Files
HealthLog/src/lib/insights/chart-tokens.ts
T
Marc-André BombeckandClaude Opus 4.7 dc01ab7934 feat(insights): metric:MOOD inline charts + evidence-contract prompt
Bucket D phase 1 of the v1.4.3 polish pass — the AI insights felt
"static" because (a) findings repeated phrasing with no concrete
numbers and (b) the inline-chart machinery existed but was prompted
permissively, so the model rarely used it.

Two concrete moves, scoped to ship in this hotfix:

1. **`metric:MOOD` is now an allowlisted chart token.** Renderer in
   `<InsightAdvisorCard>` branches on `tokenKind(token)` — measurement
   tokens still mount `<HealthChart>`, but `metric:MOOD` mounts the
   dedicated, self-fetching `<MoodChart>` instead. The mood timeline
   was previously the most under-served by the AI prose because it
   couldn't be charted inline; now the model can illustrate adherence-
   risk and mood-drift findings with the same chart users see on the
   /mood page. Drift test was updated in lockstep — the v1.4.2 guard
   that asserted MOOD was excluded now asserts it's present, plus a
   new guard asserts COMPLIANCE is still excluded (no self-fetching
   wrapper exists yet for `<ComplianceLineChart>`; ship that in v1.5).

2. **Prompt grew an EVIDENCE CONTRACT and a forbidden-phrase list.**
   The base system prompt (DE + EN both) now requires every
   `findings[]` entry to put a concrete number from the snapshot in
   `value` ("138/85 mmHg", "+0.4 mmol/L vs. 30d-avg") instead of
   adjective prose ("slightly elevated"). Findings without a snapshot
   anchor MUST be omitted — variable cardinality 0..8 explicitly
   stated, sorted by salience, NO padding. The forbidden-phrase list
   blocks generic boilerplate that signals ungrounded output: "achte
   auf ausreichend Schlaf", "drink enough water", "regular exercise",
   "consult your doctor" (except in the disclaimer field which exists
   precisely for that line). The chart-token instruction flipped from
   passive ("Du DARFST") to active ("Bevorzuge Tokens immer dann,
   wenn …") with three explicit triggers — temporal claims, threshold
   references, window comparisons — that reliably surface chart-worthy
   findings.

Carved out of v1.4.3 for v1.5: telescoping multi-resolution payload
(14d daily + 10w weekly + 21mo monthly + yearly summary per QA
research 1) and the evidence-ref output schema extension (QA 2).
Both touch the prompt's input shape and would change every cache
key + token bill — they want their own phase.

Tests + typecheck green (671/671 — one new chart-tokens drift
guard).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-08 21:41:29 +02:00

81 lines
3.1 KiB
TypeScript

/**
* Chart-include tokens for AI insight prose.
*
* The LLM can embed one of the literal strings below inside `summary` or
* `findings[].label` to inline-render the matching chart underneath the
* paragraph. Every token the model emits is matched against this allowlist
* — anything else (typos, hallucinations, attempts to inject other strings)
* is silently dropped before it ever reaches the renderer.
*
* Render path: prose → parseChartTokens(prose) → matching chart component
* per token. The visible prose is run through stripChartTokens so the
* literal token string never surfaces in the UI.
*/
// v1.4.3: allowlist extended with `metric:MOOD` so the AI can illustrate
// findings about mood drift inline. <MoodChart> is self-fetching, so the
// renderer just mounts it.
//
// `metric:COMPLIANCE` was prepared but pulled before ship — the existing
// <ComplianceLineChart> takes pre-aggregated daily data via props and
// has no self-fetching wrapper, so wiring it inline would silently
// render an empty chart. Land it in v1.5 once a self-fetching wrapper
// exists.
export const ALLOWED_CHART_TOKENS = [
"metric:WEIGHT",
"metric:BLOOD_PRESSURE_SYS",
"metric:BLOOD_PRESSURE_DIA",
"metric:PULSE",
"metric:BODY_FAT",
"metric:SLEEP_DURATION",
"metric:ACTIVITY_STEPS",
"metric:BLOOD_GLUCOSE",
"metric:TOTAL_BODY_WATER",
"metric:BONE_MASS",
"metric:OXYGEN_SATURATION",
"metric:MOOD",
] as const;
export type ChartToken = (typeof ALLOWED_CHART_TOKENS)[number];
const ALLOWED_SET = new Set<string>(ALLOWED_CHART_TOKENS);
// Greedy `[A-Z_]+` match. Uppercase letters and underscores only — anything
// else (apostrophes, spaces, lowercase, digits) terminates the token, so
// trailing junk like `metric:WEIGHT' onclick='alert(1)'` cleaves cleanly
// into the safe metric and an inert text remainder.
const TOKEN_REGEX = /metric:[A-Z_]+/g;
/**
* Find every well-formed chart token in `text` and return only those that
* are in the allowlist. Hallucinated tokens (e.g. `metric:NUKE`) are
* silently dropped.
*/
export function parseChartTokens(text: string): ChartToken[] {
const matches: string[] = text.match(TOKEN_REGEX) ?? [];
return matches.filter((m): m is ChartToken => ALLOWED_SET.has(m));
}
/** Strip the chart tokens out of the visible insight text. */
export function stripChartTokens(text: string): string {
return text
.replace(TOKEN_REGEX, "")
.replace(/\s{2,}/g, " ")
.trim();
}
/** Map a chart token to the metric param the underlying chart component
* expects. For MeasurementType-backed tokens this is a real enum value
* (`WEIGHT`, `PULSE`, …); for `MOOD` and `COMPLIANCE` it's a synthetic
* identifier the renderer routes to a dedicated chart component. */
export function tokenToMetric(token: ChartToken): string {
return token.slice("metric:".length);
}
/** Token kinds the renderer needs to distinguish. MeasurementType tokens
* feed `<HealthChart>`; `metric:MOOD` mounts the dedicated, self-fetching
* `<MoodChart>` instead. */
export function tokenKind(token: ChartToken): "measurement" | "mood" {
if (token === "metric:MOOD") return "mood";
return "measurement";
}