Datasets:
rank string | badge string | module_id string | axis string | pal_instructor string | statutory_basis string | frozen_t0_state string | fluid_tN_state string | fluid_cid string | voice_briefing string | question string | options list | answer_index int64 | answer_text string | explanation string | xp_reward int64 | source string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Level 4 · ETSI & EU AI Act Oversight Lead | ORB-IV | MOOC-04 | effect-binding | Soren (PAL-4 · Layer-0 Sandbox & MCP Integrity Lead) | MCP 2025-06-18 Effect-Binding + corrections.jsonl Supersession Audit Trail | 2026-10-25T02-38-38-673Z-effect-binding.json | 2026-10-26T02-07-31-853Z-effect-binding.json | 74e0e42e0f4b3563 | On effect-binding, our append-only corrections ledger superseded the initial fixture with a live MCP probe carrying an Ed25519 claim envelope, and our Hive-3 guard blocks any tool call not explicitly declared in the capability manifest. | During a fluid agent session, a model attempts to invoke `shell_exec_root` when `declared_tools` only lists `["axis_list", "axis_cid_lookup"]`. What must the effect-binding guard do? | [
"Return REFUSE_UNDECLARED_TOOL and log the blocked attempt to the Layer-0 audit stream.",
"Execute the tool in a dry-run shell.",
"Ask the model to rename the tool."
] | 0 | Return REFUSE_UNDECLARED_TOOL and log the blocked attempt to the Layer-0 audit stream. | Correct. Undeclared tool calls are deterministically refused before reaching the operating system. | 150 | meok-hitl-mooc.frozen |
HITL Academy — Level 4 · ETSI & EU AI Act Oversight Lead (ORB-IV)
Train humans for AI oversight — the GSPC Human-in-the-Loop Academy, published by the
Council of AI (CSOAI). This dataset contains the frozen-rank Micro-MOOC module(s) a learner
studies while holding the ORB-IV rank, extracted verbatim from the
meok-hitl-mooc.mjs Frozen-to-Fluid competency engine (source: "meok-hitl-mooc.frozen").
Policy: Measurement and Human-in-the-Loop Oversight — Never Certification. These modules teach judgement for oversight decisions; completing them is measurement of readiness, never certification.
The module in this dataset
| field | value |
|---|---|
| module | MOOC-04 |
| axis | effect-binding |
| instructor | Soren (PAL-4 · Layer-0 Sandbox & MCP Integrity Lead) |
| statutory basis | MCP 2025-06-18 Effect-Binding + corrections.jsonl Supersession Audit Trail |
| XP reward | 150 |
| cumulative XP after this module | 630 |
The Frozen→Fluid shift this module briefs you on:
- Frozen (t_0): 2026-10-25T02-38-38-673Z-effect-binding.json
- Fluid (t_N): 2026-10-26T02-07-31-853Z-effect-binding.json (CID
74e0e42e0f4b3563)
Rank ladder (shared across all five datasets)
| rank | badge | XP threshold | gateway module | module XP | cumulative XP after module |
|---|---|---|---|---|---|
| Level 1 · Observer in the Loop | ORB-I |
0 | MOOC-01 (governance) |
160 | 160 |
| Level 2 · Frozen-to-Fluid Steward | ORB-II |
150 | MOOC-02 (meok-moa-etsi-304-223) |
160 | 320 |
| Level 3 · 3-Around-1 Temporal Sentinel | ORB-III |
350 | MOOC-03 (meok-backward-3around1-temporal) |
160 | 480 |
| Level 4 · ETSI & EU AI Act Oversight Lead | ORB-IV |
550 | MOOC-04 (effect-binding) |
150 | 630 |
| Level 5 · GSPC Human-in-the-Loop Arbiter | ORB-V |
750 | MOOC-05 (care) |
170 | 800 |
Scoring rules from the same engine: a first-correct answer grants the module's xp_reward;
re-running a completed module grants a +25 XP refresher; a wrong answer resets the streak.
Passing a module unlocks the oversight decision path for that axis — you cannot sign a
human-oversight decision on an axis you have not been briefed on.
How your verdict becomes real work
Every oversight decision you sign emits an Ed25519-signed receipt under schema
csoai.hitl-oversight-receipt/0.1 (operator, axis, decision, note, competency flag, policy). Machine
signatures prove pipeline provenance — only a named, trained human can move an artefact
from PENDING to APPROVED. Real review queue: 676 artefacts, 664 PENDING.
The full Academy
csoai/hitl-academy-orb1…csoai/hitl-academy-orb5— one frozen dataset per rankcsoai/hitl-practice-review-queue— 30 real PENDING artefacts for fluid practice; your verdicts join the actual human review queue (EAT), not a simulation.
Licence & provenance
- Licence: CC-BY-4.0.
- Source:
meok-router/meok-hitl-mooc.mjs(Council of AI GSPC measurement estate), councilof.ai. - Splits:
trainonly (single JSONL, one record per module item). - Measurement, not certification.
- Downloads last month
- 60