Guides
Why Map Narratives to Propositions?
Most professional communication arrives as narratives: emails, meeting notes, reports, customer stories, press coverage, product feedback, audit findings. Narratives are powerful, but they’re also messy—mixing facts, interpretations, assumptions, and emotional framing. When you need to make decisions, align stakeholders, or measure changes over time, the mess becomes the problem.
Retelnist’s approach is to convert narrative content into propositions—small, testable belief units—each assigned a stable identifier: a P-ID. A P-ID is not just a label; it’s a handle you can use to track the same idea across documents, time periods, teams, and systems.
This guide shows how to do that conversion in a repeatable way.
Core Concepts: Narrative vs. Proposition vs. P-ID
Narrative
A sequence of statements that implies meaning through context, ordering, tone, and selective detail.
Proposition
A single claim that can be evaluated as true/false, more/less supported, or unknown given evidence. A good proposition is specific enough to test, yet broad enough to reuse.
P-ID
A persistent identifier assigned to a proposition so the proposition becomes a measurable object. You can attach evidence, confidence ratings, ownership, status, and history to a P-ID.
A helpful mental model:
- Narrative = “What happened and what it means”
- Proposition = “One claim embedded in that story”
- P-ID = “The durable reference to that claim across time”
Step 1: Define the Decision Scope (What Are You Trying to Measure?)
Before extracting propositions, establish scope so you don’t create an unmanageable cloud of claims.
Ask:
- What decisions will this mapping support? (e.g., product roadmap, risk mitigation, hiring, legal posture, brand response)
- What time horizon matters? (this quarter, this incident, multi-year trend)
- What domains are in-bounds? (security, customer satisfaction, financial risk, operational reliability)
- Who will use the P-IDs and how? (dashboards, weekly reviews, incident retros, policy updates)
Actionable tip: write a one-sentence “mapping intent,” such as:
“Convert customer support narratives into trackable propositions about product reliability and response quality.”
Step 2: Segment the Narrative Into Atomic Claims
Take a narrative and break it into claim-sized chunks. You are not summarizing; you are isolating assertions.
Look for:
- Causal statements (“because,” “therefore,” “led to”)
- Comparisons (“worse than,” “improved,” “declined”)
- Generalizations (“users always,” “the team never”)
- Obligations and norms (“should,” “must,” “policy requires”)
- Forecasts (“will likely,” “expected to”)
- Attributions (“the customer said,” “the audit found”)
Practical method:
- Copy the narrative into a working document.
- Put each sentence (or clause) on its own line.
- Highlight anything that is an opinion, assumption, or implication—those often hide propositions.
Actionable tip: if one line contains “and,” it probably contains multiple propositions.
Step 3: Rewrite Each Claim as a Clean Proposition
Now convert each chunk into a proposition with a consistent structure:
- Subject (entity)
- Predicate (what is claimed)
- Conditions (when/where/for whom)
- Measurability (how it could be tested)
Use this template:
[Entity] [claim] [under conditions/timeframe], as measured by [signal/evidence].
Examples of clean proposition phrasing:
- “The onboarding flow causes a drop-off at step 3 for new mobile users.”
- “Response time to priority-1 incidents exceeds the stated SLA during weekends.”
- “The new pricing page increases trial-to-paid conversion for small teams.”
Common rewrite fixes:
- Replace vague adjectives with measurable constructs
- “Bad performance” → “Median page load time exceeds 3 seconds on 4G networks.”
- Remove hidden bundles
- “Support is slow and unhelpful” → split into “Support first-response time exceeds X” and “Support resolutions fail to address the reported issue.”
- Avoid mind-reading
- “Customers feel ignored” → “Customers report lack of updates during ticket resolution.”
Actionable tip: keep propositions neutral. Don’t embed the conclusion inside the wording (“The team negligently ignored…”). Save evaluation for evidence and status.
Step 4: Decide When Two Propositions Are the Same (Identity Rules)
A major benefit of P-IDs is that they let you recognize the same claim across many sources. That requires rules for identity.
Treat propositions as the same if they share:
- The same core subject (same system, team, product, policy)
- The same predicate (same alleged behavior/outcome)
- The same conditions (timeframe, segment, environment)
- The same meaning even if wording differs
Treat them as different if:
- The timeframe changes materially (“in Q1” vs “since 2022”)
- The segment changes (“enterprise” vs “SMB”)
- The metric changes (“response time” vs “resolution time”)
- The causal direction changes (“X causes Y” vs “Y causes X”)
Actionable tip: keep one proposition general, and use “child” propositions for segmentation. For example:
- Parent: “Checkout failures occur above an acceptable threshold.”
- Children: “Checkout failures spike on iOS,” “Checkout failures spike during promotions.”
Step 5: Assign a P-ID and Capture Minimal Metadata
Once a proposition is stable, assign it a P-ID. The format matters less than the discipline: one proposition, one identifier.
At minimum, store:
- P-ID (unique)
- Proposition text (canonical wording)
- Owner (person/team responsible for upkeep)
- Domain (risk, product, compliance, people ops)
- Status (active, deprecated, split, merged)
- Created date and last reviewed
- Tags (segment, system, market, policy area)
Optional but highly useful:
- Confidence level (e.g., low/medium/high)
- Polarity (positive/negative/neutral impact)
- Evidence links/notes (without embedding the evidence into the proposition itself)
- Dependencies (related P-IDs, prerequisite claims)
Actionable tip: choose a single “canonical sentence” for each P-ID and treat all other phrasings as aliases.
Step 6: Attach Evidence and Track Belief Over Time
A proposition becomes measurable when you can attach evolving support. Retelnist-style mapping separates the claim from its support.
For each P-ID, maintain an evidence log:
- Observation (what was seen)
- Source type (ticket, interview, audit note, metric readout)
- Date/time
- Strength (strong/medium/weak)
- Interpretation notes (what the evidence implies and what it doesn’t)
Then update:
- Confidence: Are you more or less convinced than last review?
- Direction: Is the proposition trending toward confirmed, disputed, or unresolved?
- Next test: What would most efficiently raise or lower confidence?
Actionable tip: when evidence is mixed, do not “average it out.” Instead, split the proposition by condition until evidence becomes coherent.
Step 7: Handle Conflicts, Ambiguity, and Loaded Language
Narratives often contain contradictions. The goal isn’t to force agreement; it’s to make disagreement explicit and manageable.
Techniques that work in practice
-
Create competing propositions
- P-A: “Feature X reduces churn for new users.”
- P-B: “Feature X increases churn for new users.” Track evidence for each; don’t merge prematurely.
-
Convert judgments into testable claims
- “The rollout was a disaster” → “The rollout caused an increase in incidents above baseline” and “Customer complaints increased above baseline.”
-
Separate descriptive from normative
- Descriptive: “SLA breaches increased last month.”
- Normative: “We should revise the on-call schedule.” Different proposition types can coexist, but they should not be conflated.
Actionable tip: if a statement contains blame, translate it into process or system claims that can be validated.
Step 8: Build a Proposition Map (So P-IDs Become a System)
A library of P-IDs becomes exponentially more useful when you connect them.
Common relationship types:
- Supports: P-101 supports P-020
- Contradicts: P-044 contradicts P-044b
- Depends on: P-330 depends on P-112
- Causes/Leads to: P-210 causes P-211 (use carefully; causality needs strong evidence)
- Part-of: P-500 is a component of P-050 (parent/child)
Actionable tip: start with simple relationships (supports/contradicts/part-of) before modeling complex causality.
Step 9: Operationalize P-IDs in Your Workflow
To make mapping stick, embed it where decisions happen.
Practical integration points:
- Meeting notes: include a “P-IDs touched” section
- Incident reviews: identify which P-IDs were confirmed, weakened, or created
- Quarterly planning: prioritize initiatives that move key P-IDs
- Stakeholder updates: report changes in confidence and evidence for top propositions
- Knowledge handoffs: new team members inherit a proposition set, not a pile of documents
Actionable tip: cap your “active proposition set” per team (for example, the top 25–50 P-IDs). Archive the rest to avoid dilution.
A Simple Quality Checklist for Each P-ID
Before finalizing a proposition, verify:
- Atomic: one claim, not a bundle
- Testable: you can imagine evidence that would change belief
- Scoped: subject and conditions are clear
- Neutral: not a conclusion disguised as a claim
- Reusable: likely to appear again across narratives
- Comparable over time: can be reviewed monthly/quarterly
Closing: What You Gain by Mapping Narratives to P-IDs
Mapping narratives to propositions turns qualitative noise into a measurable belief system. Instead of debating stories, teams can:
- Identify exactly what they agree or disagree on
- Track how belief changes as evidence arrives
- Detect recurring claims across customers, quarters, or departments
- Make decisions with clearer accountability and less re-litigation
The key is consistency: extract atomic claims, rewrite them as testable propositions, assign durable P-IDs, and maintain evidence and confidence over time. Once that discipline is in place, narratives stop being the end product—and become the raw material for structured, decision-grade knowledge.