Intent Hierarchy Evidence

Answer#

This record distils the Intent layer hierarchy investigation of 2026-10-30 into the claims this release relies on. Counts below were measured over the largest consumer's built model on 2026-10-30, holding 21,548 records.

Only four Intent links are required by schema, stated as obligations, and enforced by a check at the same time: product.audience, journey.actor, a requirement except entry to its fallback, and decision.decides on product decisions, the last by provision plus check despite an optional schema. Every other link above the requirements is optional in the schema, mentioned only in provision Rationale, and unchecked. One layer down LiteSpec already accepts the every-record-points-up principle: a use case with neither serves nor fulfills fails record-link-missing at tier G. The principle was simply never applied inside Intent. Confidence is high: each cell was read from the schema, the provision text, and the check table.

The central measured gap is requirements with no origin: 719 of 928 name neither feature (56 do) nor basedOn (164 do, 11 carry both). Absence of feature alone is not a defect, because the field is defined narrowly as the feature that carries the requirement in a named phase, so a built requirement correctly carries none. The combined absence is still not a conformance defect today, because the provision puts the expectation in its Rationale while its Statement and check require only identifier, kind, statement, and dated applied values. The 719 are therefore specification-tolerated traceability defects, chargeable to LiteSpec's missing obligation and missing check rather than to the consumer's authors. One measured signal shows the anchor is already load-bearing in practice: of 150 requirements fulfilled by nothing below, the 100 without a feature all appear as known requirement-fulfillment-missing findings, while none of the 50 with a feature produced such a finding.

Features split cleanly. All 40 features carry solves, and all 8 goals are served by 16 serves links from 12 features, so the top of the chain is intact; the gap is that 28 of 40 features name no goal. Journeys name their goal as prose in 112 of 112 cases, by design: the schema requires a string, the provision requires the goal in words, and the Links table holds no journey to goal edge. A migration warning is measured: all 112 goal sentences are distinct and none contains any goal record's title, so no mechanical mapping to the eight goals exists.

The remaining populations are tolerated or by design. Five of 29 constraints and 15 of 29 research records are cited by no basedOn; no rule requires citation, so they form a basis pool, though 15 of 29 suggests the pool is written faster than decisions consume it. All 3 product decisions decide (2, 65, and 4 targets); the 71 technical decisions carry no decides by design and govern code instead, where 8 already draw the checked decision-unimplemented finding and 15 draw advisory decision-basis-missing. One of 30 problems is solved by no feature, deliberately left without one. Exactly one actor, actor/report-recipient, is named by nothing at all, a genuine but minor orphan. Fourteen of 164 intent-layer terms are cited by no record, which is tolerated. Requirement-page container conformance could not be judged from the model, because the builder emits zero requirement-page records; it needs the consumer validation run, not this record.

The investigation tested the recommendation clause by clause. The requirement origin clause is supported with one amendment: keep feature and basedOn distinct in the wording, report the anchor found as one finding, and count a decision-settled requirement as anchored. The feature clause is supported for solves, since all 40 features already comply, but serves should stay advisory: with only eight goals, forcing one goal per feature manufactures vague claims, so the expectation should attach where it informs sequencing, on dated features. The journey goal clause is argued against as a replacement and supported as an added optional link, because the sentence in words is what lets a reader judge a journey chapter without opening a goal page. Advice-first checks are supported with a written promotion trigger: zero advisory findings on merged records, checks landing one slice at a time, and promotion through the waiver mechanism with an end date.