Project Lantern: following a decision through funding, evidence and reconsideration

Project Lantern is a fictional portfolio exercise designed to make an investment decision traceable from its original request through funding conditions, returned evidence and the next decision.

The expanded scripted MVP was accepted by Rick Molony on 27 October 2026 with documented limits. Its achievement is a completed, inspectable demonstration. It does not establish that independent participants can operate it effectively or that a real portfolio would achieve better financial results.

Start with the whole portfolio

The exercise retains all eighteen original records: six exits, four delegated items and eight mature proposals. These records have different decision needs. The demonstration does not manufacture a major portfolio meeting for every item.

The synthetic baseline contains requests for 62 million of capital funding and 19.3 million of possible annual operating expenditure, against a 30 million capital envelope. Unknown actual spending remains unknown. These are illustrative scenario values, not real budgets.

Follow one consequential cycle

An initial funding round allocates 18.5 million and leaves 11.5 million in reserve. Late evidence changes a regional proposal from 4 million to 6 million. The revised allocation becomes 20.5 million and the reserve falls to 9.5 million.

Both rounds remain visible. The later decision does not erase the earlier one or imply that a recommendation was always a commitment.

The cycle then separates authorisation from release. Conditions and review capacity can hold work even after a funding decision. Returned evidence matters too: a finance example produces a 10% result against a 15% condition, so follow-on work is held and the next decision is Revise. A completed process does not mean a successful or funded bet.

How AIPF shaped the work

The five portfolio views draw on the same source records: bets, decision capacity, shared dependencies, benefits and allowed AI actions. They do not create five competing versions of the facts.

What the recorded checks establish

The acceptance record reports eighteen regression tests and ten fixture checks. They cover funding totals, unmet conditions, excessive simultaneous starts, incomplete or stale input, recovery to the complete fixture and return comparisons. A parent-link warning during import was repaired and checked.

These checks establish the behaviour of the scripted demonstration within its boundary. They do not establish independent human usability, least-privilege enforcement or recovery from every possible server failure. The repeat snapshot uses the same synthetic baseline; it is not fresh real-world business evidence.

Lessons for a portfolio owner

Preserve the chain from request to decision, terms, release and return. A changed result should be able to change the next action. Hold incomplete evidence visibly rather than filling gaps to make the meeting pack look finished.

Treat scarce competent review time as a constraint. Preserve disagreements and unknowns. Ask a person who did not assemble the records to reconstruct one decision without coaching before claiming the process transfers well.

Using this case with the guides

Read the Portfolio Guide for decision scope, then the Portfolio Implementation Guide for the connected operating cycle. HarbourBridge provides a complementary comparison-and-capacity example.

The article is self-contained; access to private collaboration tools is not required. Independent participant observations and measured efficiency remain future evidence, not implied by acceptance of the scripted MVP.