No person. The review is performed by SASFA's own agents running open-weight models on SASFA-controlled servers in South Africa. No call leaves those servers to any outside service. The environment is created for your engagement and destroyed at the end.

Who stands behind the report?

SASFA. No human reads your code, but SASFA designs, validates and signs the method, and signs the report.

What can a code review not grade?

Anything that needs the system running: the statistical quality of the generator, settlement across every bet type at volume, recovery from a power cut mid-bet. Where sasfa.md gives a runnable harness and tests, the agents run them and grade on the results. Where it does not, the clause is marked as requiring execution, and the accredited laboratory does it.

Is the pass summary a certificate?

No. It states that a named build passed SASFA's automated SANS 1718 conformance review on a given date. Certification is the NRCS letter issued on a SANAS-accredited laboratory's report; SASFA is not SANAS-accredited.

Can you review only part of the repository?

No. The review needs the repository that produces the production build, so the report can name the build it was true of. sasfa.md tells the agents where to look first; they still look everywhere, and the report states any outcome-generating or money-moving code found outside the declared paths.

Testing

How the testing works

Q: What does it cost?

R10 000 flat for the first report. Re-tests are priced on the outstanding clauses only, and never more than the first report.

Automated, private and repeatable. SASFA's agents review your repository clause by clause against SANS 1718 on servers nobody else can reach, report what passed and what failed with the fix for each, and leave nothing behind.

Request a test What SANS 1718 is

The process

01 NDA A mutual NDA is signed before anything is shared.
02 Repository access and sasfa.md You grant read access to the repository that produces your production build, and add a sasfa.md file that maps the code: where outcomes are generated, where bets are recorded and settled, where money moves, where limits and exclusion are enforced, how the build is hashed, and how the tests and generator harness run.
03 Automated review SASFA's agents clone the repository into a fresh, isolated environment and review it clause by clause against SANS 1718, guided by sasfa.md but not limited to it. Where your tests and harness can be run, they are run.
04 Private report Every applicable clause graded met, partial or missing, with the evidence and, for every partial, exactly what is satisfied and what is not. Each failed clause carries the fix. The report goes to you and nobody else.
05 Fix and re-test You fix what the report lists. Re-tests cover only the outstanding clauses and are priced on them.
06 Pass summary When every applicable clause is met, a public pass summary is published here, naming the product, the build hash, the parts and editions, and the date.
07 Deletion The environment is destroyed, you revoke repository access, and you receive a system-generated deletion report listing what was removed. Nothing of your code remains on SASFA systems.

Price

R10 000, flat, for the first report. Re-tests are priced on the outstanding clauses only, never more than the first report. The pass summary and the deletion report are included.

Privacy

No person reads your code. The agents run open-weight models on SASFA-controlled servers in South Africa, with no call to any outside service. Each engagement gets a fresh, isolated environment that is destroyed when the report is accepted. You revoke repository access yourself, and the deletion report lists what was removed.

What is reviewed

Odds and settlement

Settlement logic for every bet type against the published rules.

Outcome generation

One declared generator, isolated from everything else, with its seeding, mapping and harness.

The bet record

Each bet stored with stake, odds, selection, time and result, append-only.

Player funds and limits

Wallet, deposits, withdrawals, deposit and loss limits and self-exclusion enforced in the system.

Software integrity

Build reproducibility, hashing and the verification that production matches the reviewed build.

Events, meters and recovery

The significant-event catalogue, meters nobody can edit, and recovery of bet state after a failure.

Against SANS 1718 Part 4 (wagering record-keeping), Part 10 (server-based systems), the outcome-fairness clauses of Part 2, and the control baseline of Parts 3 and 9. The edition reviewed against is named on the report.

What goes in sasfa.md

A short markdown file at the root of the repository. It is a map, not a boundary: the agents review the whole repository and use it to know where to start.

Product: what it is, bet types and products, and which generate their own outcomes
Outcome generation: the module that generates results and the harness that exercises it alone
Bets: where a bet is recorded, priced and settled
Money: the wallet, ledger, deposit and withdrawal paths
Player protection: where limits, exclusion, age and identity gates are enforced
Build: how the production artefact is built and hashed, and the current production hash
Tests: the commands that run your tests and the generator harness
Out of scope: paths that are not part of the betting system, and why

What the report says

One row per applicable clause: met, partial or missing, with the evidence the agents found, a file and line, a test result, a harness run. A partial says exactly which part is satisfied and which is not. A missing clause says what would satisfy it. Clauses that need the system running, and could not be run from sasfa.md, are marked as requiring execution. Clauses that belong to the licensed bookmaker rather than the software are marked as the bookmaker's. The report is private and is never published.

After a pass

The pass summary is public: product, build hash, parts and editions, date. It is not a certificate. Any change to the build changes the hash, and the summary is true only of the hash it names; a re-test on the changed clauses produces a new one.

Request a test