Project State Model

Assumptions

Everything the model believes, stated before the math. If you do not buy one of these, do not buy the output.

1. A project's lifecycle is a finite ladder - announced, sited, permitted, interconnected, under construction, energized, operating - with cancelled and failed as absorbing states.
2. Each rung has an evidence class that binds it (the milestone register): a state is assigned only when evidence of the binding class exists on the record.
3. Evidence arriving after the as-of date does not exist for the run; the ladder is reconstructed from what was on record that day.
4. A state resting on one secondary claim is a weak state, and the probability says so - it is capped below 0.5 until a primary (tier-1) document binds it.

Derivation

The model is a finite-state reconstruction, not a forecast. It walks the ladder from the top rung down and stops at the most advanced state whose binding evidence exists:

where L is the ordered ladder and B_i(p) is the set of binding evidence items for rung i on project p, filtered to items dated on or before the as-of date. If no rung has binding evidence, the state is UNKNOWN - never assumed.

State probability reflects how the binding evidence splits between primary and secondary sources under the Evidence Weighting Model. With share the primary-side weight fraction:

The next-milestone probability is the state probability less a fixed 0.15 discount, floored at 0.05 - a stated policy, not a fitted transition matrix. There is no smoothing and no inference past the evidence.

Identification strategy

Why the estimate supports the claimed meaning - and where the model refuses to claim more than the data carries.

• The output is a claim about what the record supports, not about what is true in the world. The identification discipline is the binding rule: a rung is real only when its evidence class is present, so the model can never read a press release as a completed milestone.
• The primary/secondary split in the probability is what separates a documented state from a claimed one. A state that one secondary source asserts sits below 0.5 by construction, which is exactly the meaning the number carries: more likely asserted than proven.
• Absorbing states (cancelled, failed) short-circuit the ladder with probability 1 - a resolution on the record is terminal evidence, not a transition to be estimated.

Inputs and outputs

Calibration and gate evidence

Validation design

• Historical replay against resolved projects in the outcomes collection (spec Section 40): at each past date, reconstruct the state using only the evidence available that day, then check what actually happened next.

Evidence on record

• Replay results are not yet published in the sealed source. The validation design is specified; the ledger it runs against accrues as projects resolve.

Failure modes

The ways this model declines, degrades or goes UNKNOWN - and what it prints when it does.

• No binding evidence for any rung: state is UNKNOWN with the reason printed.
• All binding evidence stale: state is shown with the stale flag set - the state is not refreshed by assumption.
• Project not in the graph: every output is UNKNOWN, confidence LOW.
• A record carrying a status string but no provenance as-of date cannot bind a rung; it is dropped from the evidence set.

Sensitivity

• The answer moves only when the evidence set moves: a new primary document can advance the state several rungs in one run; nothing else touches it.
• The 0.15 next-milestone discount and the 0.5/0.49 probability split are fixed policy constants - changing them moves every probability mechanically, which is why they are versioned rather than tuned per case.