Sokros

Solutions · Workforce Matrix

The right person, provably.

Recruitment screening and workforce development on one evidence-linked record: judged against what you wrote, decided by your people.

Signals land with a person. No automatic penalty exists anywhere in the system.

01 · Recruitment

Screened against what you wrote, not a vibe.

A CV pack is evidence like any other. It is read into typed fields, run through your written criteria as recorded rules, and handed back as a trail.

The pack, read

Typed fields from the CV pack, each with its source.

CVs, certificates and right-to-work pages become the fields your screening criteria actually check. No field is filled without the line, page or cell that earned it, so a shortlisting decision quotes the pack itself.

years_supervisory4CV · p.2
cert_expiry.forklift2026-03-01cert_register.xlsx · table 3
right_to_workdocument consistentpassport page

Screening for Meridian BioTech · role R-114 · every field quoted.

The rule trail

Identity as recorded answers.

Enrolment records, ID pages and observed footage read into one rule trail: name match, document consistency, photo match, submission behaviour. Four rules, four recorded answers, and a decision a caseworker can reopen a year later.

Identification · candidate C-2041
Name matches enrolment recordpassport page · learner file
Document internally consistentfonts, MRZ checksum, issue dates
Photo matches submitted IDstill from observed presentation
Submission behaviour in rangedevice, hours, draft cadence
Verified · rule trail stored4 of 4 · reopenable

02 · Workforce development

Capability, cell by cell.

Judged against the framework you wrote down, per person, per attempt. Each cell is a judgement with its evidence attached: the capability as your framework words it, the attempt that evidenced it, and the quote or frame that shows it. A gap is named the same way.

Capability matrix · role R-114
Operates lift equipment to the written standardcert_register.xlsx · expires 2026-03-01Renewal due
Leads the site induction briefingobserved · attempt 2Evidenced
Responds to a propped fire doorsite_walkthrough.mp4 · 03:12 · debrief filedEvidenced
Completes the escalation reportattempt 1 referred · gap namedGap named
Every judgement quotedattempt-keyed

Frameworks, written

The matrix reads from a framework you author as configuration: capabilities, descriptors, evidence rules. A new framework year is a new version. Nothing resets, nothing retrains.

Attempt-keyed

Every observed task, assessment and renewal is keyed to its attempt, so growth is measured against the person's own record, and a referred attempt names its gap.

A person signs

The matrix proposes, with quotes. Managers and assessors sign, and every sign-off lands on the record beside the judgement it approved.

03 · Integrity signals

Signals with receipts, not scores without reasons.

Every indicator names what it saw and quotes where it saw it, so your panel argues over evidence it can point to.

Authorship consistency

Writing that changes register mid-cohort is worth a look. Style and structure are compared across a person's own attempts, never against strangers' work.

Submission provenance

Document metadata, template history and revision trails are read alongside the content. A script that arrived through an unexpected route is flagged with the reason attached.

Reference credibility

Cited sources are checked for being real, dated and plausible: the classic tell of outsourced or generated work, caught with the specific citation quoted.

AI-usage analysis

Indicators of generated text are reported as evidence-linked signals for your assessor's judgement, never as an automatic penalty or a single opaque percentage.

Integrity signals · learner L-2041 · unit 5CO01
Authorship consistent with attempts 1 and 2
Reference credibility2 to check

“Harvard Business Review, 2031”: publication date implausible, citation quoted for review

Generated-text indicators
Signals recorded and routed to assessor review. No automatic action taken.

04 · A human decides

The decision stays with a person.

A human decides

Integrity outputs are signals for a qualified decision-maker. The system never penalises a candidate, a learner or an employee on its own.

Attempt-keyed history

Each attempt has its own immutable record, so comparisons run against the person's genuine baseline, not a guess about it.

Private by architecture

Signals are computed inside your deployment boundary, with zero third-party AI in the marking path. Nothing anyone writes trains a model.

These are standing commitments, not settings: read the commitments →

Know who you're certifying, hiring and growing.

One record for the person: screened at the door, verified at assessment, developed against the framework. Every judgement quoted, every decision human.

A 30-minute briefing: one role or one cohort scoped, your own packs screened in front of you before anything integrates.