Demonstration — one item, end to end

ILLUSTRATIVE. NOT CUSTOMER EVIDENCE.

Every figure below is a labelled assumption constructed to show the shape of the method. No customer supplied these numbers and no engagement produced them.

This file may be shown to a prospect only with this banner intact. Using an illustrative figure as a reference result would destroy the one thing the offer sells — that our numbers carry their provenance. If a sales conversation needs a real figure, the answer is "we don't have one yet for your sector", not this page with the banner removed.

The organisation is anonymised and composite: a mid-size membership body running a content platform, with an in-house approval integration built years ago.

---

DISCOVERY

Field	Value
item_id	ITEM-017
item	Custom content approval integration
business_purpose	Route member-facing content to a second approver before publication, and keep an audit record of who approved what
current_owner	Unknown — the digital team assumed platform engineering owned it; platform engineering assumed digital did
who_uses_it	Content editors (11), compliance reviewer (1)
why_it_was_built	Built in 2019 because the platform's workflow module could not express a two-stage approval with an audit trail
current_maintenance	Breaks on most minor platform upgrades; three incidents in the last year; one developer spends roughly a week a year on it
credentials_or_access	One service account with content-publish rights; one outbound webhook token to the notification service
key_person_dependency	SINGLE_PERSON — one contractor, no longer on retainer

current_owner being unknown is itself a finding and was reported before any verdict was reached.

CURRENT COST

measure	Value	source	assumption	customer_confirmed
Maintenance hours	40 h/yr	illustrative	One developer-week, from the team's own recollection	NO
Upgrade retest	6 items	illustrative	Counted from the last upgrade's test plan	NO
Access points	2	illustrative	One service account, one webhook token	NO

All four rows read customer_confirmed = NO, which is exactly how they would appear in a real report until the customer confirmed them. An unconfirmed number is shown as unconfirmed, not dropped and not promoted.

RISK

• upgrade_impact — blocks the major upgrade; its approval hooks use an API removed in the next version
• security_impact — the service account holds publish rights far broader than the integration needs, and its dependency set has not been updated in two years
• key_person_dependency — SINGLE_PERSON, and that person has left

REPLACEMENT

Field	Value
existing_platform_replacement	The platform's own editorial workflow with a second review state, already licensed and already enabled for another content type
replacement_maturity	GA
replacement_evidence	VENDOR_DOCUMENTED at review time → SIDE_BY_SIDE_PROVEN after step 4
removal_difficulty	MEDIUM
removal_risk	MEDIUM — publication is member-facing
verdict	REPLACE
rank	NEXT at review time
verdict_rationale	Platform workflow covers both approval stages and the audit record; evidence was VENDOR_DOCUMENTED only, so this was ranked NEXT rather than NOW pending a side-by-side run

Note the rank. At Review time the evidence was vendor documentation, which under Delivery-Method.md §1 phase 3 may not be ranked NOW. It was promoted only after step 4 produced SIDE_BY_SIDE_PROVEN.

SIDE-BY-SIDE PROOF

Both ran for three weeks against real editorial traffic.

Field	Value
before	3 weeks baseline: 212 items published, 212 with a complete two-approver audit record
after	Same 3 weeks, parallel run: 212 items routed through platform workflow, 212 with a complete audit record
business_outcome	No member-facing content is published without a second approver, and we can show who approved it
functional_equivalence	EQUIVALENT_WITH_DIFFERENCES
known_differences	Platform workflow records the approval timestamp in UTC; the custom integration recorded local time. Compliance reviewer confirmed UTC is acceptable and preferable. Email notification wording differs; editors were notified.

EQUIVALENT_WITH_DIFFERENCES with two named differences is the realistic outcome. A claim of EQUIVALENT with known_differences: none across a three-week parallel run would be the suspicious result, not the good one.

CUSTOMER APPROVAL

Field	Value
customer_approval	GRANTED
customer_approver	Director of Digital (named individual in a real engagement)
what_replaces_it	Platform editorial workflow, second review state
rollback_path	Custom integration left installed and disabled by configuration flag; configuration export retained in the customer's repository at a tagged commit
rollback_period_days	45

45, not 30. The customer's compliance review runs on a six-week cycle and the window was extended to span one full cycle. The default is 30; the engagement sets it.

CUTOVER AND ROLLBACK WINDOW

Step 7 disabled the custom integration by configuration. It was not deleted. Step 8 tested the rollback: the flag was flipped back in a staging environment restored from the production configuration, and the custom path resumed — so rollback_test = PASS rather than an assumption.

Step 9 read back from production: the next 30 published items each carried a complete two-approver record.

CREDENTIAL RETIREMENT

Step 10, and the step most likely to be skipped because the feature already worked at step 9:

• Service account with publish rights — revoked
• Outbound webhook token — revoked
• old_runtime_disabled — YES
• old_source_retired — YES, after the 45-day window
• documentation_updated — YES; the runbook's approval section now points at the platform workflow

FINAL RESULT

Field	Value
final_verdict	PASS
functional_equivalence	EQUIVALENT_WITH_DIFFERENCES (both accepted)
rollback_test	PASS
production_readback	PASS

ANNUAL BURDEN REMOVED

measure	Value	customer_confirmed
CUSTOM_SYSTEMS_RETIRED	1	YES
MAINTENANCE_HOURS_RETURNED	40 h/yr	NO — customer estimate, not confirmed against records
ACCESS_POINTS_CLOSED	2	YES
UPGRADE_TESTING_REMOVED	6 items	YES
KEY_PERSON_DEPENDENCIES_REMOVED	1	YES
ANNUAL_OWNERSHIP_COST_REMOVED	not calculated	—

Two things to notice, because they are the method working rather than the method looking good:

1. Four of six measures are confirmed; one is not, and says so. The hours figure is the one a less careful report would lead with.
2. ANNUAL_OWNERSHIP_COST_REMOVED is not calculated at all, because the customer never supplied a loaded cost rate. The honest output is an absent number, not a plausible one derived from an industry average.

What was proven: the same required business outcome — no member-facing content published without a second approver, with an audit record — now runs on a capability the customer already licensed, with one fewer system, two fewer credentials, six fewer upgrade test items and no single-person dependency.

Same required outcome. Less customer-owned machinery.