Example: fragmented estate¶

Synthetic example — constructed to illustrate the report shape.

These numbers are invented to show what the report looks like for an estate with no shared harness and several diverged copies of the same control surfaces. They are not a reading of any real estate. The self-assessment is the real one.

Five service repositories, each with its own habitat, plus one legacy repository nobody is investing in. No shared harness is declared — because none exists.

The manifest¶

version: 1
team: commerce-platform

subjects:
  - id: checkout
    path: ./checkout
  - id: search
    path: ./search
  - id: inventory
    path: ./inventory
  - id: pricing
    path: ./pricing
  - id: fulfilment
    path: ./fulfilment
  - id: legacy-batch
    path: ./legacy-batch
    posture: maintenance
    justified_variance:
      - dimension: testing
        reason: COBOL batch; harness test tooling does not apply

Coverage ledger¶

Subject	Status	Report date	Age	Note
checkout	assessed	2026-08-05	9d	—
search	assessed	2026-08-05	9d	—
inventory	assessed	2026-08-06	8d	—
pricing	assessed	2026-08-07	7d	—
fulfilment	assessed	2026-08-07	7d	—
legacy-batch	assessed	2026-08-08	6d	maintenance

6 of 6 subjects assessed; 0 stale; 0 unreachable.

Dimension matrix¶

All placements are local — no habitat is declared, so nothing is inherited. * marks the weakest cell in the column.

Subject	AB	AI	WF	OM	TP	OR	OA	HR	WP	AC	AD	Te	Ob	Gv
checkout	3	3	3	3	3	3	3	3	3	2	3	3	2	3
search	3	2	3	3	3	3	3	3	3	2	3	3	2	3
inventory	3	3	3	3	3	3	3	3	2	2	3	3	2	3
pricing	2	2	3	2	3	2	3	2	2	2	2	2	2	2
fulfilment	3	3	4	3	3	3	3	3	3	3	3	4	3	3
legacy-batch (maintenance)	2*	1*	1*	2*	1*	2*	2*	2*	2*	1*	2*	1*	1*	1*

AB Agent behaviour · AI Agent input · WF Workflow · OM Operating model · TP Teams provide · OR Output role · OA Output artefact · HR Humans review · WP Work patterns · AC Agent composition · AD Agents… · Te Testing · Ob Observability · Gv Governance

Gap table¶

Subject	Habitat maturity mean	Cognitive level	Gap	Regime
fulfilment	3.14	3	−0.14	Coherent
checkout	2.86	3	+0.14	Coherent
search	2.79	3	+0.21	Coherent
inventory	2.79	3	+0.21	Coherent
pricing	2.21	3	+0.79	Ambition outpaces enablement
legacy-batch	1.50	3	+1.50	Ambition outpaces enablement

Spread¶

Across active subjects: −0.14 (fulfilment) to +0.79 (pricing) — range 0.93. Including legacy-batch, the range is 1.64; it is reported but marked maintenance, so it does not set the agenda.

Four of the five active services cluster tightly. pricing is the outlier, and fulfilment is the only subject whose habitat is slightly ahead of the team's cognitive read.

Portfolio regime¶

Fragmented — with lineage.

The evidence:

No habitats declared, and no inherited or inherited-unbound provenance anywhere. All 84 placements are local. That rules out Federated and Distributed.
Four services carry a HARNESS.md descended from a common ancestor — checkout, search, inventory and pricing. Same section structure, same rule names, same opening paragraph; diverged bodies. fulfilment wrote its own, which shares nothing.
Moderate spread (0.93 across active subjects), which separates this from Islanded — the estate is not one team pulling ahead of three, it is four copies of the same thing drifting.

Where the copies diverged¶

Artefact	Subjects	Divergence
`HARNESS.md`	checkout, search, inventory, pricing	Constraint list identical in checkout/search; inventory added two agent-composition rules; pricing is missing the CI-gate section entirely and still carries the original's TODO markers
`.github/workflows/ci.yml`	checkout, search, inventory	Same job structure; three different Python versions and two different coverage thresholds
`CLAUDE.md`	all five services	Same preamble; five different sets of project-specific conventions — genuinely divergent, not drift

pricing is the clearest lineage case: it is an older copy that never received the CI-gate section the other three have, which is most of why its gap is the widest among active subjects.

Declared variance¶

Subject	Dimension	Reason
legacy-batch	Testing	COBOL batch; harness test tooling does not apply

Listed, not hidden. legacy-batch's L1 Testing is not reported as drift and not treated as an extraction candidate.

Split ceiling¶

Common weak — Agent composition, Observability. Weak in four of the five active subjects. Neither is present in any of the copied harnesses, so every team has been left to invent them and four have not. This is the enablement backlog.

Locally weak — pricing: Agent behaviour, Operating model, Output role, Humans review, Agents…, Testing, Governance. pricing is behind across the board, and the lineage analysis says why: it is running an older copy of the shared ancestor.

legacy-batch is excluded from common-weak detection by its maintenance posture. Without that exclusion it would set the ceiling on eleven of fourteen dimensions and every one of those findings would be noise.

Confidence¶

84 placements: 48 observed, 36 inferred, 0 asked. No subject exceeds five inferred dimensions.

What this estate should do next¶

Extract the common layer — and treat the divergences as the work. Four services are maintaining four copies of one harness. The extraction is not the hard part; reconciling what the copies disagree about is. Three Python versions and two coverage thresholds encode four local decisions, and some of those are real requirements while others are accidents nobody remembers making. Go through them deliberately rather than picking the newest copy and declaring it canonical.

Start with pricing. It is the widest active gap and the cheapest to close, because it is not missing capability — it is missing the CI-gate section the other three copies already have.

Put agent composition and observability into the extracted layer, not into each service. They are weak in four of five subjects precisely because no copy of the ancestor ever had them.

Leave fulfilment's harness alone for now. It shares no lineage with the other four, it is the only subject whose habitat is ahead of the team, and folding it into the extraction would mean reconciling a fifth unrelated design for no benefit. Read it after the extraction lands, and take what is good from it then.

This estate is Fragmented, not Islanded — the distinction matters. If those five habitats shared nothing, extraction would be the wrong move and the answer would be to move people instead. See Extract or bind.

View this report as a shareable HTML page — a single self-contained file, no network assets, works opened straight from disk.