QESIS+ Agent Console

The six agents run headless. This page is the instrument that shows what they did, read from the hash chain rather than recalled from a session. It is a plain file: it opens in any browser, makes no network request, and needs no assistant, no plugin and no account to read. Two planes appear below, the runtime plane from the operational store and the index plane from the published vintage, and on this surface both are read at build time, which was 2026-07-31T11:48:40Z. On the pinned desktop artifact they diverge, and the runtime plane there carries its own older stamp for that reason.

chain height
604
entries, hash linked
link breaks
0
recomputed, not asserted
agents
6
registry closed at six
operations
30
distinct action types logged
open tasks
13
owner named on each
0

actions held at the human gate. Each one ran, produced output, and stopped. Release requires a named person through python -m qesis_agents approve <log_id> --by "R. Batista Silva". No agent signs and no agent may approve its own work.

Topology

SCOUT collect · 98 logged SENTINEL gate · 188 logged ANALYST reason · 93 logged ARCHITECT record · 134 logged HERALD publish · 45 logged · 36 held COUNSEL bind · 46 logged · 35 held Article 14 boundary. No agent signs.
Collection reaches nothing downstream without passing the gate. A warm border marks an agent with work waiting on a human.

Agents

SCOUT

collect

Research and collection, refresh monitoring

risk default
MEDIUM
web facing
yes
last run
2026-07-30T23:01:01Z
  • research_brief36
  • assess_content36
  • refresh_audit20
  • submit_intake3
  • citation_scan3

SENTINEL

gate

Integrity gate, security, EU AI Act and ISO 42001 conformity

risk default
MEDIUM
web facing
no
last run
2026-07-30T18:34:30Z
  • integrity_audit60
  • doctrine_audit54
  • validate_intake39
  • licence_audit18
  • verify_chain12
  • pending_approvals3
  • gate_publication1
  • connector_review1

ANALYST

reason

Correlation, fsQCA, epistemic classification

risk default
MEDIUM
web facing
no
last run
2026-07-30T18:16:06Z
  • correlate78
  • axis_matrix8
  • tables5
  • sql1
  • association_scan1

ARCHITECT

record

Documentation, reports, register hygiene, KPI custody

risk default
LOW
web facing
no
last run
2026-07-31T11:46:19Z
  • daily_report55
  • technical_documentation20
  • reconcile_register17
  • add_task17
  • kpi_snapshot14
  • task_queue10
  • propagate1

HERALD

publish

Content, SEO, funnel, lead generation

risk default
HIGH
web facing
yes
last run
2026-07-30T18:16:07Z
  • artifact_ladder3535 held
  • lead_scan9
  • draft11 held

COUNSEL

bind

Legal, commercial compliance, licensing, obligations

risk default
HIGH
web facing
no
last run
2026-07-30T18:16:07Z
  • triage_contract3535 held
  • licence_posture11

What waits on you

agentoperationarticle riskheldoldest
COUNSELtriage_contractArt. 14 (Human oversight)HIGH352026-07-28T17:12:48Z
HERALDartifact_ladderArt. 12 (Record-keeping)HIGH352026-07-28T17:12:48Z
HERALDdraftArt. 12 (Record-keeping)HIGH12026-07-28T17:19:01Z

The gate is wider than the article: 36 under Art. 12, 35 under Art. 14. Holding follows the reach of the output rather than the label on the record, because risk classification and write scope are separate properties and stay separate fields (L-035).

Index plane

vintage
v8.3 (2026-07-31)
supersedes
v8.2 (2026-07)
states
35
ranked
32
withheld under the coverage gate
3
uncertainty entries
2 high, 4 medium, 2 low

Composite: 0.3*WSE + 0.3*CSE + 0.17*ODI + 0.08*RGD + 0.15*REE. Withheld: HKG at 0.7, SGP at 0.7, TWN at 0.7. Coverage below 0.75 is published as a gap and never imputed (D-007).

Standing constraints

  1. D-006No credential is handled by any agent, in either direction.
  2. D-007Coverage below 0.75 is a finding. Nothing is imputed.
  3. D-004Quarantine, never delete.
  4. D-012Public surfaces carry methodology, never licensed data.
  5. D-019The registry is closed at six. A seventh displaces an existing one.
  6. D-027Anything writable lives outside the sync mount.
  7. L-015The em dash is banned in prose and the gate runs on this page too.
  8. L-023A counter reports the thing it is named after. Drafts count as drafts.
  9. L-040A capability is claimed only on the surface where it was verified.
  10. L-047A token in a presentation attribute is discarded and paints black.