Skip to main content
CyberOS LEARN is the capability ledger — what each member can do, how good they are at it, and how that translates into promotions, bonus-pool weighting, and sabbatical readiness. It plays three distinct roles in the platform. As a skills catalogue, it maintains an evidence-backed, closed-enum skill tree across all disciplines. As the VP (Voting Power) engine, it computes a deterministic quarterly score per member that REW uses to weight the bonus-pool distribution. As the Hội đồng Chuyên môn (Specialist Council), it runs the structured peer-review workflow that makes every promotion decision transparent, human-driven, and defensible under EU AI Act Art. 14.
LEARN is a P1 module — currently in planned/design phase and not yet generally available. Features described here reflect the planned P1 design.

​
Skills catalogue

LEARN maintains a closed skill catalogue — members cannot invent free-text skills. Every skill belongs to a discipline, carries a 1–5 mastery scale, and requires concrete evidence before a level claim is accepted.

​
Disciplines and skill tree

Engineering

Languages, frameworks, infrastructure, security, and architecture skills. ~60 skills at P1.

Design

UX research, visual design, motion, and design systems. ~25 skills at P1.

Operations

Project management, process design, compliance, and people ops. ~25 skills at P1.

Business

Sales, partnerships, customer success, and strategy. ~20 skills at P1.

Finance

Accounting, FP&A, tax, and corporate governance. ~20 skills at P1.

​
Mastery levels and evidence requirements

Each level from 1 to 5 requires at least one piece of linked evidence. The system accepts four evidence types:
For level-4 to level-5 jumps, a council review is mandatory regardless of how many endorsements you have. The threshold is a versioned parameter — your manager can confirm the current requirement for your discipline.

​
Certification expiry alerts

LEARN monitors the expires_at field on every external certification and fires learn.cert.expiring events at 90, 30, and 7 days before expiry. The CUO/CHRO-skill agent surfaces these as proactive notifications to the member. Renewing a certification before it lapses preserves continuity of the mastery level it supports.

​
Voting Power (VP)

VP is a quarterly score that quantifies each member’s concrete contribution to the company. REW reads VP at quarter close to weight the BP bonus-pool distribution.

​
What VP measures

VP is a deterministic pure function of three input dimensions:
All weights are versioned parameters — changing them requires a three-way CEO + CFO + CHRO co-sign, and the change cannot retroactively re-score historical quarters.
Closed issues, their complexity tags, and the peer ratings attached to each contribution. A member who leads high-complexity work scores higher than one who closes an equal number of low-complexity tasks.

​
VP roll-up and REW integration

LEARN recomputes VP for all members monthly. The vp_snapshot table records each member’s total and per-discipline breakdown alongside the parameter_version_id used for that computation. A CI replay job asserts byte-identical output for the last 12 snapshots on every parameter change — if any snapshot drifts, the change is rejected.
VP roll-up is automatic and deterministic. You cannot manually override a VP score. If you believe an input signal is wrong (e.g. a PROJ contribution was miscategorised), correct the source record in PROJ or KB — the next monthly recompute will pick it up.

​
Hội đồng Chuyên môn (Specialist Council)

The Specialist Council is CyberOS’s adaptation of the Vietnamese academic peer-review format for professional promotion decisions. It ensures every promotion is backed by structured evidence and multiple independent judgements — not a single manager’s opinion.

​
Council workflow

1

Manager Nomination

Promotion cases are opened manually by a manager nomination or a member’s own submission. The system never auto-triggers a promotion case, regardless of how high a member’s VP score is.
2

Parameter Version Snapshot

At submission, LEARN snapshots the current promotion gate criteria (VP threshold, mastery requirements, minimum endorsements, time-in-level) into the case record. Even if parameters are updated mid-review, this case is judged against the criteria that were in effect when it was submitted.
3

Panel Assignment

HR/Ops assigns 3–5 peer judges plus a council chair. The system applies a diversity heuristic — panels should span at least two teams and two seniority levels. The CHRO confirms the final panel before deliberations open.
4

Individual Scoring (Private)

Each judge scores the candidate across five dimensions in a private workspace. Judges cannot see each other’s scores during deliberation — this mirrors the Vietnamese academic hội đồng format and prevents anchoring bias.
5

Aggregate Computation

Once all judges submit, LEARN computes the median score. The median is robust to a single extreme outlier on a small panel. The aggregate score and a recommendation (ADVANCE, HOLD, or REFINE) are written to the PROMOTION_OUTCOME row.
6

Outcome Forwarded to HR

HR receives only the {summary, recommendation} pair — no per-judge scores, no individual feedback text. This is a structural guarantee: HR’s database role lacks SELECT on the council_vote table in LEARN’s vault schema.
7

CEO Final Approval

The CEO reviews the outcome summary and makes the final promotion decision. The CEO can override the council’s recommendation, but must record a written rationale — the audit log distinguishes council-approved from override outcomes.
Per-judge individual scores never leave LEARN. This is the psychological-safety contract that makes honest peer review possible. HR’s access role is structurally blocked from reading individual judge data — only the aggregate outcome summary and recommendation are ever forwarded to HR.

​
Promotion gate criteria — example (L2 → L3 engineering)

EU AI Act Art. 14 — Human Oversight: The Specialist Council is the decision-maker for promotions, not an algorithm. LEARN’s computation (VP score, criteria checking) is advisory evidence. The human council deliberates, and the CEO issues the final decision. This design satisfies the human-oversight requirement for high-risk employment-decision systems under Annex III §4.

​
Career path visualisation

Every member can view their own skill progression and exactly what remains to qualify for the next level.
Your career-path dashboard shows each gate criterion for your target level, whether you currently meet it, and which specific evidence gaps remain. You can click through to the relevant PROJ issues, KB posts, or certification records that count toward each criterion.

​
Manager workflow

Skill-level proposals are evidence-driven and surface automatically — you confirm or return them, you do not initiate them from scratch.
1

Evidence Accumulates

As a member closes PROJ issues, authors KB posts, earns certifications, and collects endorsements, LEARN automatically updates the evidence record attached to their mastery claims.
2

Proposal Surfaces

When evidence crosses the threshold for the next mastery level, LEARN surfaces a proposal to the member’s manager: “Evidence supports upgrading Mai’s postgres skill from level 3 to level 4.”
3

Manager Confirms

The manager reviews the linked evidence and either confirms the upgrade or returns it with a note explaining what additional evidence is needed.
4

Promotion Case (Manual Nomination)

When you believe a member is ready for a level change (e.g. L2 → L3), you open a promotion case manually. The system checks current gate-criteria readiness and flags any gaps before the case is submitted to the council.