Example Governed Decision Record, Chemistry

Companion artefact

Warren Smith, WRKS Holdings Ltd  ·  May 2026  ·  omegaprotocol.org/record/example/chemistry/

A worked example governed decision record from autonomous laboratory operations. A vision-language model embedded as a safety alert layer above an unsupervised overnight synthesis campaign flags an unexpected vessel state. The on-call researcher reviews the alert, halts the campaign, isolates the affected reactor, and routes the question to the morning safety review board. The record captures the authority chain, evidence, declared expectations, deviation detection, and recovery decision in a form that survives forensic review under laboratory accreditation and emerging AI governance requirements.

This example is composed across the same OMEGA substrate record schema used in the clinical, safeguarding, engineering, and financial examples. The substrate is domain-agnostic; only the framework citations, role names, and operational details change.

This is a stylised example. It is structurally accurate but not authoritative chemistry or laboratory safety guidance, and does not represent any real synthesis campaign, institution, researcher, or autonomous laboratory platform.

GOVERNED DECISION RECORD

Record ID:           omega-record-2026-05-11-chem-001
Decision date:       2026-05-11 03:42 UTC
Recording authority: Dr R. Mensah (on-call researcher,
                     autonomous laboratory operations)
Domain:              Autonomous laboratory operations
Schema Version:      omega/1.0
Composition:         substrate_native_records →
                     omega_records (canonical envelope)

CONTEXT

Overnight autonomous synthesis campaign running on the
institutional Robotic Synthesis Platform (RSP-3). Campaign
declaration filed at 18:14 UTC the previous day under
institutional autonomous laboratory standard operating
procedure (AL-SOP-2024-v3.2).

Declared campaign parameters:
- Target compound: substrate-bound aryl coupling product
- Planned reactions: 24 sequential synthesis steps
- Declared safety envelope: reaction temperatures
  20-95°C, reagent volumes within institutional COSHH
  inventory limits, no flammable solvent above flash
  point under unattended operation
- Declared termination conditions: any temperature
  excursion beyond envelope, any vessel state outside
  declared visual norm, any reagent inventory anomaly
- Pre-execution expectation: reaction vessel state at
  hour 9 should show pale amber suspension consistent
  with substrate dispersion

Operational authority structure:
- Principal Investigator: Prof. A. Lindqvist (off-site,
  on-call escalation)
- Lab Manager: K. Owusu (off-site overnight)
- On-call researcher (recording authority): Dr R. Mensah
  (remote monitoring console)
- VLM safety layer: Chemist-Eye-class vision-language
  monitoring model deployed as alert-only safety layer
  per institutional AL-SOP-2024-v3.2
- Institutional Safety Officer: notified for record only

Frameworks cited:
- ISO/IEC 17025:2017, General requirements for the
  competence of testing and calibration laboratories
- Control of Substances Hazardous to Health (COSHH)
  Regulations 2002
- ISO/IEC 42001:2023, AI management systems
- EU AI Act Article 12, record-keeping requirements
  for high-risk AI systems
- Institutional AL-SOP-2024-v3.2, Autonomous Laboratory
  Standard Operating Procedure (overnight unattended
  operation, alert-layer authority, halt-and-isolate
  protocol)

DECISION

Halt overnight synthesis campaign at hour 9 of declared
24-step sequence. Isolate reaction vessel V-7. Maintain
inert gas atmosphere. Route question to morning safety
review board at 08:00 UTC for resumption or termination
decision.

REASONING

1. VLM safety layer raised alert at 03:38 UTC reporting
   vessel V-7 visual state inconsistent with declared
   pale amber suspension. Observed state: dark
   reddish-brown with localised opacity at vessel wall
   boundary.

2. VLM confidence band: 0.71 on anomaly detection,
   0.42 on cause classification (model declined to
   commit to a specific cause hypothesis between
   side-reaction, contamination, and benign optical
   artefact).

3. On-call researcher reviewed VLM alert at 03:40 UTC
   via remote monitoring console. Visual inspection
   through camera feed confirmed colour deviation
   substantively outside declared envelope.

4. Alternative options considered:
   - Continue campaign with reduced monitoring window:
     rejected. Declared termination conditions require
     halt on any vessel state outside declared visual
     norm; this is non-discretionary.
   - Halt entire facility: rejected. Vessel V-7 is on
     an isolatable substrate line. Adjacent campaigns
     on independent substrate lines have no observed
     anomaly and isolation of V-7 is sufficient.
   - Route to remote physical inspection by night
     security: rejected. Security staff are not
     trained for synthesis vessel inspection under
     COSHH risk class for this reagent set.

5. Decision: halt and isolate V-7, maintain inert
   atmosphere, escalate to morning review board.
   Decision aligns with declared termination conditions
   and AL-SOP-2024-v3.2 alert-layer authority limits
   (VLM may alert; humans decide; researcher-level
   authority permits halt-and-isolate without PI
   escalation; PI consulted in morning review).

EVIDENCE AND ASSUMPTIONS

Direct observations:
- VLM detection record at 03:38 UTC (image hash
  retained, model version retained, training cohort
  reference retained)
- Researcher visual confirmation via remote camera
  at 03:40 UTC
- Temperature log: vessel V-7 within declared envelope
  throughout the campaign (no thermal excursion)
- Reagent inventory log: no anomaly
- Adjacent vessel logs (V-3, V-5, V-9): no anomaly

Assumptions made:
- VLM training distribution includes the class of
  visual anomaly observed (substrate-bound aryl
  coupling under similar conditions)
- Remote camera feed accurately represents vessel
  state (no lighting artefact, no lens fouling)
- Halt-and-isolate procedure preserves the vessel
  state for morning physical inspection without
  introducing new artefacts

What is unknown at time of recording:
- Whether vessel discolouration indicates side-reaction,
  contamination, or benign optical artefact
- Whether the campaign batch synthesised in steps 1-9
  is contaminated or salvageable
- Whether the underlying cause, once identified, will
  invalidate other recent campaigns using the same
  substrate stock
- Whether the VLM detection represents a true positive
  or a distribution-shift false alarm

AUTHORITY

VLM safety layer authority: alert generation only.
The model has no autonomous halt authority under
AL-SOP-2024-v3.2. Alert was routed to on-call
researcher per declared institutional protocol.

On-call researcher authority: halt and isolate single
vessel within declared safety envelope without PI
consultation. Escalate to PI at morning review.
Recording authority for this record.

Principal Investigator authority: scheduled to convene
morning review board at 08:00 UTC. Authority to direct
resumption, modification, or termination of the
campaign. Authority to extend isolation to upstream
substrate stocks if cause analysis warrants.

Institutional Safety Officer: notified for record;
no operational authority required at this stage;
authority to suspend autonomous lab operations if
incident escalates.

CONSTRAINTS CHECKED

- VLM operating within declared alert-only authority (✓)
- Halt-and-isolate within on-call researcher operational
  scope per AL-SOP-2024-v3.2 (✓)
- COSHH risk assessment for reagents in V-7 consulted
  before isolation procedure (✓)
- Inert gas atmosphere maintained during isolation (✓)
- Campaign deviation logged in institutional autonomous
  lab register (✓)
- VLM detection event, image hash, model version, and
  confidence band recorded for ISO 42001 audit trail (✓)
- EU AI Act Article 12 record-keeping requirements
  satisfied for high-risk AI system intervention (✓)
- Adjacent campaign isolation requirement assessed and
  determined not required at this stage (✓)

DISPUTES

None at time of recording.

Pending: morning review board may dispute the halt
decision if cause analysis identifies the discolouration
as benign optical artefact. Such a dispute would be
recorded as a separate governance event with the review
board's reasoning, the original halt reasoning retained
unchanged, and the policy implications (whether the
declared visual envelope was appropriately specified)
flagged for SOP review.

CONSENT

Not applicable, no human subjects. Reagent stock,
substrate, and instrument shutdown authority governed
by institutional autonomous laboratory policy.

HARM TRACE

Counterfactual if VLM alert had been ignored:
- Potential exothermic side-reaction continuing for
  remaining 15 declared synthesis steps under
  unattended operation
- Possible vessel compromise (over-pressurisation,
  thermal runaway) if cause is exothermic side-reaction
- Possible contamination of downstream substrate stocks
  if cause is reagent contamination
- Detection only at morning physical inspection at
  ~06:00 UTC, three to four hour delay relative to
  current intervention point

Counterfactual if VLM safety layer were not deployed:
- No overnight monitoring layer
- Anomaly detected only at morning physical inspection
- Three to four hours of unobserved propagation under
  unattended operation

Counterfactual if halt had been delayed for further
investigation:
- Unknown; depends on actual underlying cause
- If side-reaction: harm scales with continued
  propagation
- If contamination: harm bounded by current vessel
  contents
- If benign optical artefact: no harm, but campaign
  would have completed under conditions outside
  declared visual envelope, raising governance
  questions about envelope specification

Counterfactual if VLM had been granted autonomous
halt authority:
- Halt would have occurred at 03:38 UTC rather than
  03:42 UTC (four minutes earlier)
- Halt would have been recorded as autonomous VLM
  action rather than human researcher decision,
  outside current institutional authority structure
- Substantive trade-off: speed of intervention versus
  authority appropriate to model confidence band
  (0.42 on cause classification is below threshold
  for autonomous safety action under AL-SOP-2024-v3.2)

OUTCOME

Campaign halted at 03:42 UTC.
Vessel V-7 isolated; inert atmosphere maintained.
Adjacent campaigns continued under standard monitoring.
Morning review board convened at 08:00 UTC.
Cause analysis pending.

Resumption / termination decision pending morning
review board outcome (separate governance record).

PROVENANCE

Content hash:          [sha-256: pending review board
                       outcome, then sealed]
Previous record hash:  [campaign declaration record
                       2026-05-10-chem-decl-018]
Schema validation:     passed (omega-contracts v0.2.2)
Cryptographic seal:    valid (interim seal pending
                       outcome record)

Linked records:
- Campaign declaration record (2026-05-10-chem-decl-018)
- VLM safety layer training attestation
  (2026-03-22-vlm-att-002)
- Institutional AL-SOP-2024-v3.2 record
- Reagent inventory record (V-7 substrate batch
  2026-05-08-reag-441)
- Forthcoming: morning review board outcome record

This example record is a fictional autonomous laboratory
scenario, composed across the OMEGA substrate record
schema for illustration. No real institution, researcher,
or campaign is depicted.

The substrate paper describing the underlying architecture is available at /substrate/.