Methodology
Last updated: 2026-07-13
AutoKaam is a small editorial operation. One editor, a transparent pipeline, and a hard rule against fabrication. Here is exactly how an article goes from source to published page.
Six-stage pipeline
1. Source identification
Each article begins with a primary source, vendor announcement, SEC filing, official press page, named analyst note, first-person LinkedIn / X / Reddit post by an operator, GitHub release notes, or a public conference recording. Articles without a documented primary source do not enter the writing pipeline. The source URL is captured in frontmatter (sourceUrl) and renders as the source-quote block on the article page.
2. Triangulation
Numbers (pricing, valuations, headcount, model context windows, benchmark scores) are cross-referenced against at least one second source, vendor pricing page + the specific announcement, or two independent operator reports. If sources disagree, both ranges are published and the disagreement is called out in the article body. Where the source itself is rumor or unverified, the article is tagged with a clear 'Unverified' kicker.
3. AI-assisted drafting (disclosed)
First-pass research scaffolds, extraction, and structural edits may use AI assistance. Models and routes can change, so no model output is treated as evidence. Claims must resolve back to primary sources, and release requires explicit editorial review. The repository documents the current deterministic checks; AI never replaces verification or final judgment.
4. Human editorial review
Explicit owner review is a release requirement. The review checks factual accuracy against cited sources, internal consistency of numbers and claims, removal of fabrications (no invented quotes, fake company names, or hallucinated benchmarks), source attribution, and whether the operator verdict is supported rather than merely provocative.
5. Quality gate
Two deterministic checks, scripts/seo/ai_tell_gate.py and scripts/seo/quality_gate.py, validate style, frontmatter, evidence signals, and operator-verdict fields. The release command also runs lint, strict TypeScript checks, regression tests, and the production export. A failing check stops the release; passing does not deploy automatically.
6. Release + correction policy
Scheduled GitHub Actions are currently disabled. Reports publish through an explicit, gated release after review rather than on a daily quota. Stale pricing, retired vendors, or superseded model versions are updated or retired when verified. Material corrections carry a visible dated note on the original URL.
What we do not do
- ✕We do not republish vendor press releases as 'news' without adding operator analysis.
- ✕We do not run anonymous personas. After the 2026-04-29 persona collapse, every byline is either Aditya Sharma (analytical) or AutoKaam Editorial (desk drops).
- ✕We do not invent operator quotes, testimonials, or named sources.
- ✕We do not publish AI-generated content without human editorial review, even for short FAQ entries or headlines.
- ✕We do not accept sponsored placements inside article bodies. Optional display ads load only after reader consent.
- ✕We do not rank tools based on advertising relationships. Comparison pages present feature, pricing, and procurement data without commercial bias.
- ✕We do not chase head-term SEO at the cost of editorial substance. Narrow, source-anchored coverage beats broad, generic guides.
Source-of-record
The repo behind AutoKaam is public: github.com/Declan142/autokaam. Every article's edit history, commit log, and sourcing pipeline lives there. The qa-gate script, content schema, and prompts are all open. If you need to audit a claim, the trail is open.
For article-level corrections, contact [email protected]. For author bibliography, see Aditya Sharma's profile. For the editorial standards layer, see the editorial policy.
Owner: Aditya Sharma · Contact: [email protected] · This methodology is reviewed quarterly and on every significant change to the publishing pipeline.