Skip to content
TwishivPlatforms

Worked example

Reading CVs against a brief

Four applicants for one role at the same fictional company that receives the invoices and sends the quotes in the other examples. Two of them score badly. Only one of them deserves to.

A recruiter with sixty applications and an afternoon does the same thing every time: skims for keywords, keeps the ones that look close, and moves on. It is not carelessness. It is arithmetic. The cost is that a good engineer with a terse one-page CV loses to a weaker one who wrote more.

Software is genuinely good at this part — reading every CV properly, against every requirement, without getting tired at application forty. What it is bad at is the thing it is usually sold for: turning that reading into a verdict on a person.

So this example does the first and refuses the second. It reads all four CVs against the brief, shows the evidence for every line, and hands a recruiter a shorter job than they started with. It never rejects anyone.

  • Evidence or nothing

    Every verdict quotes the line it came from. Where there is no line, it says so rather than inferring one.

  • Silence is not a no

    A requirement the CV never mentions is recorded as unaddressed, not as absent. These are different facts and they are kept apart.

  • The score shows its working

    Weights, statuses and arithmetic are all on the page. A percentage nobody can check is worse than no percentage.

  • The decision stays with a person

    The system reads and sorts. It does not reject anyone, and there is no threshold at which it starts.

The two to compare

Rahul scores 46%. Vikram scores 37%. Those numbers look like the same kind of answer and they are not. Vikram’s CV names its whole stack, and the things the role needs are not in it — he stays at 37% even if every question goes his way. Rahul’s CV is one page of skills with no detail, so most of what he is missing is information rather than ability, and he reaches 93% on the same test. Open them both.

Try it

Four applicants, one brief

Synthetic applications to Meridian Industrial Components Pvt Ltd. Pick any of them — every requirement is shown with the line of the CV it was judged on, and the questions the system would not answer for itself.

The brief

Senior Data Engineer

Meridian Industrial Components Pvt Ltd · Bengaluru · Hybrid — three days on site · Contract, 12 months · 6+ years

Must have · weighted 3

  • Python
  • Advanced SQL
  • Snowflake
  • Airflow
  • AWS
  • Bengaluru, hybrid

Nice to have · weighted 1

  • dbt
  • Streaming
  • Manufacturing domain
  • Engineering degree
  • Mentoring juniors

Applicants

3 needing a person

Sorted as received, not by score. Ranking the queue by a number is how the number stops being checked.

Priya Raghavan

8 years · Data Engineer at a logistics platform · Bengaluru

Outcome

Ready to submit

87%

of the brief is evidenced in the CV

Another 9% is unmeasured — the CV does not address it either way. If every open question came back well this candidate would reach 96%.

How this number is made: each must-have counts 3, each nice-to-have 1. Evidenced scores full, partial scores half, and anything not evidenced scores nothing. That is the whole calculation — it is arithmetic over the table below, not a model’s opinion, and it is not a validated prediction of whether this person would do the job well.

Every requirement, and what the CV actually says

The verdict and the evidence for it sit in the same row on purpose. A conclusion you have to go somewhere else to check is a conclusion nobody checks.

  • Python

    Technical · must have3 / 3Evidenced

    Asked for: 5+ years, production pipelines

    “Built and maintained ingestion pipelines in Python serving 40+ downstream tables”— Senior Data Engineer, Kestrel Logistics, 2021–present
  • Advanced SQL

    Technical · must have3 / 3Evidenced
    “Owned the warehouse modelling layer — window functions, CTEs, incremental merges”— Kestrel Logistics, 2021–present
  • Snowflake

    Tools · must have3 / 3Evidenced
    “Migrated the warehouse from Redshift to Snowflake over two quarters”— Kestrel Logistics, 2022
  • Airflow

    Tools · must have3 / 3Evidenced

    Asked for: authoring and operating DAGs

    “Author and on-call owner for around 120 Airflow DAGs”— Kestrel Logistics, 2021–present
  • AWS

    Technical · must have3 / 3Evidenced

    Asked for: S3, Glue, Lambda

    “S3, Glue and Lambda across the ingestion estate”— Kestrel Logistics; previously Arcwell Systems
  • Bengaluru, hybrid

    Location · must have3 / 3Evidenced

    Asked for: three days on site

    “Bengaluru. Open to hybrid.”— Header of the CV
  • dbt

    Tools · nice to have1 / 1Evidenced
    “dbt for all transformation models, with tests in CI”— Kestrel Logistics, 2022–present
  • Streaming

    Technical · nice to have1 / 1Evidenced

    Asked for: Kafka or Kinesis

    “Kafka consumers for shipment telemetry”— Kestrel Logistics, 2023
  • Manufacturing domain

    Domain · nice to have0 / 1Not demonstrated
    “Logistics and warehousing data across both roles”— Whole CV

    The CV names her sectors and manufacturing is not among them. Close adjacency is not the same as the thing asked for, and counting it would be the exact unearned match this page argues against.

  • Engineering degree

    Education · nice to have0 / 1Not addressed

    The CV does not address this — no evidence either way.

    The CV does not address this either way.

  • Mentoring juniors

    Experience · nice to have0 / 1Not addressed

    The CV does not address this — no evidence either way.

    The CV does not address this either way.

20 of 23 weighted points evidenced = 87%

What to ask before deciding

Generated from the gaps above rather than from a template, so each one is a question this CV actually raises. Every item here is something the system declined to assume.

  1. Your CV does not list a qualification — what did you study, and where?
  2. Your CV does not describe mentoring juniors — have you held that responsibility?

What would go to the client

Assembled from the rows above, so it cannot say anything the evidence does not. Where there is no evidence for a line, the line says so instead of being filled in.

Priya Raghavan — Senior Data Engineer

  • 8 years' experience — Data Engineer at a logistics platform
  • Python (5+ years, production pipelines sought): Senior Data Engineer, Kestrel Logistics, 2021–present
  • Advanced SQL: Kestrel Logistics, 2021–present
  • Snowflake: Kestrel Logistics, 2022
  • Airflow (authoring and operating DAGs sought): Kestrel Logistics, 2021–present
  • AWS (S3, Glue, Lambda sought): Kestrel Logistics; previously Arcwell Systems

Advocacy note

  • Proven experience. 8 years, most recently Data Engineer at a logistics platform.
  • Relevant skills. Python, Advanced SQL, Snowflake, Airflow, AWS
  • Outcomes. From the CV: “Built and maintained ingestion pipelines in Python serving 40+ downstream tables”
  • Validation needed. 2 items to confirm at screening.
  • Why review. Evidence score 87% against the brief, rising to 96% if the open items confirm.

Audit trail

What the system said, and what the person decided afterwards. In screening this matters for a reason the other examples do not have: if anyone ever asks why a candidate was not put forward, the answer should be a record rather than a recollection.

  1. 2026-09-24Brief readsystem11 requirements identified for Senior Data Engineer
  2. 2026-09-24CV readsystem9 of 11 requirements addressed
  3. 2026-09-24Scoredsystem87% from demonstrated evidence, ceiling 96%
  4. 2026-09-24Ready to submitsystemEvery must-have evidenced. 2 minor items still worth confirming.

Synthetic data. Meridian Industrial Components Pvt Ltd, the role and all four applicants are invented — these are not anonymised real CVs — and the requirement reading is pre-computed for this page rather than produced by a model.

Being straight about it

What is real here and what is not

Real

  • The weighting, the scoring arithmetic and the thresholds
  • The four evidence states, and the rule behind each exception
  • The screening questions, generated from the gaps
  • The submission summary and advocacy note, built from the evidence
  • The audit trail

Not real

  • The company, the role and all four applicants — invented for this page, not anonymised from real CVs
  • The CV reading — pre-computed here, not produced by a model
  • Any connection to an ATS or a job board

And one thing worth saying plainly

A percentage next to a person’s name is not a measurement of them. It is a count of how much of one brief their CV happens to evidence, and CVs are written unevenly by people with unequal amounts of practice at writing them. Screening software that hides that arithmetic, or lets it decide, discriminates quietly and at volume.

That is why nothing here auto-rejects, why the unmeasured share of every score is shown next to the score itself, and why a blocked candidate still carries the reason and the evidence. If we built this for you, that is the shape it would keep.

All three examples run on the same engine. The rules, the exception model, the approval routing and the audit trail are shared code — only the policy on top differs. That is the actual offering: not a shelf of products, but one capability pointed at whichever process is costing you most. Invoices and quoting are the other two.

What is your team reading by hand?

CVs, claims, applications, tenders, support tickets — anywhere a person reads the same kind of document against the same criteria all day is worth an hour of conversation.