Imported from zakariaf/Hifz-Companion (.claude/skills/ui-science-source-row/SKILL.md). Install upstream with npx skills add zakariaf/Hifz-Companion --skill ui-science-source-row. Copyright stays with the author.

ui-science-source-row

The source row is the atomic row of the "The science we follow" screen: it renders exactly one verified CLAIMS.md row — a plain-language claim headline, its named-and-dated source (author/year/venue, or a hadith collection + number), an evidence-grade tag rendered as honest confidence-about-the-evidence language, and an optional tappable link that clearly leaves the app. It is a calm ListTile/Card-shaped leaf; it never authors a fact (the register is the only author), never shows a grade as a star rating or "proven" badge, and never depends on the network to be trustworthy.

This row is where Pillar 5 (private & offline by feel) and the science foundation's honesty chain become visible to a ḥāfiẓ: the full citation is on-device text so the row reads identically in airplane mode, the grade describes the strength of the evidence and is kept strictly separate from any certainty about your Quran, and a [TRAD] source is named with its collection/number and framed as methodology — never a ruling. One row = one verified, sourced, graded claim.

When to use

Use this skill when building or placing:

• a science-screen source row — the claim headline + named source + grade tag + optional "the evidence" expansion (docs/science/11-the-in-app-science-screen.md §3, §4);
• the evidence-grade tag — the [MA]/[RCT]/[EXP]/[CS]/[OBS]/[TEXT]/[TRAD] label rendered as plain confidence words + a non-color glyph (docs/science/11-the-in-app-science-screen.md §5, §7);
• the on-device citation block — author, year, title, venue as readable text, with an optional tappable URL that visibly leaves the app and degrades gracefully offline (docs/science/11-the-in-app-science-screen.md §2, §4; docs/design-system/10-privacy-and-trust-ux.md §3, §7);
• a [TRAD] source row — a hadith shown with collection + number + grading, framed as methodology, issuing no ruling (docs/science/11-the-in-app-science-screen.md §4, §8);
• the "why does the app say this?" affordance — a per-claim or per-number tappable that opens the row's source.

Do NOT use this skill for → use the sibling instead:

• authoring/grading the claim itself, the seven-column CLAIMS.md row, the REFERENCES.md entry, or the whole science screen's structure and offline-payload contract → use domain-claims-register-and-science-screen (this skill renders one row the register already owns; it never invents a claim).
• the grade → lay confidence-phrase mapping and the standalone certainty badge / grade legend → use ui-certainty-label (this row composes that badge in its "the evidence" expansion; it does not re-author the phrase mapping).
• the engine math / rule a claim describes (due = min(ideal, ceiling), FSRS D/S/R, the cycle ceiling, "never safe to drop") → use domain-scheduling-engine-rules; the row reads the registered description, it never re-derives the rule.
• the calm, non-coercive, sect-neutral, servant-to-teacher wording of the headline and the no-ruling framing of a [TRAD] row → use domain-adab-and-religious-integrity.
• per-locale numerals (year/number), FSI/PDI bidi isolation of a Latin author/venue/URL inside RTL text, and RTL mirroring of the row → use eng-rtl-and-bidi-layout; this skill only requires you use them.
• adding the localized strings for the headline / grade label / "leaves the app" hint to the ARB bundle → use eng-add-localized-string.
• the verified offline pack the bundled citation text ships inside → use domain-asset-pack-integrity; this row only relies on it.

The source row is attribution made visible — it shows who found a thing, when, and how strong the evidence is. A row that grades the user's Quran, sells, promises a retention number, fetches "latest research," or hides a citation behind a link that is meaningless offline is the wrong component.

The canonical pattern

1. The row renders a registered claim — it is a view, never an author. Each row reads its claim text, value/rule, source citation, and grade from the bundled CLAIMS.md data; nothing on the row is hand-written science. If a string here asserts a fact not registered with ≥1 verified source and a grade, that is a release-blocking defect, not a copy nit. docs/science/11-the-in-app-science-screen.md §1 (the screen "renders the register — it is a view, not a source of truth"); authoring lives in domain-claims-register-and-science-screen.

2. Two-layer copy: a plain headline, an optional "the evidence" expansion. The face of the row is a plain headline in the ḥāfiẓ's own frame (murājaʿa / pages / juz / cycle), type.body, common words, active voice; the named source, grade, and any caveat live in an optional expansion ("the evidence") so detail is on demand, never a wall of jargon. Technical terms (FSRS, retrievability) appear only with a one-line plain gloss. docs/science/11-the-in-app-science-screen.md §3 (plain, common words; two-layer copy; the user's frame); copy register owned by domain-adab-and-religious-integrity.

3. Name and date the source on the face of every claim. The row shows the author(s), year, venue (e.g. Cepeda et al., 2006 — meta-analysis), never a vague "studies show." The full reference is present as readable on-device text so it is useful with no connection. docs/science/11-the-in-app-science-screen.md §4 (named, dated source + grade tag on every claim); strings via eng-add-localized-string.

4. The grade is honest confidence language — never a star rating, never a promise. Compose the certainty badge (ui-certainty-label) in the "the evidence" expansion: it renders the grade as plain words about the evidence — "among the best-established findings in memory science" for [MA]; "a single controlled study" or "named traditional scholarship" for weaker grades — kept strictly separate from any certainty about the user's Quran. The grade is conveyed by its text tag ("meta-analysis") plus an optional non-color glyph, never by color alone and never as ★★★★★, a "proven" badge, or a retention percentage. docs/science/11-the-in-app-science-screen.md §5 (grades → confidence language; grade ≠ certainty) and §7 (grade never color-only, WCAG 2.2 SC 1.4.1).

5. A [TRAD] source is named by collection + number and framed as methodology — no ruling. A traditional row (e.g. the decay ḥadīth, Ṣaḥīḥ al-Bukhārī 5032) shows its collection, number, and grading, stays madhhab- and sect-neutral, positions the app as servant to the teacher and the sanad chain, and shows "needs scholarly review" plainly where sign-off is pending. No row interprets a verse or issues a fiqh verdict. docs/science/11-the-in-app-science-screen.md §4 (religious claims sourced by name/number) and §8 (methodology, not rulings); domain-adab-and-religious-integrity.

6. The link clearly leaves the app — and is meaningless-but-harmless offline. A tappable URL is an optional convenience, never the citation itself: it is visibly marked as an external link (a non-color "opens externally" glyph + a localized "opens in your browser" / "leaves the app" hint), and offline it simply does nothing — the on-device reference still carries full trust. The tap opens the system browser via the platform launcher; the app makes no in-app fetch of the source. docs/science/11-the-in-app-science-screen.md §2 (bundled, static, offline; source links degrade gracefully; no runtime fetch) and §4; external-link honesty in docs/design-system/10-privacy-and-trust-ux.md §3 (specific, verifiable; link is the check) and §7 (verification links are mixed Latin/RTL runs, bidi-isolated, ≥48dp).

7. No coercion, no engagement mechanics on the row. The row explains; it never sells, nags, frightens, or counts. No "5 of 12 sources read" progress, no badge for expanding the evidence, no urgency, no guilt, no decay-as-threat framing. A calm, declarative line does the work. docs/science/11-the-in-app-science-screen.md §6 (calm, no coercion; no engagement mechanics on the science); domain-adab-and-religious-integrity.

8. Accessible, RTL-native, locale-numeral by construction. The row meets WCAG 2.2 AA — ≥4.5:1 text contrast across light/sepia/dark, resizable text at 200% with reflow (no truncation, no horizontal scroll), and the grade never color-only. The year and any number render in locale numerals (Extended Arabic-Indic ۰۱۲۳ for fa/ckb, Arabic-Indic ٠١٢٣ for ar, via intl NumberFormat + type.numeral, never raw ASCII); the Latin author/venue/URL run is bidi-isolated (FSI/PDI) so it never reorders inside the RTL line; placement uses EdgeInsetsDirectional / logical start/end and the external-link glyph sits at the logical end. docs/science/11-the-in-app-science-screen.md §7 (WCAG 2.2; grade never color-only; RTL + numerals + bidi isolation); numeral/isolation mechanics via eng-rtl-and-bidi-layout.

9. Offline, no-AI, on-device by construction. Every word of the claim, citation, grade, and confidence label ships inside the app binary or the verified core pack and renders locally — no live feed, no model generating an explanation, no "latest research" fetch. Corrections ship as an app/pack update, never a silent remote edit. docs/science/11-the-in-app-science-screen.md §2 (bundled, static, offline, no AI; versioned with the app); relies on domain-asset-pack-integrity for the verified pack.

10. One screen-reader phrase, locale-tagged. Merge the row into one localized phrase so a screen reader hears the headline, the source, the confidence label, and "external link" as one node in the active locale — the year and Latin runs voiced correctly via their locale/bidi tags, the external-link glyph announced as words ("opens in your browser"), decoration excluded. docs/science/11-the-in-app-science-screen.md §7 (semantic label per locale); merge/exclude mechanics consistent with the app's component semantics rule.

Do / Don't

Checklist

Before this source row / grade tag / citation block is done:

• The row renders one verified CLAIMS.md entry (claim text, value/rule, source, grade); no hand-written science, no orphan fact — authoring lives in domain-claims-register-and-science-screen (docs/science/11-the-in-app-science-screen.md §1).
• Two-layer copy: a plain headline (type.body, user's frame) on the face; the named source, grade, and caveat in an optional "the evidence" expansion; FSRS/retrievability glossed once (docs/science/11-the-in-app-science-screen.md §3).
• The named, dated source is on the face (author/year/venue) as readable on-device text, useful with no connection (docs/science/11-the-in-app-science-screen.md §4).
• The grade is plain confidence language about the evidence (e.g. "among the best-established findings" for [MA]), conveyed by text tag + a non-color glyph — never ★★★★★, a "proven" badge, a retention %, or color alone (docs/science/11-the-in-app-science-screen.md §5, §7 / SC 1.4.1).
• A [TRAD] source shows collection + number + grading, is framed as methodology (no fiqh ruling, sect/madhhab-neutral, servant-to-teacher), with "needs scholarly review" shown plainly where pending (docs/science/11-the-in-app-science-screen.md §4, §8; domain-adab-and-religious-integrity).
• Any URL is an optional convenience visibly marked as leaving the app (external glyph + "opens in your browser" hint), opens the system browser (no in-app fetch), and degrades gracefully offline; the citation never depends on the network to be trustworthy (docs/science/11-the-in-app-science-screen.md §2, §4; docs/design-system/10-privacy-and-trust-ux.md §3, §7).
• No coercion or engagement mechanics: no "N of M read," no badge for expanding, no urgency, guilt, or decay-as-threat; calm, declarative copy only (docs/science/11-the-in-app-science-screen.md §6).
• Offline / no-AI by construction: claim, citation, grade, and confidence label ship inside the binary / verified core pack and render locally; no runtime fetch, no model generation; corrections ship as an app/pack update (docs/science/11-the-in-app-science-screen.md §2; domain-asset-pack-integrity).
• Accessibility / RTL: ≥4.5:1 contrast across light/sepia/dark, resizable text at 200% with reflow (no truncation / horizontal scroll); year & numbers in locale numerals (Extended Arabic-Indic fa/ckb, Arabic-Indic ar via intl + type.numeral); Latin author/venue/URL bidi-isolated (FSI/PDI); EdgeInsetsDirectional / logical start/end, external-link glyph at logical end — no hardcoded left/right (docs/science/11-the-in-app-science-screen.md §7; eng-rtl-and-bidi-layout).
• The whole row is one ≥48dp affordance where it taps to a source, MergeSemantics-merged into one localized phrase (headline + source + confidence label + "external link"), each run locale/bidi-tagged, decoration ExcludeSemantics (docs/science/11-the-in-app-science-screen.md §7).
• On-screen wording matches the value/rule the engine actually uses for the claim — no drift between the row's copy and the app's behavior (docs/science/11-the-in-app-science-screen.md §1).
• Widget + golden tests cover the row for an [MA] and a [TRAD] claim across fa/ckb/ar on the real bundled UI fonts (never Ahem), a grayscale/deuteranope check that the grade survives without color, and an HttpOverrides offline guard asserting tapping the link triggers no in-app network call (eng-write-dart-test).

The source row makes one honest, calm claim — here is who found this, when, and how strong the evidence is — so a sceptical ḥāfiẓ, teacher, or scholar can see attributed scholarship, not the app's opinion. If a row ever tempts you toward a star rating, a "proven" badge, a retention promise, a grade-of-the-user, or a citation that dies offline, that is the non-negotiables being violated, not a missing feature. The standard is iḥsān, because the work is built free, lillāh.

Files

• template.dart — copy-paste scaffold: the domain-blind EvidenceGradeTag leaf (confidence text + non-color glyph, never color-only), the SourceCitation leaf (on-device author/year/venue text with locale numerals + bidi-isolated Latin runs and the optional "leaves the app" external link), the feature-layer ScienceSourceRow mapping a registered claim → the two-layer row with MergeSemantics, and the per-locale RTL / offline / no-color golden test stub — with // TODO markers and every token/rule referenced by name.
• references.md — the exact governing doc sections that own this row, each with the one thing to take from it, and the sibling skills.

Related skills: domain-claims-register-and-science-screen (the CLAIMS.md register + the whole science screen this row renders one entry of — the only author of a claim), ui-certainty-label (the grade → lay confidence-phrase badge this row composes in its "the evidence" expansion), domain-scheduling-engine-rules (the engine rule a claim describes but the row never re-derives), domain-adab-and-religious-integrity (the calm, sect-neutral, no-ruling, servant-to-teacher copy and the [TRAD] framing), eng-rtl-and-bidi-layout (locale numerals, FSI/PDI bidi isolation of the Latin author/venue/URL, RTL mirroring), eng-add-localized-string (the ARB strings for the headline, grade label, and "opens in your browser" hint), domain-asset-pack-integrity (the verified offline pack the bundled citation text ships inside), eng-write-dart-test (the per-locale RTL / CVD / offline golden harness on the real bundled fonts).