Fiscora's career guides teach the visible game: how to research a band, anchor a number, evaluate equity, price freelance work, read a severance agreement. That advice is sound and it operates at a stage most candidates never reach.
This article is about the stage before it — and about the moment at the other end, when access to work is withdrawn by a system that owes you no explanation.
The gate before the interview
Application screening is now overwhelmingly automated. Resume parsers, keyword matching, assessment scoring, ranking models, and in some markets video analysis. These produce determinations of material consequence — access to livelihood — and they routinely satisfy every element of the register's most serious governance harm: material consequence, automated determination, no specific reasons, no appeal to a human with authority to reverse.
The register calls it algorithmic governance without due process: consequential decisions become unappealable because they became automatic. Hiring is the largest instance most people encounter, and the least visible, because a rejection is indistinguishable from not being chosen. You cannot tell whether a person read your application and preferred someone else, or whether a parser could not read your PDF.
What you can do is narrow, and it is not nothing:
- Ask whether the decision was automatedIn several jurisdictions this changes your rights. Ask regardless — the answer, or the refusal, is a fact.
- Ask for the specific reasonNot "we went with another candidate." Which criterion, at which stage. Most employers will not answer; some will, and the asking costs one email.
- Ask whether a human reviewed the rejectionReview and authority to reverse are different things, and worth separating in the question.
- Treat repeated silent rejection as a systems problemIf you are rejected at the same stage repeatedly with no human contact, the fault may be in machine-readability rather than in you. That distinction matters psychologically and practically.
The attribution trapThe most damaging effect of unexplained automated rejection is not the rejection. It is that the candidate has no way to distinguish "I was not suitable" from "the system did not parse me," and defaults to the first. A structural failure gets absorbed as a personal one, silently, at scale.
Labour disassembled until it is deniable
The second harm is labour fragmentation and invisible work: human labour is disassembled until it is deniable, then denied. The sequence is the argument. Work is broken into tasks small enough that no single one looks like employment; the tasks are mediated by a platform describing itself as a marketplace; and the resulting relationship is denied to be employment because it no longer resembles it.
The duty attaching to this is disclosure-shaped and, read from the worker's side, gives you three things to demand by name:
| What is owed | What it means in practice |
|---|---|
| Disclosure of human involvement | Where output presented as automated is materially human-produced, that must be stated. If your work is being sold as an algorithm's output, the buyer is being misled and your contribution is being erased. |
| A stated pricing rule | How your pay is calculated, not just what it came to. A rate that varies by an undisclosed rule is not a rate. |
| Appealable deactivation | Removal from the platform is loss of income. It should give reasons and permit appeal to a human who can reverse it. |
The third is the one that matters most and is most often absent. Deactivation is typically framed as an account action governed by terms of service — a private matter between a company and a user. Framed as what it actually is, it is a determination of material consequence made about a person's livelihood, usually automatically, usually without specific reasons.
Our position is that the framing is the mechanism. Calling it an account action rather than a determination is what removes the expectation of reasons. Insisting on the second framing, in writing, is the single most useful thing a platform worker can do — not because it always works, but because it establishes what kind of thing is being disputed.
The rating that became the employer
The third harm is metric colonisation, and in work it has a specific life cycle worth naming:
- A rating is introduced to inform clients. Reasonable: buyers want signal.
- The rating begins to allocate work. Ranking in search, eligibility for jobs. Still defensible.
- The rating becomes a threshold. Below a number, you are removed. The measure now governs employment.
- The rating becomes the job. Workers optimise the metric over the work — accepting only safe jobs, avoiding difficult clients, managing the score rather than doing the task well.
The governing duty here is metric honesty, and its second limb is the operative one for workers: do not couple a single metric to an automatic material consequence without human review. A rating threshold that triggers automatic deactivation is precisely that coupling. It is worth naming in any appeal, because it identifies the defect rather than merely disputing the outcome.
It is also worth understanding what a rating cannot see: a client who was unreasonable, a job that was misdescribed, a delay caused by the platform itself. Those absences are systematic rather than random, and they fall hardest on workers who take harder jobs.
Ẓulm: dependency meeting unaccountable power
The governing instrument uses a term for the situation where these converge — Ẓulm, defined as oppression, and specifically as dependency meeting unaccountable power. Both halves are required. Power you do not depend on is tolerable; dependency on power that is accountable is ordinary employment.
The definition is useful because it identifies which lever exists. Where you cannot reduce the dependency, the only remaining move is to increase the accountability — reasons, appeals, records, escalation. Where you can reduce the dependency, that is usually the stronger play, and it is the practical case for the thing career guides already recommend: more than one client, more than one platform, a portable record of your own work.
Severance is the same structure at the end of the relationship. The instrument's consent-integrity duty states that consent obtained when a person cannot practically refuse — mid-transaction, at a moment of financial need, or as a condition of receiving something already owed — is not consent. A release of legal rights, signed while unemployed, on a short deadline, without independent advice, is close to a textbook instance. That does not make it void, and knowing the category is what tells you to slow down and get advice.
Building a record they cannot delete
The final harm is digital amnesia and its inverse: a civilisation that records everything preserves nothing, and forgets nothing it should. In work it appears as an asymmetry — the evidence you would need to defend yourself disappears when your account closes, while adverse records persist indefinitely.
The practical response is to hold your own copy:
- Export your work history periodicallyCompleted jobs, ratings, client feedback, earnings. Platforms rarely provide this after deactivation, which is when you need it.
- Keep artefacts outside the platformWork samples, correspondence establishing scope, anything demonstrating what you actually delivered.
- Log adverse events when they happenDate, what occurred, who said what. Memory is not evidence; a contemporaneous note is close to it.
- Ask what is retained about you, and for how longWhere a subject-access right exists, use it. Where it does not, ask anyway.
Honest limits
Most of this does not work the first time. Asking a platform for the specific reason for a deactivation will usually produce a template. The value is cumulative: a documented pattern of unanswered requests is the raw material for a regulator, a journalist or a collective claim, none of which is available to someone with no record.
Individual technique is a poor substitute for collective standing. Everything here is a workaround for an accountability gap. Historically the things that closed such gaps were organisation and regulation, not individual skill at writing letters. Teaching the workaround is worth doing and should not be mistaken for the fix.
Automation removed real prejudice too. The steelman we would accept: discretionary human hiring was frequently arbitrary and discriminated openly, and structured assessment reduced some of that. The argument here is for reasons and appeals, not for returning to a manager's unexamined judgement.
And rights here vary more than in any other category in this estate. Employment protection, platform-work classification and rights over automated decisions differ enormously between jurisdictions and are changing quickly. The structure generalises; the leverage is local, and you should check yours.
Revision log. Sep 2026: first publication. Written as the mastery track for the career desk's structural issues, referenced from the "More Details" section of Fiscora's career guides. Reviewed annually.