Why we’re publishing this

Most newsrooms that use AI either hide it or bury a generic disclosure on a site-wide policy page. Neither serves readers. Lodi411 takes a different position: if artificial intelligence touches a piece of journalism, the reader deserves to know which tools touched it, at what stage, and under whose editorial judgment.

Lodi411 began publishing per-article AI methodology notes in LodiEye analyses in March 2026, after the underlying AI production pipeline was formalized, internally reviewed, and validated against working articles. The formalization step mattered: publishing a methodology note only makes sense once the methodology it describes is stable, reproducible, and auditable. From March 2026 forward, every LodiEye analysis carries a note naming the specific models used, the functions they performed, and the human editorial responsibility for the output.

This document explains the standard behind those per-article notes. It is the operating manual Lodi411 follows across its full production pipeline.

About Lodi411 and its founder

Lodi411 is a self-funded hyperlocal news, civic-data, and community-information platform serving Lodi and the surrounding San Joaquin County region. It operates without philanthropic grants, major-donor subsidies, or institutional parent funding. Its cost structure is designed to be sustainable at the scale of a mid-size California city through community advertising, membership, and earned revenue — not through recurring outside subsidy.

Lodi411’s founder brings more than 40 years of experience as a software developer and senior technology leader at Microsoft, Apple/Claris, Yahoo, and eBay, along with several startup ventures across that career. That background — deep full-stack engineering, large-scale data systems, search and discovery platforms, marketplace infrastructure, and product leadership at consumer-internet scale — is directly reflected in how Lodi411 is built and how it approaches AI. The methodology described below is the work of a practitioner who has shipped software in high-stakes environments for four decades, applied to the specific problem of sustaining serious local journalism in a market that traditional newsroom economics no longer support.

The combination matters: self-funded operation forces discipline about what technology actually earns its place in the pipeline, and a career of shipping production systems forces discipline about how that technology is validated, reviewed, and disclosed.

Our core principle

Within that principle, AI is used extensively — because the alternative, in a lean hyperlocal newsroom serving a mid-size California city, is not “purely human journalism.” The alternative is less journalism, shallower analysis, and fewer accountability stories. Technology is how Lodi411 covers civic ground that traditional small-market newsrooms can no longer afford to cover at all.

How the pipeline was formalized before disclosure

Before Lodi411 began publishing methodology notes with LodiEye articles in March 2026, the underlying pipeline was taken through a formal review:

• Documented: every stage — source discovery, credibility validation, analysis, drafting, editing, final review, publishing — was written down as a repeatable process with named tools, checkpoints, and human-decision gates.
• Tested against working articles: the pipeline was run against prior and in-progress LodiEye pieces to confirm that the methodology description matched what was actually happening in production, not an idealized version.
• Reviewed for failure modes: known risks of AI-augmented journalism — hallucinated sources, fabricated quotes, false corroboration across model outputs, over-reliance on LLM synthesis without primary-document review — were explicitly mapped to mitigations in the pipeline.
• Aligned with emerging standards: the methodology was compared against disclosure guidance developing at Poynter, LION, Partnership on AI, and the academic AI-disclosure literature, to ensure Lodi411’s practice met or exceeded the evolving bar.

Only after that formalization step did Lodi411 begin attaching per-article methodology notes. Disclosure without a disciplined pipeline behind it would have been a marketing gesture; disclosure after formal review is an accountability instrument.

The article production pipeline

1. Research and source discovery

Lodi411 uses Perplexity AI and Anthropic’s Claude (primarily Opus and Sonnet models) to identify and retrieve sources across federal agencies, California state agencies, municipal records, peer-reviewed literature, regional news reporting, and primary documents.

In this stage AI is used to:

• Locate candidate sources across large public-document corpora (Bureau of Reclamation studies, DWR allocation notices, city council agendas, audit reports, grand jury filings, campaign finance disclosures).
• Retrieve real-time data on deadline events, forecasts, and announcements.
• Identify relationships between documents that manual search would miss.

AI is not used to:

• Decide what is newsworthy.
• Determine the editorial framing of a story.
• Replace direct review of primary documents by the human editor.

2. Credibility validation

Every factual claim that reaches a draft is cross-referenced across multiple independent sources. AI tools assist this process by running parallel checks across a source hierarchy that Lodi411 applies consistently:

1. Federal government datasets (USBR, USGS, Census, BLS, EPA, FBI UCR).
2. State agency publications (DWR, SWRCB, CDPH, CA DOJ, Cal EPA).
3. Peer-reviewed research and academic institutional analysis.
4. Primary institutional documents (city audits, CAFRs, budget books, staff reports).
5. Corroborating reporting from established regional news outlets.

Multiple models verify critical figures independently. A number, date, or threshold that appears in a LodiEye article has been checked by at least two models against at least two independent sources in the hierarchy above, and confirmed by the human editor against the primary document.

3. Analysis and synthesis

This is where AI contribution is most significant and where Lodi411’s disclosure is most explicit. Claude Opus and Sonnet are used collaboratively with the human editor to:

• Identify causal relationships across datasets that span multiple domains (for example, linking Colorado River allocations to Central Valley groundwater stress through the MWD dual-source substitution mechanism).
• Develop analytical frameworks (three-channel models, impact matrices, comparison tables).
• Distinguish direct from indirect effects on the Lodi readership.
• Stress-test arguments by running opposing-view critiques through separate model sessions.

When an analytical framework originates in collaboration with a model, the per-article disclosure says so. When a framing is the editor’s own, the disclosure says that too. Readers can see which analytical moves were human and which were machine-assisted.

4. Drafting

AI assists with drafting, structuring, and formatting, including headline candidates, section organization, transition paragraphs, narrative framing, and house-style compliance. Every draft is edited by the human editor. No sentence reaches publication without editorial review.

On politically sensitive material — allegations, disputed claims, contested characterizations — the human editor enforces a stricter standard: claims are framed as allegations rather than findings, sources of allegations are named, and multiple models review the framing for balance before publication.

5. Editing and final review

Before publication, the completed draft is reviewed by multiple AI models for:

• Factual consistency against the source set.
• Source attribution accuracy.
• Logical coherence of the argument.
• Balanced presentation of contested claims.
• Adherence to the Lodi411 house style specification.

Model review is a check on the human editor, not a replacement for the human editor. Where models flag concerns, the editor adjudicates. Where the editor and the models disagree, the editor’s judgment governs, and the disagreement is sometimes itself disclosed in the methodology note.

6. Publishing

HTML conversion, metadata generation, schema tagging, and internal linking follow the Lodi411 house style specification and are AI-assisted. Publication timing, placement, and promotion are human decisions.

Image generation

Lodi411’s use of AI image generation is deliberately narrow.

Documentary images in Lodi411 articles are either original photography, licensed stock photography with credit, public-domain government imagery with credit, or screenshots of public documents with source attribution. When an AI-generated preview image is used on social media, it is clearly stylized and title-bearing — not a depiction of a scene.

This narrow use reflects a specific editorial judgment: AI-generated imagery presented as documentary content corrodes reader trust faster than almost any other AI misuse, and the marginal benefit to a small newsroom is not worth the cost to credibility.

Data pipeline: datasets, validation, and visualization

A substantial share of Lodi411’s distinctive value comes from its data infrastructure — civic datasets that power story research, interactive maps, and reader-facing visualization tools. AI is used across this pipeline.

Dataset location and acquisition

AI assists in locating authoritative datasets across:

• Financial: municipal CAFRs, budget books, audit reports, bond disclosures, economic development subsidy reports, pension actuarial reports, campaign finance filings.
• Geospatial: parcel data, zoning layers, general plan maps, FEMA flood maps, tree canopy imagery, aerial and drone imagery, transportation networks, utility service boundaries.
• Civic and government: city council agendas and minutes, planning commission actions, building permits, business licenses, code enforcement records, public records act releases.
• Public safety and justice: police incident data, CAD logs, crime statistics, court filings, grand jury reports.
• Environmental: water quality monitoring, groundwater levels, air quality, tree inventories, watershed data.
• Demographic and economic: Census, ACS, BLS, EDD, housing market data.

AI accelerates the identification of the authoritative source (the canonical agency publication rather than a derivative), the most recent release, and the appropriate vintage for the analysis at hand.

Format conversion and normalization

Public datasets arrive in inconsistent formats: PDF tables, scanned documents, proprietary GIS exports, malformed CSVs, HTML tables, legacy file types. AI tools assist with:

• Extracting tabular data from PDF and scanned source documents.
• Normalizing field names, units, and date formats across datasets.
• Converting geospatial data between formats (shapefile, GeoJSON, KML, MVT).
• Reconciling addresses and place names across jurisdictional datasets.
• Generating schema documentation for internal reuse.

Every converted dataset is spot-checked by the human editor against the source document before it is used in an article or published as a reader-facing tool.

Validation

Before a dataset informs reporting or a public visualization, it is validated through:

• Cross-reference against the originating agency’s published summary statistics.
• Internal consistency checks (totals, subtotals, categorical coverage).
• Comparison against prior-period releases to detect schema or methodology changes.
• Outlier inspection with source-document verification for flagged records.

AI runs these checks at scale; the editor reviews the exceptions.

Analysis

AI assists in statistical analysis, including time-series decomposition, clustering, anomaly detection, geospatial overlay analysis, and comparison across jurisdictions. Analytical conclusions are reviewed against primary documents and, where appropriate, discussed with subject-matter sources before publication.

Visualization and reader-facing tools

The interactive maps, dashboards, and visualization tools Lodi411 publishes for its audience — tree canopy maps, crime pattern visualizations, budget explorers, civic meeting trackers — are built on this validated data foundation. AI assists in:

• Generating chart concepts and visualization frameworks.
• Writing and reviewing JavaScript/Node.js code for the mapping and dashboard stack (Leaflet, Google Maps, Kendo UI, Algolia search).
• Drafting user-facing tooltips, legends, and explanatory copy.
• Testing for accessibility and clarity.

The human editor makes final decisions on what is published, what is labeled as provisional, and what is withheld pending further verification.

What we won’t do

To make the boundaries of this methodology explicit:

• We will not publish AI-generated text as if it were reported by a human journalist.
• We will not publish AI-generated imagery as documentary content.
• We will not publish analytical conclusions that the human editor has not independently reviewed against primary sources.
• We will not use AI to fabricate quotes, sources, or events.
• We will not use AI to impersonate community members, public officials, or sources.
• We will not publish a dataset-derived claim without tracing it back to the originating agency document.

Why this matters for Lodi and for local journalism

Small and mid-size cities across California have watched their newspapers shrink, consolidate under out-of-region owners, or disappear entirely. The institutional answer from national philanthropy has been to rebuild traditional newsrooms under nonprofit ownership, at cost structures that require permanent major-donor subsidy and that do not scale to cities the size of Lodi.

Lodi411’s bet is different: that a lean, technically sophisticated, self-funded newsroom using AI responsibly — with transparent methodology, a strict source hierarchy, and unambiguous human editorial responsibility — can cover civic ground in a mid-size California city at a cost structure that community advertising and membership can sustain, without recurring philanthropic subsidy. Forty years of building production software at Microsoft, Apple/Claris, Yahoo, eBay, and several startups informs every technical choice behind that bet. This methodology is how we hold ourselves to it.

Transparency is not a marketing posture. It is the condition under which readers can trust AI-augmented journalism at all. Lodi411 publishes a per-article methodology note with every LodiEye analysis — a practice adopted in March 2026 after the underlying pipeline was formalized — because the reader’s ability to evaluate the work depends on knowing how the work was made.

Questions, corrections, and feedback

Readers who want to examine the source documents behind a LodiEye article, question a methodology choice, or flag an error are invited to contact the editor directly at editor@lodi411.com. Corrections are published openly. Methodology changes are versioned and disclosed.

This standard is a living document. It will be updated as tools change, as our practice matures, and as the broader field establishes norms that should be adopted or challenged. The current version is dated April 17, 2026.