Vault Config & Intent Schemas

Config Fields — Clarification Session Questions

Fields below template_skip_patterns are asked during Step 6 (first session only) when empty. The question column shows what to ask the speaker.

Full Config Schema

See the canonical schema and field reference in ../../vault-profile/references/schemas-config.md. Current writes preserve config schema v2, including the owner-validated pptx_directory_exclusions; config v1 is owner-migration input, not writable session state. That file also documents the migration path for the exclusion field and for vaults created before the unified shownotes block.

Confirmed Intents Schema

Stored in the confirmed_intents array of the tracking database. Populated during clarification sessions when the speaker confirms a pattern is intentional. Persist records only with the ingress owner's upsert_confirmed_intent mutation; see the owner mutation contract.

{
  "confirmed_intents": [{
    "schema_version": 1,
    "pattern": "delayed_self_introduction",
    "intent": "deliberate",
    "rule": "Use two-phase intro: brief bio at slide 3, full re-intro mid-talk",
    "note": "Speaker confirmed this is an intentional rhetorical device",
    "confirmed_date": "2026-08-01",
    "source_talk": "2026-08-01-example.md"
  }]
}

The complete record has required exact integer schema_version: 1 and string fields pattern, intent, rule, and note, plus optional speaker-confirmation provenance. Provenance may use one of singular source_talk or legacy alias talk, or the non-empty unique string array source_talks; confirmed_date is canonical YYYY-MM-DD, and retrofit_targets is an optional non-empty unique string array. Existing speaker-owned intent labels outside the three recommended labels remain valid non-empty classifications and must not be discarded during an exact-record update. Unknown fields are rejected.

Improvement Goals Schema

Stored in the improvement_goals array of the tracking database. This is the artifact that closes the coaching loop: the speaker picks 1–2 focus areas, and a later ingress run checks whether the targeted issue actually moved. Without it the system diagnoses but never verifies that the speaker acted.

Artifact owner: vault-ingress owns the tracking database, record shapes, and migrations. Authorized writer: vault-clarification creates and retires goals during a session through upsert_improvement_goal. Reader/updater: vault-ingress reads active goals and changes only the verification fields through patch_improvement_goal_verification (status, current_value, last_checked, checked_by, verification_state, verification_reasons) — never the owner shape or fixed baseline. Both operations use the owner mutation contract. On a record whose schema_version it does not recognize, vault-ingress treats it as read-only and skips verification.

Schema v2 binds every catalog-derived goal to the exact pattern-scoring generation accepted by the speaker. The full baseline snapshot is copied from a validated schema-v4 or schema-v5 speaker-profile.json; Section 15 is narrative context, never the machine source. Schema-v5 derived labels may inform the offered goals only when their exact classification domain is available; schema v4 remains occurrence-only. The helper at scripts/goal_generation_provenance.py makes the mechanical comparability decision for both the owner and reader.

{
  "improvement_goals": [{
    "id": "reduce-shortchanged",
    "schema_version": 2,
    "issue": "Shortchanged — rushing the final third under time pressure",
    "kind": "antipattern|underuse|pacing|other",
    "antipattern_id": "shortchanged (null unless kind=antipattern)",
    "metric": "fraction of recent talks exhibiting Shortchanged",
    "baseline_value": "4 of last 6 (0.67)",
    "target": "at or below 1 of 3 (0.33)",
    "set_date": "2026-06-11",
    "set_by": "vault-clarification",
    "status": "active|improving|achieved|stalled|regressed|retired",
    "current_value": "",
    "last_checked": null,
    "checked_by": null,
    "verification_state": "pending|current|needs_rebaseline|unverifiable",
    "verification_reasons": [],
    "supersedes_goal_id": null,
    "baseline_provenance": {
      "lane": "pattern_scoring",
      "pattern_baseline": {
        "schema_version": 2,
        "as_of": "2026-06-11T12:00:00+00:00",
        "scope": "global",
        "active_batch_excluded": false,
        "excluded_filenames": [],
        "eligible_statuses": ["processed", "processed_partial"],
        "pattern_scoring_generation_status": "current",
        "pattern_scoring_generation_reasons": [],
        "pattern_catalog_fingerprint": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa",
        "pattern_scoring_schema_version": 5,
        "scored_talk_count": 12,
        "pattern_score_sum": 72,
        "average_pattern_score": 6.0,
        "eligible_talk_count": 12,
        "opportunity_coverage_identity": "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb",
        "raw_score_comparison_status": "available",
        "raw_score_comparison_reason": null
      }
    }
  }]
}

• kind ties the goal to a coaching surface: antipattern (a speaker-selected antipattern occurrence to reduce), underuse (a speaker-selected pattern occurrence to increase), pacing (hit the time/slide budget), other (free-form). Raw occurrence rows do not classify recurrence, signature status, or underuse.
• antipattern_id is set only when kind is antipattern; it is null for underuse, pacing, and other goals (no antipattern exists to reference).
• baseline_value is the human-readable metric value accepted by the speaker. For a pattern goal, compute it from the same validated profile cohort carried in baseline_provenance.pattern_baseline; never parse it from Section 15.
• target is the speaker's own stated aim, not a generic standard.
• baseline_provenance.lane is pattern_scoring for antipattern and underuse, pacing for pacing, and independent for other. Pacing/independent records omit pattern_baseline. An other goal cannot be used to evade pattern provenance.
• Pattern baselines must be non-empty post-batch full-cohort snapshots (active_batch_excluded: false, excluded_filenames: []). A fingerprint or scoring schema mismatch produces needs_rebaseline; scoring-v5 also requires one available opportunity_coverage_identity for raw-score comparability. A missing current baseline or legacy schema-v1 pattern goal is unverifiable. Neither state permits an outcome status.
• Rebaselining is an explicit speaker decision: preserve the old record, retire it, and create a new schema-v2 record with supersedes_goal_id. Never overwrite or restamp the old baseline.
• Existing schema-v1 antipattern/underuse goals remain read-only and unverifiable. Existing schema-v1 pacing and truly non-pattern other goals remain independent of catalog releases.
• Verification rubric (how vault-ingress sets status) lives in ../../vault-ingress/references/processing-rules.md Improvement Goal Verification.