Workshop 01 — The Train Preacher

The trap — the point of this exercise

Every row above comes from a party to the event publishing its own conduct. A 5W+H is not a neutral instrument: point it at one curated source and it will describe what the publisher wants visible, in a tidy six-row format that looks like fact.

Three habits to take from this:

1. Mark the status of every answer. An unmarked 5W+H is indistinguishable from a verified one.
2. Ask whose interest the framing serves — here, a title that promises a backlash rather than merely describing an encounter.
3. Finish the exercise by listing what the source cannot answer. That list is the bridge to Exercise 5.

Small detail worth surfacing: the title says train; at least one passenger's words, as transcribed, say bus. A minor discrepancy, but exactly what a transcript check is for.

The source is the publisher, and the publisher is a party to the event. This is first-party material: the ministry uploaded footage of its own member's encounter. It is primary in the sense that it is the event rather than commentary about it — but it is also selected, edited and titled by an interested party.

Two defensible gradings, and the reasoning matters more than the letters:

Neither is wrong. The letters hide the decisive point: reliability and independence are not the same thing. The clip can be entirely authentic — high reliability as an artefact — while being a poor basis for any judgement, because there is no independence in it at all.

Three consequences to reach:

• A source that is also the sole evidence must be flagged as single-source, and the confidence of anything built on it capped accordingly.
• The transcript is a separate, third-party artefact with its own failure modes: no punctuation, redacted profanity, and no speaker attribution. The clip never establishes who said which line. Treating a machine transcript as a faithful record is a mistake, and in this material it is a visible one.
• Cumulative metrics are not indicators. 717,090 views describes distribution, not threat. It is also an incentive: on this platform, provocation is what gets shared. Engagement figures measure the publisher's reach, and reward the behaviour the title advertises.

What the source cannot tell us: who else was involved; what happened before or after; whether the encounter was typical of anything; whether the subject is still active, still in Australia, or still practicing; who held the camera; what was cut; how the passengers experienced the exchange; whether anyone complained, was charged or was hurt; and whether any of it has any current relevance at all.

Two assumptions sit inside the task. That the preacher is the threat source; and that a 2019 clip is evidence of a present condition. Both are testable, and testing them is the exercise.

Five candidate framings — each leads to different collection, different products and different risks:

The direction of risk, and the currency of the material, do most of the work here. A team that produces H1 by default has answered a question nobody asked, and may have written a record of lawful expression while believing it wrote a threat assessment.

Legal frame — check the premise, do not assume it:

• Summary Offences Act 1988 (NSW) — s 4 (offensive conduct) and s 4A (offensive language). The prescribed maximum for the conduct offence is six penalty units or three months' imprisonment. Re-read the current text before relying on it.
• Safe access zones — Public Health Act 2010 (NSW) Part 6A, inserted by the Public Health Amendment (Safe Access to Reproductive Health Clinics) Act 2018 (NSW), assented 15 June 2018 and commenced on assent. A safe access zone is the clinic premises and the area within 150 metres of them; ss 98C–98E cover interference with access, communications about abortion likely to cause distress or anxiety, and capturing or distributing visual data.
• The trap: those provisions operate inside a safe access zone. A moving train carriage is not one. The first instinct is usually "the abortion laws — charge him", and that instinct is wrong on these facts. That is the exercise working exactly as intended.

The distinction we must hold, and state: offensive is not unlawful; unlawful is not a security threat; and a threat assessment is not a charge.

Threat to police — the thread teams usually miss

The customer is the police, and the police are also a party who can be exposed. Nothing in the clip shows hostility towards police: the risk to officers here is situational and structural, not directed at them — which is exactly why it is easy to overlook.

• Physical. A confined carriage, a hostile crowd, and an arrival to enforce means the crowd's anger can transfer to whoever represents authority. Officers would be working shoulder to shoulder with the same passengers who are already shouting at a stranger.
• The camera changes the calculus. The subject's operating model is to film encounters and publish them. If police attend, the encounter becomes content — and the edited, titled product, not the officer's account, is what circulates. A lawful and proportionate intervention can still be presented otherwise, and the intervention itself may become the story.
• Legal and reputational. An intervention against a minority-position speaker, filmed and titled for engagement, is a reputational event regardless of its merits, and is reusable by political actors.

The practical consequence is that for an officer on the day, the tactical option is also a communications decision. That is a genuine finding for a police customer — and it is not something reachable by asking "is he dangerous?".

Verify the statute directly rather than trusting a summary: Summary Offences Act 1988 · Safe Access to Reproductive Health Clinics Act 2018

Induction, in one line: reasoning from what you observe to the pattern those observations imply. It produces hypotheses, not proofs — so every inference below is stated with the alternative it competes against.

Inference 1 — the subject was not alone

The camera settles on him, follows him, and holds him in frame at close range while he moves and gestures. He is mid-speech, addressing passengers rather than the lens. So a second person was present, and filming deliberately. The first question we want to ask — is he alone? — is therefore already answered: at the moment of the encounter, this was at minimum a two-person activity. Extend it cautiously: the footage was cut, titled and published, so roles exist beyond the speaker — camera, edit, publish.

Distinguishing evidence: whether the camera follows him off the carriage or between services; whether other uploads from the same channel show the same framing hand; whether a voice near the microphone belongs to the operator; whether the ministry ever describes a team, or credits a videographer.

Inference 2 — the testimony may be a plant

A man volunteers a conversion story that fits the message exactly, arriving as the clip turns from hostility towards redemption. Published evangelism content uses precisely this beat.

This is a hypothesis, not a finding, and the material cannot settle it. Both sides, honestly:

• For: the narrative fits the message exactly; it arrives where a producer would want it; it is the only sympathetic voice in the clip; and the clip is authored by the interested party.
• Against: spontaneous testimony is genuinely common in this activity; the man sounds unrehearsed, is repeatedly interrupted and never finishes; a staged confederate would usually be placed closer to the microphone; and the transcript cannot reliably tell us who is speaking at any given moment.

Distinguishing evidence: whether he is present before the confrontation begins; whether he appears in other uploads from the channel; whether the ministry ever credits him; whether any second recording of that carriage exists.

The discipline

• State the inference, then the alternative it competes with, then the evidence that would separate them. An unopposed inference is a guess.
• Neither inference goes into a product as fact. Both belong in the gap register until collection supports them.
• Watch the pull: once you suspect staging, the whole clip starts to look staged. That is Confirmation Bias in a detective's coat — and it is the likeliest trap in this exercise.

The priority rule

Gaps that would change the judgement come first. Gaps that only add colour are not gaps. The view count, the speaker's nationality, his denomination and his theology are all readily available, all interesting, and none of them move the assessment. A register padded with those is a register that has not been thought about.

Two rows here exist only because of Exercise 4 — induction told us there is at least a second person, and handed us a hypothesis we cannot test. That is the normal rhythm: infer, then collect.

Note what the first two rows have in common: they are cheap to fill. The reason this task stalls is not that the answers are hard to get — it is that the task was accepted without asking what it was for.

What we should not collect

• Anything about the individual beyond overt, public, lawful material, unless a real tasking, authority and legal review exist. The fictional scenario does not grant them, and the exercise does not require them.
• A profile of a person's beliefs, associations or expression in place of evidence of conduct. Belief is not a threat indicator; conduct, frequency and harm are.
• Platform metrics as if they were intelligence. They are the publisher's distribution figures.

This is a template, not an assessment. It demonstrates form: the reframed question, explicit confidence, the separation of fact from judgement, and the gaps left open. Nothing here may be lifted into a real product without the collection, authorities and handling that a real product requires.

Task as reframed. Does the subject present a current public-safety or public-order risk to NSW transit users, and if so, in which direction?

BLUF. The question as posed cannot be answered from the material available: the only substantive source is a single, first-party, edited upload from 2019. We assess with high confidence that the source is inadequate for a judgement about the subject, and with high confidence that there is no evidence of a present condition — noting that this reflects an empty evidence base, not a finding of safety.

Key judgements.

• Currency — no finding possible. No collection establishes whether the subject is active in NSW today. Absence of evidence is a collection gap, not evidence of absence.
• Source — single, interested, edited. The clip is authentic as an artefact and unusable as a sole basis for judgement. Confidence: high.
• Risk direction — unresolved. The observed pattern, if current, contains risk in both directions: to the public from a disruptive approach in an enclosed space, and to the speaker from an agitated crowd. Confidence: low about this subject; moderate as a pattern statement about the activity.
• Intent and character — no judgement offered. No lawful collection exists to support one, and none is needed for the decision the customer faces.

What would change the call. A current incident report from the operator; evidence of a repeated practice on NSW services; a complaint history in either direction; an assault at either end of an encounter; evidence of organised activity; or one independent recording of the same encounter.

Recommendation to the customer. Do not accept the task as framed. Establish the decision it serves. If the concern is transit disruption, the operator's incident data answers it. If the concern is speech that offends, that is a legal and policy question, not an intelligence one — and framing it as a threat assessment creates a record that will not bear the weight.

Confidence and caveats. Judgements about the source are high confidence. Judgements about the subject are low confidence and rest on a single-source base with no independent corroboration. Collection gaps: currency, organisational status, pattern, complaint history.