The `residual` field carries every honesty disclosure in this repo and is not in any schema

[avrabe]

Found while adding a residual to FEAT-033 (#196). rivet validate --explain says:

INFO: [FEAT-033] field 'residual' is not defined in schema for type 'feature'
  help: ... remove the 'residual' field, or rename it to a declared field
        [acceptance-criteria, baseline, phase].

Measured: 17 features carry a fields.residual, and all 17 emit that INFO — 17 of the 31 INFO diagnostics rivet validate produces.

Why this matters more than the noise

residual is not a stray field. It is where this repo puts every honesty disclosure: what a feature does not do, what was measured and found wanting, what an assessor should not conclude from a green. FEAT-091's trailer limits, FEAT-093's ruleset-reset gap, FEAT-096's undetected asymmetric case, FEAT-097's four vacuous 100%s — all of it lives in a field the schema does not know exists.

Consequences today:

• Nothing can require it. An accepted feature with no residual validates identically to one with a thorough residual. FEAT-033 shipped with none for months, and that is precisely the defect FEAT-097: four coverage rules report 100% over an empty population #196 is fixing.
• Nothing validates it. A typo (residuals:, residual_notes:) is silently a different field — indistinguishable from absence.
• The correct-by-house-pattern artifact is permanently nagged as possibly-a-typo, which trains the reader to skip INFOs.

What needs deciding

1. Can a local schema extension add a field to a built-in type? schemas/research-ext.yaml adds link types and explicitly notes it does so "without redeclaring any existing artifact type" — link types are documented as unioned by name. Whether fields merge the same way, or whether declaring feature locally replaces the built-in type, is unknown and needs testing before anything is written.
2. If it can, should residual be required for status: accepted features? That is the interesting question and the reason this is not just noise cleanup. A required residual is a mechanical honesty gate — but it is also trivially satisfiable with "none", so it would need FEAT-088's rule 2 shape (declared and meaningful) to be worth anything, and that may not be expressible in a schema.

Deliberately not acted on in #196 — the answer to (1) is unknown and (2) is a design decision, so bundling either into a PR about ASPICE scope would have been scope creep with an uncertain outcome.

Refs: FEAT-097

[avrabe]

Both questions answered; the fix landed in #200 (8a7de29).

1. Can a local schema extension add a field to a built-in type?

Yes, and fields MERGE rather than replace — which was the risk that made this worth probing rather than just trying. Replacing would have dropped acceptance-criteria from every feature in the repo.

Verified on an isolated project before touching this one, then confirmed on main:

$ rivet schema show feature
Fields:
  phase                    string       optional  [phase-1, phase-2, phase-3, future]
  acceptance-criteria      list<string> optional
  baseline                 string       optional
  residual                 string       optional      <-- added by schemas/dev-ext.yaml

The key is artifact-types: (not types:), and the entry needs a description. Controlled in both directions: disabling the schema brings the INFO back, enabling it removes it.

On this repo: INFO 32 → 14, WARN unchanged at 188 — it removes exactly these diagnostics and nothing else.

2. Should it be required for status: accepted?

No, and this is the more interesting half. A required field is satisfiable with the word "none", so it would gate on the presence of a string rather than on disclosure — a checkbox that reads as rigour and enforces nothing, which is this repo's documented dominant failure class. FEAT-088's rule 2 shape (declared and justified) is what would be worth having, and it is not expressible as a field constraint. Recorded in the schema file header so the next person doesn't re-litigate it.

What actually improved

Not the suppressed diagnostics — the discrimination. Measured on a three-artifact probe carrying residual, residuals, and residual_notes:

	reported
before	all three, identically
after	only the two typos

A misspelled disclosure was silently a different field — silently absent — which is precisely the failure residual exists to prevent.

One thing this turned up

rivet schema validate reports two errors on the built-in safety-case preset, pre-existing and unrelated to this change. They are on goal-has-support and goal-has-context — the two rules at 40% — so I chased them, because a broken rule would have meant the 40% wasn't real.

It is real. The coverage engine resolves supported-by/scoped-by correctly and G-001/G-002 are covered via genuine supports/scopes links; only the validator fails to resolve inverse names. J-004's investigation stands. Filed upstream as pulseengine/rivet#876.

Closing.