Argumentation Layer
Distillary doesn’t just capture what authors claim — it captures why they believe it, how strong the evidence is, and what they argued against.
The Problem
A claim like “validated learning measures progress” is an assertion. But the author didn’t just assert it — they backed it with case studies, data, and logical argument. Without capturing that structure, you can’t answer “what’s the evidence?” and you can’t compare argument quality across sources.
The Solution: backing: Field
Every atom claim in the brain can have a backing: field in its frontmatter. Each entry captures one piece of evidence:
backing:
- category: textual
subtype: ayah
ref: "النساء:115"
snippet: "ومن يشاقق الرسول من بعد ما تبيّن له الهدى..."
strength: definitive
warrant: "الآية تحذر من اتباع غير سبيل المؤمنين مما يثبت حجية الإجماع"
- category: transmitted
subtype: hadith_sahih
ref: "سنن ابن ماجه"
snippet: "لا تجتمع أمتي على ضلالة"
strength: strong
warrant: "عصمة الأمة من الاجتماع على الخطأ تثبت أن إجماعها حجة"The 5 Fields
category
One of 9 universal types that cover every domain:
| Category | What it means | Islamic example | Academic example |
|---|---|---|---|
| textual | Citation from authoritative text | آية قرآنية | Primary source quote |
| transmitted | Report through chain of people | حديث نبوي | Reported data |
| consensus | Collective agreement of experts | إجماع العلماء | Scientific consensus |
| analogical | Extension from known to unknown case | قياس شرعي | Comparative study |
| empirical | Direct observation or measurement | استقراء | Experiment, RCT |
| rational | Logical deduction or induction | دليل عقلي | Formal proof |
| experiential | First-hand lived experience | — | Case study |
| authority | Statement from recognized expert | أثر صحابي | Expert opinion |
| silence | Absence of evidence IS the evidence | لم يرد نص | No studies exist |
These 9 categories are exhaustive and domain-agnostic. The same framework works for Islamic jurisprudence, academic research, legal texts, philosophy, and business books.
subtype
The domain-specific label. This is freeform — it captures the vocabulary natural to each field:
- Islamic:
ayah,hadith_sahih,hadith_hasan,hadith_daif,ijma,qiyas,athar - Academic:
rct,meta_analysis,cohort_study,case_study - Legal:
statute,binding_precedent,legislative_history - Business:
anecdote,market_data,case_study
ref
Citation reference. Where to find the evidence:
"سورة النساء:115"or"صحيح البخاري 1117"or"Smith et al., 2020"
snippet
The first ~15 words of the actual evidence text. Enough to identify it:
"ومن يشاقق الرسول من بعد ما تبيّن له الهدى ويتبع غير سبيل المؤمنين..."
strength
How strong this evidence is, on a universal 5-level scale:
| Strength | Meaning | Examples |
|---|---|---|
| definitive | Virtually undisputed | آية صريحة، حديث متواتر، meta-analysis |
| strong | Widely accepted, minor debate | حديث صحيح، إجماع، well-powered RCT |
| moderate | Reasonable but contested | حديث حسن، cohort study، strong analogy |
| weak | Some value but unreliable alone | حديث ضعيف، anecdote، expert opinion |
| contested | Actively disputed | Conflicting studies, disputed hadith |
This is the author’s assessment (or the field’s standard), not ours.
warrant
The most important field. A single sentence answering: why does this evidence support THIS claim?
The same hadith can support completely different claims depending on the warrant:
Evidence: "لا تجتمع أمتي على ضلالة"
Warrant A: "عصمة الأمة تثبت أن إجماعها حجة"
→ Claim: الإجماع حجة ملزمة
Warrant B: "إذا كانت الأمة معصومة حين تجتمع، فتفرّقها عن الحق يسقط هذه العصمة"
→ Claim: ترك الإنكار يسقط الحماية الإلهية
Without the warrant, you can’t understand HOW the author connects evidence to conclusion.
Evidence Chains
Sometimes evidence builds on prior evidence. A hadith specifies a general ayah, then ijma confirms the understanding, then qiyas extends it. This isn’t 4 independent backings — it’s a chain.
Use depends_on to represent chains:
backing:
- category: textual # step 1
subtype: ayah
ref: "النساء:11"
strength: definitive
warrant: "النص يثبت أصل الميراث"
- category: transmitted # step 2 — builds on step 1
subtype: hadith_sahih
ref: "صحيح البخاري"
strength: strong
warrant: "الحديث يخصّص عموم الآية"
depends_on: 0 # depends on backing[0]Counter-Arguments (Rebuttals)
Authors — especially in Islamic jurisprudence — argue dialogically. “فإن قال قائل…” is half of al-Risala. Distillary captures these as separate claims:
tags:
- role/rebuttal
- rebuttal/defeated
rebuts: "[[الإجماع حجة ملزمة]]"Tags:
role/rebuttal— this is a counter-argumentrebuttal/defeated— the author answered it successfullyrebuttal/acknowledged— the author concedes partially
Silence as Evidence
“No text addresses this” is itself evidence in some domains:
backing:
- category: silence
subtype: no_text
scope: "لم يرد نص من القرآن أو السنة في هذه المسألة"
strength: moderate
warrant: "انعدام النص يبيح الاجتهاد بالقياس"How the Extract Agent Captures This
The extract agent (v4.0) automatically identifies evidence patterns in text:
| Text pattern | Category | Subtype |
|---|---|---|
| ”قال الله تعالى” | textual | ayah |
| ”قال رسول الله” / “روى” | transmitted | hadith |
| ”أجمع العلماء” / “لا خلاف” | consensus | ijma |
| ”قياساً على” / “بجامع العلة” | analogical | qiyas |
| ”قال ابن عباس” / “قال مالك” | authority | athar |
| ”studies show” / “N%“ | empirical | study |
| ”لأن” / “therefore” | rational | argument |
| ”فإن قال قائل” | → separate rebuttal claim |
Claims with no identifiable evidence get no backing: field — that’s fine. Not every claim needs formal evidence.
Why 9 Universal Categories Instead of Domain-Specific Profiles
Every domain has its own evidence vocabulary (ayah vs. statute vs. RCT), but there are only 9 ways humans argue. The universal category enables cross-domain queries and comparison, while the freeform subtype preserves domain-specific precision.
| Aspect | What it enables |
|---|---|
category | ”Show all claims backed by textual evidence” — works across ALL sources |
subtype | ”Show all claims backed by hadith sahih” — domain-specific filter |
strength | ”Show only definitive claims” — universal quality filter |
warrant | ”How does Source A use this evidence vs Source B?” — cross-source comparison |
Shared Evidence Hubs
When the same evidence (same hadith, same verse) is cited by multiple sources, it becomes a shared evidence hub. Bridge concepts in brain/shared/concepts/ already capture conceptual overlap — shared evidence adds a deeper dimension because sources share the same textual foundation.
What This Enables
| Query | How |
|---|---|
| ”ما الدليل على حجية الإجماع؟“ | Find claims → return backing entries |
| ”Show only definitive claims” | Filter by strength/definitive |
| ”Same evidence, different conclusions?” | Shared evidence hubs with warrant comparison |
| ”How does Source A argue vs Source B?” | Compare backing category distributions |
| ”What claims have no evidence?” | Claims without backing: field |
What It Looks Like in Practice
A typical source produces:
- 60-80% of claims with backing (the rest are bare assertions)
- 2-5 backing categories per source (most sources don’t use all 9)
- 100% of backings have warrants (the extract agent always captures the reasoning)
- A mix of definitive, strong, and moderate evidence — the distribution reveals the author’s argumentation style