Every instance writes an append-only log of the person, the step, the values, the decision, the notifications and the acknowledgement of each write-back, with the version of the app.
| Entry | Holds | Example |
|---|---|---|
| Step | node, role, person, start and end time, app version | Decide · buyer · Priya Sharma · 14:05:41 · v3 |
| Read | handler, filter, rows returned, access rules applied | dh_po_lines · group 101 · 31 rows · policy PURCH-101 |
| Value | components shown and entered, with their values at the time | dd_decision = split · txt_note = 'supplier capacity' |
| Decision | the criterion that held and the arrow taken | decision: split → re-source |
| Notification | channel, recipient, card, buttons pressed | Teams · supplier contact · asked · replied 11:30 |
| Write-back | interface, values sent, acknowledgement or error | BAPI_PO_CHANGE · 4500012345 · RETURN S · 14:06:12 |
| Override | administrator action with justification | released stuck instance · 'system maintenance' · E. Clarke |
The log is append-only and hash-chained, so every entry stays as written; a correction is a new entry that refers to the old one. Each entry carries the app version and the hierarchy version, so a decision taken under last year's approval bands is read with last year's bands. Files uploaded during a step are referenced from the log and kept under the retention rule.
Auditors read the log themselves, on an auditor screen that shows instances, decisions and configuration history with the same filters a process owner uses. In the showcase estate the external auditor read 5 entries during the quarter's audit and sent the finance team zero requests.
The showcase estate's SOX certification of a revenue view.
The pattern is a collaborative app with three levels in series and a rule that the third level is external when the view is in scope for the regulation. The review pack is a set of handlers and a generated screen. The same pattern serves a data product publication (with a hard block below a trust score), a policy change (with an automated impact analysis as stage 2) and a schema change on personal data (with legal and the data protection officer as stages).
A person sees in an app what the source systems allow them to see. The data handler applies the source's own policy when it reads: rows by organisation, plant or region; columns masked or hidden; values redacted. On top, the app adds the level's or the role's scope. The log records, for every read, which policies applied, so the answer to 'what could this person see on that day' is in the record.
Configuration rights are separate from run rights: who may edit an app, who may publish, who may switch write-back on per object, who may override a stuck instance. Each is a role of the platform, logged like everything else. The trust center sets out the access model, the AI boundary and the certifications.
| Right | Who | Logged as |
|---|---|---|
| See rows and columns | from the source policy plus the role scope | read entry with policies applied |
| Act at a step | the role or level, alternates, delegates | step entry with both names under delegation |
| Edit the definition | configurator role for the app | definition change with version and note |
| Publish and promote | process owner; application team at the gate | promotion entry with approval |
| Switch write-back on | application team, per object, after a test write | write-back enablement entry |
| Override an instance | administrator | override entry with justification |
Every change to a hierarchy, a screen, a handler, a criterion or a variable is a versioned entry with the person, the time and the note. Retiring a level or a screen is a flag with its dependents listed; the definition stays readable. The history of an app answers an auditor's question about a rule the same way the log answers a question about a decision: what was the rule on that date, who changed it, and why.
Global variables carry the values regulations care about, such as approval bands, so a change to a band is one logged entry that every app inherits from its effective date.
Yes: per app, per instance or per period, as files with the hash chain, or read directly on the auditor screen.
Nobody. A correction is a new entry that refers to the earlier one. Retention rules and erasure requests remove personal data by rule, with the removal itself logged.
Personal fields are tagged in the catalog; an erasure request runs as an app that finds every occurrence, removes or masks it and records the run. The showcase estate completed one across 847 records in 47 minutes with the data protection officer's signature.
The trust center lists the current status and dates of SOC 2, ISO 27001 and the SAP program, and how to request the security packet.
Bring one process and the systems it touches. In the first session we register the data handlers, draw the levels or the lanes and run the first screen on your live rows.