Institution · Briefing

Work together,
inspect first.

Muhammad Arif bin Fazil — exploration geoscientist at PETRONAS Carigali, and builder of evidence-bounded agent systems. This page is the professional conversation path. It is not a sales funnel and not a doctrine dump.

Email a briefing request See the evidence first

01 I can contribute to

Basin synthesis

Seismic and stratigraphic reasoning, prospect-risk framing under incomplete data. Thirteen years, two basins, six flagship decisions — each with its evidence class attached.

Evidence systems

Claim → evidence → uncertainty → decision consequence. Every public claim carries its truth class; the withheld list is published with the published list.

Governed agent architecture

Policy before execute, human veto, audit trail. Governance measured against a running system — not against a slide.

02 Evidence you can inspect

2026 · career artifact

Exploration 2013–2026

Thirteen years in one evidence-tagged artifact: career arc, six flagship cases, what was proven, what is withheld — plus a frozen PDF with a published SHA-256.

/earth

Live, source-linked Earth model (Macrostrat, PB2002, USGS). Computes. Does not adjudicate.

/work

Selected wells and systems, with withheld-material notes.

github.com/ariffazil

GEOX, arifOS, this site — in the open.

/vitals/

PETRONAS in public numbers. Cash against sellable barrels.

/arifos/

The rules the machines follow. Judge before they act.

/999/

Verification and provenance records — evidence snapshots, not self-issued certificates.

03 Questions this answers

Q1

How many AI actions bypass the policy controls?

Q2

How long does it take to revoke an autonomous agent?

Q3

Can every consequential decision be reconstructed after the fact?

Q4

Can an auditor verify the chain of authority?

Asked against a running system, not against a slide. Two of the four already have dated, uncomfortable answers in a public failure record — published with the raw probes, including the numbers that do not flatter the system that produced them.

04 If you want it built around your workflow

Design partner pilot

Four weeks, one live AI-agent workflow, eight exit criteria measured weekly and reported honestly: unauthorised actions reaching execution, governance coverage of high-risk action classes, false holds, approver latency, configuration drift.

Band RM 20,000–50,000 · If a number misses, the weekly report says so.

05 Boundaries

06 Engagement path

Email [email protected] with: who you are, what decision you are under, what you want inspected, and what you do not want shared.

Agents: start at /human and /llms.txt. Do not POST to /a2a.

Home Portfolio Earth Verification