Make the invisible workflow visible.

The timesheet project is not blocked by coding effort. It is blocked by turning Kim's day-to-day judgement into a clear rulebook, then connecting that rulebook to Launch27, Toby's sheet, Slack, and the dashboard.

Daily approval274 records
Rules we knowpartial
Kim judgement still neededhigh
Toby handoff93 edits/day
Column meaningsunresolved
Record identityblocked
Proxy writesconfirmed
Update endpointready
Read + edge testspending
Current pain
4-6h

Kim's daily time across approval clicking and post-Toby edits.

Daily volume
274

Approximate daily bookings Kim reviews in Launch27.

Post-Toby
93

Approximate manual edits after Toby's adjustment process.

Target state
5m

Kim verifies exceptions and mismatches rather than doing every click.

Past State: Human Effort Carries The Workflow

Launch27 stores the raw entries, but Kim supplies the interpretation. Toby supplies adjustment logic separately through the sheet.

1

Field staff use Launch27

Check-in/out, travel, distance, and timing are entered inconsistently.

2

Kim reviews one-by-one

She checks duration, Slack, missing logs, staff type, job type, and manual duration.

3

Kim exports CSV

She formats the sheet and prepares daily tabs for Toby.

4

Toby adjusts sheet

His script populates adjustment columns, then the tab becomes ready for Kim.

5

Kim edits Launch27 again

She applies adjustments record-by-record, then mentally verifies completion.

Future State: Automation Does The Routine Work

The future flow keeps Kim as rule owner and exception reviewer. The system handles repeatable reads, writes, handoffs, and verification.

1

Pull from Launch27

System reads pending time logs and supporting booking fields where available.

2

Apply Kim's rulebook

Clean cases are approved; uncertain cases become dashboard exceptions.

3

Prepare Toby sheet

Daily export is formatted and Toby is notified when ready.

4

Apply Toby edits

Once done, the system writes approved adjustments through Mark's proxy.

5

Verify and surface mismatches

Kim sees only exceptions, mismatches, and anything the system refuses to guess.

Before / After By Process

There are three distinct processes. The first saves the most time and must be understood first.

Past: Kim as the processing engine

Daily approval

Kim clicks and edits hundreds of rows using SOP plus experience.

CSV and sheet prep

Kim exports, moves columns, formats tabs, and sends to Toby.

Post-Toby edits

Kim reads adjustment rows and manually updates Launch27.

Verification

Mostly manual confidence and spot checks inside the same fragile workflow.

Future: Kim as rule owner and reviewer

Daily approval

Automation approves only cases covered by Kim's rulebook.

CSV and sheet prep

System prepares the sheet consistently and notifies Toby.

Post-Toby edits

System applies known edits through the proxy after Toby is done.

Verification

Dashboard shows matches, mismatches, audit history, and exceptions.

Data Flow: Where Information Moves

The core identity problem is simple but critical: Launch27 writes use (booking_id, team_id). Toby's sheet must preserve or recover that identity.

Launch27 source of truth

Raw time logs, statuses, check-in/out, enroute times, distance, booking details, manual duration.

booking_id team_id manual_duration?
Read via proxy if available; otherwise CSV/report fallback.
Store mirrored records, decisions, exceptions, run logs, audit history.
Write approved changes back through Mark's proxy.
Timesheet System future control layer

Supabase + dashboards.maid2match.com.au hold the local workflow, audit log, exceptions, and verification results.

rule outcome audit trail verification
Launch27 time logs + bookings Proxy reads? writes confirmed Supabase records + audit Dashboard exceptions + verification Google Sheet Toby adjustments Slack proof + notifications read mirror show export read edits

The Three Processes

Keep these separate. They are related, but each has a different source of uncertainty.

Process 1

Daily Approval

Biggest time saver. Blocked by Kim's rulebook: below schedule, special staff, job types, whole-day decisions, Slack proof, manual duration.

Process 2

CSV Export + Toby Handoff

Smaller time saver but needed for the full workflow. Blocked by final sheet layout and notification preferences.

Process 3

Post-Toby Edits + Verification

High value, but blocked by column meanings, sheet row identity, Toby done trigger, proxy read/verification behavior.

Who Does What

A swimlane view helps separate operational rules from technical plumbing and sign-off.

Person / System
Approval rules
Sheet handoff
Toby edits
Verification
Escalation
Kim
Owner Supplies the real rules and edge cases.
Currently formats and sends the sheet.
Currently applies Toby's changes in Launch27.
Reviews exceptions and mismatches.
Contacts FS, support, dispatch, TLs.
Kiara
Confirms quality and payroll risk tolerance.
Confirms finance reporting impact.
Helps define working paper / verification standard.
Oversight Defines acceptable confidence.
Escalation / finance sign-off.
Toby
No direct rule ownership.
Receives prepared sheet.
Owner Runs adjustment script / populates columns.
Needs clear "done" signal.
Clarifies sheet structure.
Automation
Applies only explicit rules.
Exports and formats once layout is final.
Writes changes through proxy.
Re-pulls and compares values.
Surfaces exceptions; does not guess.
Mark / Proxy
Provides write/read capability.
No sheet role.
Write endpoint is confirmed.
Read endpoints / partial failure still need testing.
Rate limits and auth/audit need clarity.

Decision Board

This is the short list that explains why the project feels stuck. The next meeting should unlock the left column.

Must answer with Kim
Q09 Experience-based edge cases not in SOP.
Q18 Below-schedule and within-10% semantics.
Q19 Special staff rules.
Q23 Final Toby spreadsheet layout.
Q54 Column H / L meanings.
Q55 Whole-FS-day vs row-level decisions.
Jason / Toby next
Q26 How Toby signals the day is done.
Q34 What revert actually does.
Q56 Sheet row maps to (booking_id, team_id).
Q57 What if humans edit during automation.
Proxy tests separate
Q50 Read time logs and booking manual duration?
Q51 What does active:false do?
Q52 Array partial-failure behavior.
Q44 What user does Launch27 show for proxy edits?

Rulebook Areas Kim Needs To Make Explicit

This is the meeting heart. We need to turn "Kim knows what to do" into written rules.

Area What Kim decides today What the system needs Why it matters
Duration Within 10%, over schedule, under schedule. Precise thresholds and actions. Core approval logic.
Proof Slack extension, support update, manual duration. Where to search and what counts as valid proof. Prevents bad auto-approvals.
People Contractors, TLs, flat-rate staff, exceptions. How to identify staff category. Some logs should be skipped, not approved.
Job types Inductions, one-on-ones, travel, late cancel, bond clean. How to identify each type and cap values. Rules depend on context, not just time.
Grouping Whole-day and multi-staff checks. When rows must be considered together. Prevents row-by-row automation from doing the wrong thing.

Tomorrow's Kim/Kiara Session

Keep it practical. Ask Kim to teach the current process, not to imagine automation.

Best framing for Kim

"We are documenting how you already decide what to approve, edit, skip, or follow up. Short examples are enough. If it depends, tell us what it depends on."

What not to ask

Do not ask Kim how the automation should be designed. Ask what she sees, checks, decides, edits, and escalates today.

Part 1

Approval rules

Walk through normal, over, under, missing, and special cases. Capture examples.

Part 2

Evidence and data

Where she checks Slack, manual duration, booking status, staff role, job type, and whole-day context.

Part 3

Toby handoff inputs

Clarify what Kim knows about the sheet, then take unresolved sheet questions to Toby.

Milestone View

The project becomes easier once each milestone has one concrete output.

1. Kim/Kiara session

Output: explicit approval rulebook and examples.

2. Toby session

Output: final sheet columns, row identity, and done trigger.

3. Proxy tests

Output: read capability, active flag meaning, partial failure behavior.

4. Brief update

Output: final pre-build handoff document with no hidden judgement calls.

5. Build starts only after this

Output: scaffold and first shadow-mode workflow, once rules are known.

One-Sentence Mental Model

The simplest way to explain the whole thing.

We are replacing Kim's repeated Launch27 clicking with a system that applies her documented rules, refuses uncertain cases, hands clean data to Toby, writes approved edits back through the proxy, and gives Kim a dashboard for the few things that still need human judgement.