Help wanted
This site gathers advice from public community discussions, put together by a volunteer with the help of an AI agent. Nothing on it has been checked by a professional yet. The most useful help is a review from someone who works in security, banking, fraud or compliance.
How to contribute
Contributions go through the site's page on GitHub. If you don't use GitHub, email ahumanflourish@gmail.com instead.
- To report a mistake or suggest something, open an issue. No code needed: say which page, what is wrong or missing, and a source if you have one.
- To make a change yourself, open a pull request. The second half of this page explains how each kind of change is made.
- To find something to work on, see the list of open tasks.
- To volunteer as a reviewer, email ahumanflourish@gmail.com or open an issue, saying which guides you can review and what your background is. See the next section for how reviews are credited.
Volunteer as a reviewer
This is the most useful thing you can do. Every item currently says Checked by: Nobody yet, and that label only changes when someone with relevant professional background checks the item.
Reviewers wanted:
- Security practitioner, for the personal-attack and institutional items.
- Finance or compliance professional, for the refund-rule and deposit-insurance items (Regulation E, broker guarantees, FDIC, NCUA, SIPC).
- Someone to keep the open questions current, as new discussion and evidence appear.
What a "Checked by" credit requires
A "Checked by" credit tells readers that a professional looked at an item, so it is not given on someone's word alone. All three of these are required:
- Something checkable about you. A public professional profile or an employer page that names you, which a maintainer can confirm belongs to you. It is linked on the item, so readers can check it too. Describing your own background is not enough, however plausible.
- Your own explicit request to be credited. You have to ask, in your own words, to be named as the reviewer of that item. Nobody is credited because they made a helpful comment.
- A statement of exactly what you checked. For example "the Regulation E deadlines in the 'why' section", not "the item".
A helpful comment alone cannot grant a credit. The label stays "Nobody yet" until all three requirements are met and a maintainer has confirmed the first one.
Once they are met, a maintainer sets the item to expert_reviewed and records the review on it:
review_status: expert_reviewed
reviews:
- name: Your name, as it appears on your public profile
background: Your role, plus the address of the profile or employer page the maintainer confirmed
date: 2026-11-20
scope: Exactly what you checked, e.g. "Reg E timelines and the authorized/unauthorized distinction"
A review says what was checked, not that the item is guaranteed correct. Reviews older than six months are shown as due for re-review. A reviewer can ask at any time for the credit to be removed.
Links and phone numbers in steps
Readers act on the links and phone numbers in the steps, so these get extra care:
- Any change to a link or a phone number in a step must be checked by a maintainer against the institution's own website before it is merged, whoever proposes it and whatever else the change contains. A review credit does not exempt a change from this.
- Step links must stay on the approved list of official domains. The build enforces the list and fails on a step link to any other site. Adding a domain to the list is a separate change that a maintainer makes after checking that the domain really belongs to the institution.
- Do not bundle a link or phone-number change with other edits. Send it on its own so it can be checked on its own.
Making changes yourself
The rest of this page is for people editing the site's files. All the content lives in a handful of plain text files in the data/ folder: one for the steps and guides, one for incidents, one for open questions. Edit the relevant file and open a pull request. An automatic check runs on every pull request and says what is wrong if an entry is malformed.
Add an advice item
Add an entry to data/advice.yaml. Copy an existing item as a starting point.
Required fields: id, threat_model, title, advice, why, protects_against, confidence, review_status, updated.
Rules for every item:
- Say why it works and what failure it protects against.
whyis the mechanism. An item without a mechanism is not ready. - Pick the confidence honestly.
established: rests on law, regulation, official terms, or documented precedent. Needs at least one entry undersources. If you don't have one yet, setneeds_source: true; the site shows a "Needs source" tag.reasoned: follows from a clear model of the mechanism, but not directly tested.speculative: plausible, with weak or contested evidence.- Use
confidence_notewhen parts of an item differ, such as an established rule applied to a new situation.
- New items start as
review_status: unreviewed. - Paraphrase thread comments and link to them under
thread_refs. Don't quote at length. - Check for duplicates first. If the advice already exists, add your source or comment link to the existing item.
- If it's genuinely uncertain, it's a question, not advice. Put it in
data/questions.yaml. - No personal financial details about yourself or anyone else, including what a named commenter says they keep or hold.
- Set
updatedto today's date. Date any number that will go stale ("as of Nov 2026").
To put an item on the printable checklist, set checklist: true and write a one-sentence checklist_text that someone could act on without reading the guide.
Dispute an item
Don't delete or silently rewrite advice you disagree with. Add or edit the item's sides:
sides:
for: Who argues for it and why, in a sentence or two.
against: Who argues against it and why.
synthesis: What a reader should do given both.
Link the comments or sources for each side under thread_refs or sources. The site shows both sides and the synthesis, and tags the item "Disputed". If you think the confidence level is wrong, say so in the pull request with the evidence.
Add or update an incident
Add an entry to data/incidents.yaml. The log is a small set of real, dated events that inform the advice, not a news feed.
kindiscurrentfor a Hackening-era event, orprecedentfor an older event used for calibration.- Rate
ai_involvementseparately for each stage you have information on (discovery,attack,agent_behavior), usingconfirmed,suspected,unlikelyorunknown. Useconfirmedonly when the victim, the supplier, an official investigator, or a reputable outside organisation such as an established security firm says AI was used in that stage of that event. A report from unnamed sources or an individual researcher with no established organisation behind them issuspected. Put who said it innote. - Open the source before you write a figure, and put a date on any figure that can change ("as of 5 Nov 2026"). Do not rely on a summary of a source, and do not cite an AI-generated summary.
- Do not credit an AI-generated comment as a contributor or as support for an estimate. If one has to be mentioned, call it "an AI-generated comment in the thread".
- List the advice it informs under
related_advice. - Set
needs_research: trueif details are still missing or the cause is still under investigation.
Picture on the Incidents page
type picks the drawing shown with the story. Choose the one that matches what happened, not who did it. If the story has a real count (so many hit, out of so many), add count as well and the feed draws counted dots instead.
outage: a service stopped working and the cause is a failure, or not yet known.overload: a service failed because it was overwhelmed or a change went wrong; not an attack.lockout: attackers locked an organization's own systems until it pays (ransomware).heist: money stolen from a bank or other institution, often by way of a supplier.crypto-theft: crypto taken from wallets or an exchange.devices: many internet-connected machines of the same kind taken over.software-flaw: a flaw found in software that very many systems run.supplier: one supplier was attacked and the customers that depend on it went down with it.one-to-many: one attacker or one tool hitting many separate targets.agents: AI agents doing things inside accounts they were given access to.breakout: AI agents leaving the limits they were meant to stay inside.data-theft: personal or company records copied out.records-lost: the record of who owns what was lost or cannot be matched up.advisory: an official warning covering many possible targets, with no single event behind it.
Add a basics link
data/basics.yaml links out to existing security guides. Don't rewrite them. The commentary field says what matters more or less under the new threat model.
Check your change
npm install
npm run build # validates every data file, then builds the site into dist/
npm run check-links # checks that every URL resolves
The build fails, and says where, on a missing required field, an unknown field name, an unknown value for a rating, or a reference to an advice id that doesn't exist.