LensReading which lens this session carries.

Auditor evidence pack

Model at deploy

Research & Development · Data Science & Modeling · line RD5A · Model Development. This pack names every action taken on this unit, who or what took it, the authority that permitted it, and the model, prompt and policy versions in force at that moment.

Chain verified

Every record links to the one before it and hashes to its stored value. 5 records were re-hashed on this page load, in order, starting from the genesis value. Nothing was read from a cached verdict.

hash = sha256(prevHash + "|" + JSON{seq, unitRef, actor, action, outcome, occurredAt})

Records
5
Append-only, sequenced
Agent actions
3
Taken by software
Human actions
2
Taken by a named person
Refusals
0
Blocked by the authority register
Corrections
0
Later record supersedes an earlier one
Legal hold
None
Free to age out on schedule

The unit

What the work was, and where it sits

Reference
RD5A-1018
Subject
Pierre Nakamura
Requested by
Tomas Duarte
Value
$61k
Materiality
medium
Currently held by
Model Training Agent (agent)
Sub-function owner
Director, Data Science
Orchestrator
Modeling Orchestrator

Every action, in order

Actor, authority, versions in force, and the hash link to the record before it

1
Validation AgenttaskCleared

Cleared validate

2026-08-18 02:42:00ZAuthority RD5-AUTH-A1 · cap $250,000Jurisdiction UKAutonomy A3Retention TECH-3Y
Prompt
task-execution@p21
Policy
segregation-of-duties@v4
Tools
document-fetch@t9, read-only-lookup@t12
in ea8061a3...a4ed04out b52b39ee...0df010artifact ART-RD5A-1002-1 genesis a1e902d0...45b3f7
2
Explainability AgenttaskCleared

Cleared explainability pack

2026-08-18 03:58:00ZAuthority RD5-AUTH-A1 · cap $250,000Jurisdiction UKAutonomy A3Retention TECH-3Y
Prompt
task-execution@p21
Policy
segregation-of-duties@v4
Tools
document-fetch@t9, read-only-lookup@t12
in e9b56ae5...91e90dout 4d39ce5c...ff1ff4artifact ART-RD5A-1002-2 a1e902d0...45b3f7 758b2037...3fcb46
3
Director, Data Sciencegate-approverCleared

Cleared reviewer approve

2026-08-18 05:13:00ZAuthority RD5-AUTH-H1 · cap $250,000Jurisdiction UKAutonomy A0Retention TECH-3Y
Model
human@n-a
Prompt
human@n-a
Policy
segregation-of-duties@v4
Tools
none
in da7922c8...b71da7out 52bfd035...527382artifact ART-RD5A-1002-3 758b2037...3fcb46 0e5d325f...f9905f
4
Model Governance AgentpolicyCleared

Cleared register

2026-08-18 06:29:00ZAuthority RD5-AUTH-A1 · cap $250,000Jurisdiction UKAutonomy A2Retention TECH-3Y
Prompt
gate-approval@p8
Policy
segregation-of-duties@v4
Tools
read-only-lookup@t12
in dca3d298...3945b9out 8238d660...1ed891artifact ART-RD5A-1002-4 0e5d325f...f9905f 9783af89...6debdf
5
Director, Data Sciencegate-approverIn progress

Holding at deploy

2026-08-18 07:44:00ZAuthority RD5-AUTH-H1 · cap $250,000Jurisdiction UKAutonomy A0Retention TECH-3Y
Model
human@n-a
Prompt
human@n-a
Policy
segregation-of-duties@v4
Tools
none
in 1ef4bbc2...fb7219out 3aeb4742...6d27ccartifact ART-RD5A-1002-5 9783af89...6debdf cbe8ed5a...748181

Authority relied on

Who was allowed to do this, up to what value, in which jurisdictions

RD5-AUTH-A1ActiveModeling Orchestrator

Any unit on a data science & modeling line where every intake check passed, the policy in force is current, and the value sits inside the cap below.

Cap $250,000US · UK · DE · SG · AU · FRDelegated from Director, Data ScienceExpires 2026-12-03
RD5-AUTH-H1ActiveDirector, Data Science

Anything the agent grant refuses, plus any override of an agent outcome. An override must state what the agent concluded and why it is being set aside.

Cap $2,000,000US · UK · DE · SG · AU · FRDelegated from Research & Development leadershipExpires 2026-09-26

Versions in force

Exactly what was running when these actions were taken

KindIdentifierVersionStatusOwner
modelreasoner-core5.4currentPlatform model council
modelextract-lite2.9currentPlatform model council
policysegregation-of-dutiesv4currentInternal audit
promptgate-approvalp8currentGovernance office
prompttask-executionp21currentShared service engineering
toolsetread-only-lookupt12currentPlatform engineering
toolsetdocument-fetcht9currentShared service engineering

Gates on this line

The control points this unit had to clear

Reviewer approve
Held for a named human when the unit exceeds the model decision envelope, when confidence falls below the floor, or when a policy clause is engaged.
Approver: Director, Data Science · not reached
Deploy
Held for a named human when the unit exceeds the model decision envelope, when confidence falls below the floor, or when a policy clause is engaged.
Approver: Director, Data Science · cleared on this chain

Records schedule and holds

How long this evidence must be kept, and whether it is frozen

TECH-3Y3 years
Technology change and access evidence
Security review and change audit requirement.
Deleted at three years from the change closing.

What this pack proves, and what it does not

Read this before you rely on it

It does prove
  • Every action carries a named actor, an actor kind, and a timestamp.
  • Every action names the authority grant that permitted it and the cap that grant carries.
  • Every action stamps the model, prompt, policy and tool versions in force at that moment.
  • Corrections are additive. A later record supersedes an earlier one and says why; the earlier record is never edited or deleted.
  • The chain is re-hashed live on this page, so a single altered field is detected immediately.
It does not prove
  • It is not tamper-proof against an attacker with write access to the store, who could rewrite the whole chain from genesis. Making that impossible requires publishing chain tips to a store this application cannot write to. That anchor is not built, and this page does not claim it.
  • The evidence in this demonstration is modeled from a simulated estate, not captured from production systems.
  • Identity here is the actor name recorded on the action. Cryptographic agent identity and signed attestation are a separate piece of work.