How the Launch Risk Score is built
A passive check of one public URL. The finding score runs from 0 to 100 and is shown beside how much was checked. 100 means no verified finding among the checks that were completed; it is not a claim that the site is secure. Rubric version 2.0.0, coverage version 1.
The score
score = max(0, 100 − the weight of every failed check)
Weights: critical 40, high 25, medium 10, low 5, info 0. A check that did not run (blocked by robots.txt, a spent budget or a timeout), failed with an error, or does not apply costs nothing and is listed with its reason; the coverage section below says what each one means. If no check could run at all, there is no score. The same results always give the same score.
What the scan does
- Reads only. It loads the page, the scripts the page loads,
robots.txtand a few well-known paths. It never logs in, writes, submits a form or reads a row of data, and it never uses a key it finds. - Supabase checks use only the public anon key the site itself ships: row-count requests that return a number, not data, and two read-only settings requests.
- Two depths. The free scan's quick scan runs without proof of control and requests only what a visitor's browser loads: the page, the scripts the page loads and the
robots.txtfile. The checks that need more are listed as waiting for proof, and a quick scan gives no score. The other requests are sent only once the site's owner has proven control, or for an order. - Polite. It honours
robots.txtand waits at least one second between requests to the same host. - Masked. Secret values in the evidence are masked; a key that is found is reported, never used.
Finding score and coverage
The finding score says what was found. The coverage says how much was checked. Both are shown together, from the same calculation that the scanner, the report and the app use.
- Completed: the check ran to a verdict over what it covers. A verified finding is still a completed check.
- Partial: the check ran but read only part, for example 12 of 40 public scripts. A partial pass means nothing was found in what was read, not that nothing is there.
- Not checked: the check did not run (
robots.txt, a spent budget, a timeout, a refusal or a missing page). It is unexamined, and it is never treated as a pass or as not applicable. - Error: the check ran and failed.
- Not applicable: the check does not apply, and the result gives the reason. This is used only when what it depends on was checked well enough to show it is absent, for example the public scripts were read in full and none names a Supabase project.
- Coverage unknown: a result from before coverage was recorded. Its score is shown as stored and is not recalculated.
When no applicable check is left incomplete, the score is the headline. When some did not, the result is partial and no headline score is given: the result lists what was not examined. When nothing completed, the score is unavailable. Unseen defects are never estimated, and the score is not a probability of anything.
Worked examples, produced by the scoring code:
- One check passes, the other eight did not run. Partial result: 1 of 9 applicable checks completed. The finding score so far (100/100) reflects only what was examined, so no headline score is given.
- Nine of nine checks complete. Finding score 100/100 from the public page scan: 9 of 9 applicable checks completed. Fewer points means more verified findings; it is not a probability of security.
- Scripts read in full, no Supabase project named. Finding score 100/100 from the public page scan: 6 of 6 applicable checks completed, 3 not applicable. Fewer points means more verified findings; it is not a probability of security.
- A stored result without coverage. Coverage unknown: this older result did not record which checks ran.
This is a scan of the public page, separate from an audit of the code, and it applies no framework-specific expectation. A backend file that is not visible from the page is missing coverage, not evidence that a policy is in place.
The checks
- A failed check shows
- It does not show
Service-role token in public JavaScript
critical, −40 points js_service_role_token
- A fetched public script or page contains a Supabase service_role JWT or sb_secret_ key.
- That the token is still valid; the token is never used.
.env-style files served
critical, −40 points env_files_exposed
- A well-known env path answered a GET with KEY=VALUE-style text.
- What the values are (never recorded) or whether they are live.
Secret keys in public JavaScript
high, −25 points js_secret_keys
- A fetched public script or page contains a string in the format of a provider secret key (audit secret patterns).
- That the key is live or has any privilege; the key is never used.
Tables answering an anon count request
high, −25 points supabase_tables_anon_count
- A table named in the scripts answered an anon-key HEAD request with a row count above zero.
- What the rows contain (never requested), or the state of tables not named in the scripts.
Auth settings readable without login
medium, −10 points supabase_auth_settings
- The project auth settings endpoint answered without login and reported sign-up open with email auto-confirm on.
- That any account was created (no sign-up is ever submitted).
Storage buckets listed without login
medium, −10 points supabase_storage_buckets
- The storage bucket list endpoint answered an anon-key request with at least one bucket.
- That any object is readable (objects are never listed or fetched).
Security headers
medium, −10 points security_headers
- The page response lacks at least one of: HSTS (https), CSP, X-Content-Type-Options nosniff, frame protection, Referrer-Policy.
- That the site is exploitable; headers on other pages or paths were not read.
Source maps served
medium, −10 points source_maps_exposed
- A script's .map file answered a HEAD request with a non-HTML content type.
- What the map contains (not downloaded).
Supabase project and anon key present
info, no points supabase_exposure
- The page or its scripts name a Supabase project URL and an anon or publishable key (public by design).
- That any data is reachable; see the table check.
Limits
One page and its scripts, not a crawl. Keys are found by their format, so a key assembled at runtime is missed. A good score is not a clean bill of health: the full App Audit reads the code itself.