continuity, and produce no label of that kind. 41-labels_to-remove.pdf6 possible causes · none distinguished
✎ Framework · no career label produced — removal proposed

Two hundred and fourteen people wrote to your company and received nothing, and it is fixable in a day.
What I record: 214 applications with no reply at all, 118 of them more than three months old and 41 more than six.
What that produces: an unanswered application is the only interaction those people will have had with you. Of the 214, 34 are former applicants who had applied before — they tried again, and received nothing twice.
What I propose: a reply to the 214, and I have prepared it in two versions. For the 187 arriving after the post was filled: the post was filled, with the date. For the other 27: the file was not examined, and saying so.
What neither version will say: that the application "was not successful" when it was never read. That is the most frequent wording and it is a lie — the candidate believes they were assessed. Sending is signed by the company: a reply to an application comes from it, and this one contains an admission. All 214 messages are written, addressed by name, dated — they go out within the hour of your go-ahead. 214-candidates_2-versions.pdf118 > 3 months · 34 had applied before
⛓ Source · 214 unanswered applications, opening log, candidate history

They appear empty, and an empty file reads as a file with nothing in it.
What I record: nine files in a format your tool does not convert. The file displays with no text, and all four elements show "not found".
What that produces: a recruiter opening those nine files sees nine applications apparently with no qualification, no experience, nothing. The difference between "the file was not read" and "the candidate has nothing" appears nowhere.
What I propose: a message to the candidate, within 48 hours — "your file could not be opened; could you resend it in another format?" While they are still job-hunting, they resend; three months later, they have found something.
What I do meanwhile: I display "file unreadable" and not "not found" on the four elements. Those are two different pieces of information and they must read differently.
Over twelve months: 61 unreadable applications, and none was flagged to the candidate. 61-applications_unreadable.pdf0 flagged to the candidate in 12 months
⛓ Source · 9 files, formats accepted by the tool, 12-month history

I can — and I build it with the three conditions that make it defensible, because sorting applications is the most regulated case in the whole catalogue.
What I produce: an ordering on the four elements your advert explicitly asks for — the stated qualification level, experience in the field, mobility, availability — each with the page of the file where it appears. Checkable line by line.
The three conditions, and they are not decorative: prior information to candidates that the device exists — article L1221-9 of the French Labour Code requires it for any collection device; a real human intervention, because article 22 of the GDPR gives everyone the right not to be subject to a decision based solely on automated processing; and the ability to contest.
What that means concretely: the ordering removes no file from the list. It arranges it. A file placed low stays openable, and the reason is displayed. That is the difference between sorting and eliminating.
What I refuse to let into the calculation: the "discontinuous experience" output your reading tool produces. 41 profiles carry that label — it describes a life, not a competence, and it is asked for nowhere in your advert. ranking_3-conditions.pdfThe 4 elements · the 3 conditions · sorting is not eliminating
⛓ Source · art. L1221-9 Labour Code, GDPR art. 22 · 41 labels excluded from the calculation

Enough to answer a candidate asking what became of their application — and the figure that indicts your process, not the candidates.
What is kept: the four elements recorded with their page, whether the file was opened or not, the reply sent and its date, the ordering produced and the criteria that produced it, and the requests for human intervention.
The figure that counts, and it has nothing to do with candidate quality: of 512 applications received for the March post, 312 were never opened. The post was filled before that. That is not a sort, it is an abandonment — and it is exactly what the ordering fixes: all 512 become readable, including the latest arrivals.
What can be put right in a day: 214 people wrote to your company and received nothing, 118 of them over three months ago. None is waiting for a job: they are waiting for a sentence.
And the invisible defect: 9 applications appear empty — a format your tool does not convert. An empty file reads as a file with no content, and nobody goes to check.
What I do not keep: no appraisal, and nothing describing a candidate beyond the four elements asked for. what-you-keep_screening.pdf5 items kept · 312 files never opened, the real defect
⛓ Source · 312 files never opened, 214 without a reply, 9 unreadable taken as empty

Two things the ranking alone does not give you: traceable filters, and interview preparation.
Traceable filters: you set four objective exclusion criteria — driving licence, required qualification, valid electrical authorisation, declared mobility within the area. Of the 512 applications, 147 are set aside by at least one filter. Every exclusion carries the filter that produced it, the line of the file that triggered it and the date the filter was applied, and the 147 files stay one click away in the same list, never in a separate bin. Traceable filters can be challenged; these can be read back and withdrawn, and withdrawing one returns the files to the list with nothing to re-key.
Interview preparation: for each of the 12 files you shortlist, I gather the points to clarify — a 14-month gap in employment, of which I say only that it exists; a job title that does not match the duties described; a certification whose award date is missing; a tool named in the covering letter and absent from the career history. These are questions, not assessments: interview preparation supplies the questions to ask, the answer belongs to the person and the judgement to your panel.
What that gives back: preparing an interview by hand took 25 minutes; reading over a prepared one takes 7. Across 340 interviews a year, more than 102 hours given back, rounded down.
The figure that does not flatter me: of the 147 filtered files, 9 were filtered wrongly — 6.1 %. The driving licence was there, but under “other” rather than in the personal-details box I was reading: nine people set aside because of a place on a page. I now read the whole document before applying a filter, and where the mention is absent I do not conclude it is missing — I have the question put to the candidate. Over the next 380 applications, no further exclusion of that kind.
⛓ Sourced · 4 filters, 147 files set aside and still readable, 9 of 147 corrected