Why we built it
Special instructions have always let you brief the engine before a run — "ignore the manager rows", "our broken-shift gaps under 20 minutes are paid handovers". They work, but they're per-upload: you have to remember to say it every time, and you only find out you needed to after a result comes back with a flag you know is wrong for your organisation.
The feedback we kept hearing was the same: "the engine was right about the award, but it didn't know this one thing about us — and I had to re-explain it on every upload." Endless Learning closes that loop at the exact moment you spot the gap.
How it works
On any timesheet result, every compliance check now carries an Endless Learning button alongside Explain, Recheck, and Mark as Resolved. Click it and a split view opens:
- On the left — exactly what the engine concluded: the verdict, the evidence, the award basis it cited, and the recheck reasoning if you've run one. No guessing about why it flagged what it flagged.
- On the right — your input. Write the nuance or detail the engine was missing, in plain language, the same way you'd brief a new payroll officer.
Save it, and the note is set against your account — not just this timesheet. You can scope a learning to the specific check it corrects, or apply it across every check.
What happens to a learning after you save it
Saved learnings are injected into the engine's context on every future analysis and every recheck you run. They behave like standing guidance: the engine applies them as defaults for how your organisation operates, and when a learning changes a verdict, the evidence quotes your note verbatim — so a verifier can always trace why a check passed or failed the way it did.
Per-run special instructions still take precedence. A learning is your standing default; an instruction on an upload is an explicit order for that run. If they ever conflict, the instruction wins.
You stay in control
Every learning you save is visible right inside the modal, scoped to the check you're looking at. Remove one with a click and it stops applying immediately. Nothing is inferred silently — the engine only learns what you explicitly teach it, and the audit trail records the engine's reasoning at the moment you corrected it.
Where this is going
This is the first step in making the engine genuinely yours. The award rules are universal; your operation isn't. The more results you review, the sharper your account's engine gets — without prompts to re-write, settings pages to maintain, or onboarding calls to schedule. Open a result, find a flag that misses your reality, and teach it. Once.