How should the wiki enforce its journalism-standards checklist?

One-line summary: The journalism-ethics literature finds that aspirational codes (SPJ) depend on professional culture, while structural codes (Reuters Founders Share Company, BBC Trust) embed enforcement in governance. The wiki's planned vault/_meta/JOURNALISTIC_STANDARDS.md checklist is currently aspirational — its enforcement model is undefined. Possibilities: ingest-time validation, periodic audit pass, lint-extension, paired-review skill. The answer materially affects whether the checklist actually changes ingest behavior.

The question

Once the operational checklist (built from journalism-practitioner-codes-canonical-tenets + 2026-05-13-academic-research-journalism-standards-political-reporting) exists in vault/_meta/JOURNALISTIC_STANDARDS.md, how is its application enforced? Without an enforcement mechanism, the checklist suffers the same fate as the SPJ Code — aspirational but not binding, depending entirely on whether the operator (the LLM driving ingest) chooses to apply it. The journalism literature finds that aspirational codes have limited behavioral effect without enforcement infrastructure.

Why it matters

This is the central operational question for the journalist-agent the user wants to build. The substantive research is done (the checklist content is largely sketched in the two foundation sources); the open question is the mechanism by which the checklist actually disciplines LLM-driven ingest. Three failure modes if enforcement is undefined:

1. The checklist becomes performative. Per Chadha & Koliska 2015 (Journalism Practice, 68 citations), newsrooms exhibit "limited and strategic transparency that enable[s] them to appear transparent without offering substantive insights into the journalistic process." The wiki could exhibit the analogous failure: the checklist is cited without being applied.
2. The checklist degrades silently under "I already know the answer" pressure. The LLM may judge that a particular page doesn't need the right-of-reply check because the answer is obvious. The same shortcut is what produces the bias-failure modes the checklist was designed to catch.
3. The checklist doesn't survive ingest-pending scaling. When the queue is long, partial application is more likely. Without explicit enforcement, abbreviated application is undetectable.

What we currently believe

• Aspirational codes alone are insufficient — SPJ explicitly notes its code is "not legally enforceable", and the journalism literature treats this as a known limitation rather than a feature.
• Structural codes work better — Reuters Founders Share Company (1984) and BBC Trust both enforce ethics via governance mechanisms that survive individual-actor compliance failures. From 2026-05-13-autoresearch-practitioner-codes-for-journalism: "The Reuters Founders Share Company holds a single golden share in Thomson Reuters Corporation. Its directors, called Trustees, ensure compliance."
• Organizational discipline works in narrow contexts — AP's "subject to review, and could result in disciplinary action, ranging from admonishment to dismissal" is enforceable inside the AP. The wiki has no analogous discipline structure for the LLM operator.
• Organizational culture matters more than written policies — George et al. 2019 (Journalism) found that "organisational culture, more than written policies, in maintaining professionalism." For the wiki, this would mean the human user's ongoing practice of demanding adherence (and pushing back when adherence is missing) IS part of the enforcement layer.

Evidence we have

• The academic and practitioner-code literatures both find aspirational-only enforcement is weak.
• The wiki's existing operational checklist (the politics/SCOPE "Sourcing posture for contested factual claims" section) is currently aspirational — it depends on the LLM applying it during ingest. Empirically, in this conversation's earlier ingests of the MN/CA fraud sources, the SCOPE rule was applied more in the round-2 ingest (after the user pushed back) than in the round-1 ingest (where the LLM defaulted to mainstream-source-leaning). This is some evidence that user-discipline-as-enforcement works.
• The Chadha & Koliska 2015 "performative transparency" finding is the closest analog warning sign.

Evidence we need

• An empirical pattern across multiple political ingests of how often each checklist element is applied vs missed. This requires the checklist to first exist in operational form.
• Comparison of LLM-ingest behavior with and without explicit checklist enforcement at ingest time.
• An understanding of what enforcement mechanisms work in LLM-driven workflows generally — analogous patterns from CI/CD, code review, or other governance domains.

Candidate enforcement mechanisms

Five options, with tradeoffs:

Option A: Ingest-time validation pre-condition

Modify ingest-pending so that for political-thread sources (or all contested-fact sources), the LLM cannot complete the ingest without explicitly documenting which checklist elements were applied to which wiki claims. The log.md ingest entry includes a checklist self-assessment.

Pro: Cannot be skipped silently — log entry won't pass review without it. Con: Adds substantial overhead to every ingest; the LLM can still produce a self-assessment that's wrong; doesn't catch subtle application failures.

Option B: Periodic audit pass (new skill)

A new /audit-political-page <slug> skill that the user invokes occasionally on a sampled wiki page. The skill walks through the checklist element-by-element and produces an audit report — pages that fail surface their failure modes.

Pro: Catches subtle failures that ingest-time self-assessment misses; preserves audit-trail; can be applied retroactively to existing pages. Con: Periodic, not continuous. Pages between audits may drift. User has to remember to run it.

Option C: Lint-extension

Extend the existing wiki-lint pattern with checklist-specific rules. A /lint-political-page could check for: missing right-of-reply citations on subject pages, missing framing-bias notes on secondary-source citations, missing primary-document anchors on dispositive claims.

Pro: Cheap to run; mechanical checks are reliable. Con: Only catches mechanical-checklist failures; misses the substantive checklist elements (like "evidence-weighted not format-equivalent balance" which requires LLM judgment).

Option D: Paired-review skill (two-LLM)

A new /review-political-page <slug> skill where one LLM reviews another's wiki page output against the checklist. The reviewer LLM is given the page + the checklist + an instruction to challenge missing-or-incomplete elements.

Pro: Closest analog to the journalism "editor reviews reporter" enforcement model. Can catch substantive elements that mechanical lint misses. Con: Reviewer LLM has the same biases as the writer LLM. Two-LLM coverage is better than one but not by as much as one might hope.

Option E: User-as-editor (current de facto state)

The user reviews wiki output and pushes back when checklist elements are missed. Over time, the LLM's ingest behavior improves (within a session at least). This is the current de facto state and is what produced the round-2 improvements in this conversation.

Pro: Already works to some degree; no new infrastructure needed; matches the George 2019 "organizational culture > written policy" finding. Con: Doesn't scale; depends on user attention; doesn't catch what the user doesn't notice; LLM behavior may not persist across sessions without explicit re-grounding.

Probable best answer: a combination — Option C (cheap mechanical lint) + Option B (audit pass for harder elements) + Option E (user-as-editor as the ultimate backstop). Option A as a stretch goal if the LLM's self-assessment can be made reliable enough. Option D adds value but at higher token cost.

How to resolve

• Build the operational checklist first (vault/_meta/JOURNALISTIC_STANDARDS.md).
• Start with Option E (current state) and observe failure modes over the next 5–10 political ingests.
• Add Option C (lint-extension) for the mechanical checks that fail most often.
• Add Option B (audit pass) for the substantive checks that lint can't reach.
• Defer Option A and Option D until empirical data on Options C/B suggests they're needed.

Related

• journalism-practitioner-codes-canonical-tenets — what the checklist content is
• no-surprises-rule — one specific checklist element
• citizen-journalism-vs-mainstream-state-program-fraud — the meta-concept this question serves
• politics/SCOPE — where the current de facto enforcement lives
• 2026-05-13-autoresearch-practitioner-codes-for-journalism + 2026-05-13-academic-research-journalism-standards-political-reporting — sources