Every Figure Read From What Is Running

Read the settings, the schedules, the logs and the licence records off the running systems, and write a source sheet naming where each number came from.

Leave a field with no source visibly empty, with the reason on screen: a zero says checked and nothing there, empty says not yet looked, and the next step differs. Show passwords and access keys as a name and a location. Keep business figures such as member counts and revenue off the view, and settle compliance in the same pass.

Plans Written Specific Enough to Fail on Paper

Write the plan with the exact screens, the exact data fields and the exact sequence, so a wrong assumption shows while it is still on paper.

Start from a proven build and treat it as the specification: copy its structure, screen layout and navigation, and change only the visual appearance per customer. Follow the same fixed order on every project, and write down what each run taught once, in the file the next run reads.

Every Plan Draws on Systems Already Built

Bring the domain knowledge into the reading. Scheduling, bookkeeping, media and estate systems have each been built and operated here, so the pass asks the questions that decide those builds from day one: which record is the master, where the write path runs, what the existing licence and permission model already enforces.

Carry the recorded failure modes over as constraints on the new plan: an import that deletes every record it is not given, an asset path that yields a silently empty library rather than an error, a promotion that counts only once the digest matches. Each one enters the plan as a step, and whoever measures the result never changes it.

5
reader agents over the source artifacts
6
parallel build tracks, one shared contract
9
questions mapping each task to one route
3 of 5
data files shipping empty on purpose

How the Facts Become a Plan Six Tracks Can Build Against

Read the Evidence 3 steps
01

Rule Zero: Every Figure Cited to an Artifact

Transcribe every container name, port, image tag, domain, database, hour and euro from an artifact the repository already holds. The named sources are the deployment manifest, the compose files, the edge site files, the time ledger, test reporter output and the commit history.
02

Provenance Per File, Absence Shown as Absence

Write a provenance document stating, per data file, where each figure came from, and for a file that ships empty why empty is the correct state: no test-report generator, no counterparty correspondence, no invoice ever issued. Render a field without a source as a designed empty state with the reason on screen.
03

Two Records Compared, Every Disagreement Listed

Compare the records against each other and list where they disagree: a stale manifest, a DNS record pointing at a box no longer owned, a runbook marked unexecuted that the zone files say is done. After an infrastructure migration, re-derive every stack, service, database and drift item from the manifests.
Fix the Contract Before Anyone Builds 3 steps
04

One Scaffold Contract, Interfaces Stated Verbatim

Fix one shared scaffold contract before any build track starts: the file tree, the owning track per path, and every shared TypeScript interface written out verbatim. State the interfaces rather than describe them, so a wrong signature is visible before the first track runs.
05

Six Partitioned Tracks, One Owner Per Path

Run the workflow in its fixed order: brand capture, five readers over the source artifacts, one shared contract, six partitioned build tracks, verification, then four review lenses. Treat the reference implementation as the specification: copy structure, card anatomy, group disclosure, wrapper widths and the route table, and swap the skin. Each track owns its own paths.
06

A Plain Email Becomes a Documented Work Item

Pull the request from the mailbox with Microsoft 365 mail intelligence inside the repository session, analyse it against the existing codebase, and write it up as a documented work item. A five-item batch that arrived on 2026-08-04 shipped as one commit to a preview address running the production containers against separate test data, promoted once the requesters agreed.
Beyond the Codebase 2 steps
07

The Same Readers Run Over Material That Is Not Code

Run the same pass over document folders and media collections. Consolidating four media products into one repository recorded the exact source commits and per-path provenance in a history document, with the timeline reconstructed for those paths and the asset-root failure mode written down: a wrong asset path yields a silently empty library. Each collection ends with the same verdict per item, with the reason beside it.
08

Time Records and Test Output Are Artifacts Too

Reconstruct the engagement from the time ledger: per-phase money and asset detail, per-repository hour panels, declared and derived hours reconciled against each other. A figure with no recorded session behind it stays off the view. Take the quality facts from test reporter output per application, and derive untested surfaces as gaps.
Baseline and Compliance 2 steps
09

One Observation Week, Then a Route Per Task

Have everyone doing office work record their recurring tasks for one week, part-time staff, assistance, accounting and dispatch included, because whoever performs a task knows its duration. Score each task for minutes per run, runs per month, follow-up effort and six suitability criteria, then walk the nine-question decision path in order, first match winning.
10

DSGVO and EU AI Act Settled in the Same Pass

Review the systems already in production against DSGVO and the EU AI Act: classify the risk tier and identify where personal data is processed. Extend the review to an EU-hosted operating concept, the practical difference between team and enterprise licensing terms, the data-privacy conditions per connected service, and an approval map naming who signs off what.

Frequently Asked Questions

What does a fact-gathering pass actually hand over?
Three things. A scaffold contract fixing the file tree, the owning track per path and every shared interface verbatim, so six tracks build in parallel. A provenance record naming the artifact behind every figure. And a drift list of what disagrees between two records, such as a stale manifest or a DNS record pointing at a box no longer owned.
What happens to a number that has no source?
It renders as a designed empty state with the reason shown on screen, and the provenance document says why empty is correct. On one build, three of five data files shipped empty on purpose: no test-report generator existed, no counterparty correspondence existed, no invoice had ever been issued.
Does this only work on a codebase?
The same readers run over media libraries, document folders, and time and billing records. A consolidated media repository carries per-path provenance and exact source commits, and a billing journey reconstructs an engagement from the time ledger so declared and derived hours agree. A figure with no recorded session behind it never appears.
Is the DSGVO and EU AI Act review a separate project?
It runs in the same pass, before AI reaches company data. The systems already in production are checked, the risk tier classified, and the places personal data is processed identified. Tasks involving personal data are flagged and reviewed. Deliverables are checklists written per stakeholder role plus onboarding material, so people outside IT can act on them.