You already have the material: documents, notes, links, data. The work is choosing the output form and the steps between — report, brief, presentation, or memo — and getting there without losing the evidence. This hub maps the territory and routes you into a working method.
Drafted by an AI agent. Read in full and signed off by Lizikk. Evidence last verified 2026-09-08.
Turning research into a finished artifact is not a writing problem — it is a sequencing problem. The material exists; what's missing is the order of operations: pick the evidence worth keeping, decide who the output is for, choose the form that fits the decision at hand, and check the result against its sources before anyone acts on it. Teams that skip a step ship a beautiful artifact nobody trusts, or an honest artifact nobody reads.
Every path through this territory crosses the same six processing steps, whatever the output:
| Inputs | Processing | Outputs |
|---|---|---|
| PDFs and documents · web pages and links · interview notes and transcripts · raw data and metrics | Filter the evidence · define the audience · build the structure · choose the form · human approval · quality check | Executive report · visual brief · presentation · web page · decision memo · social content pack |
Two failure modes account for most bad deliverables, and both are avoidable:
Answer two questions and this routes you to a working method — with the output form pre-selected so the tool starts from your goal, not a blank page.
Each of these is a documented method with reusable assets — steps, prompts, checklists, and the real inputs and outputs from our own runs. Not import files, not templates: methods you can audit.
The map above covers more ground than these two resources. In progress: a compression workflow for long documents to visual briefs, a decision matrix for report-vs-presentation-vs-web-page, and a planner tool that outputs a full deliverable plan from your constraints. Each ships only when it carries the same evidence standard as the two above — a documented method, real runs, and the original files.
Paste a draft and get two numbers: words per non-paragraph element, and bold spans per 100 words. Handles Chinese, where counting words takes a stated convention before any score exists — ideographs are counted by character and converted with a stated, arbitrary divisor. Runs in the page. No account.
We tested 48 _redirects rule forms on a local Pages simulator, three requests each, with four controls. 26 forms never took effect — but only 11 of those were silent at every layer. Those 11 share one shape, and it is not the shape the documentation warns you about.
Method v1.0
Six rules for pinning down an AI evaluation so tightly that you cannot bend it after the results arrive — each one traced to the specific independent-review round that caught us breaking it. Includes a method template, a pre-flight checklist, and a deviation-log format that stays honest when the method turns out to be wrong mid-run.
Paste a Cloudflare Pages _redirects file and see which rules were observed to fail — separated by whether anything warns you. The dangerous class is the one that gets counted as valid and never matches. Runs in the page. No account.
The 301 is the easy part. The hard part is knowing which URLs disappeared — which needs an append-only ledger of what you actually shipped, a gate that reads the build product rather than the sitemap, and a redirect written in a form that does not fail silently. Method, checklist, ledger format, and the four source and destination forms that failed in our matrix — two of them named at parse time, one of them silent at every layer.
A 14-point audit for any report — AI-drafted or not: sources exist, quotes and numbers are real, gaps are declared. Run it in the page, copy the result, take the failed items into a revision. No account.
A documented method with reusable assets for turning messy research notes into a decision-ready executive report — steps, prompt, skeleton, acceptance checklist, and a worked example with the real input and the real Studio-generated output.
