What is scarce?
Attention and comprehension, not links.
08 / 09 · Knowledge Discovery & Ranking Product
Threadline
Infinite content. Finite attention.
Threadline is not another feed. It helps a reader maintain an evolving understanding of a topic: what is genuinely new, what contradicts prior thinking, which claims have evidence, what is repetitive, and which four things are worth reading now.
KNOWLEDGE + ECOSYSTEM
How do you decide what information deserves someone's limited attention?
This case is not ‘Tonight for articles.’ It models claims, information gain, creator attribution and a shrinking queue.
01 / THE PRODUCT PROBLEM
Start with the decision the user cannot make well today.
Threadline is not another feed. It helps a reader maintain an evolving understanding of a topic: what is genuinely new, what contradicts prior thinking, which claims have evidence, what is repetitive, and which four things are worth reading now.
The portfolio test: a client should understand the human problem before seeing a model, agent or framework.
02 / HOW RIKA THINKS
From ambiguity to a decision system.
The five-step reasoning pattern stays consistent; the actual product logic is specific to this problem.
What makes a read valuable?
Relevance, trust, novelty, contradiction value, information gain and attention cost.
What does a feed optimize poorly?
Recency and clicks can reward repetition, outrage and source concentration.
What should the product create?
A small reading queue connected to a claim/idea graph and the reader's knowledge state.
What should learning mean?
Update from read/save/dismiss/follow and whether an item actually changed the reader's understanding.
03 / TRANSFERABLE PROBLEMS
The pattern travels. The constraints change the solution.
This is how I would reuse the thinking without copying the product.
Academic literature discovery
Shared ranking problem; citation authority, methodological quality and novelty dominate.
Competitive intelligence
Shared overload problem; business impact, recency and company relevance become stronger weights.
Executive briefing
Shared finite-attention problem; urgency and decision impact matter more than topic curiosity.
Customer feedback prioritization
Same ‘too many signals’ pattern; frequency, severity and strategic segment replace source novelty.
04 / USERS + BUYER
Who receives value, who operates it, and who pays?
AI products often fail when one persona is treated as ‘the user.’
Knowledge worker following a fast-moving topic
Reads across newsletters, blogs, research, publishers and saved links.
Creator / publisher
Wants attribution, qualified readers and direct follow relationships—not AI replacing the original work.
Professional information product / research community
Needs a differentiated discovery layer that respects sources and builds repeat topic habits.
JOBS TO BE DONE
Functional, emotional and social value.
Tell me the few pieces that materially improve my understanding of a topic.
Reduce the guilt and anxiety of an endlessly growing reading backlog.
Help me form and communicate an informed point of view rather than merely repeat the loudest feed.
CUSTOMER JOURNEY
The question changes as evidence arrives.
Define thread
Choose a topic/question and depth.
Reader model records what is already known.
Ingest
Normalize sources, creators and canonical URLs.
Duplicates and syndication collapse.
Map ideas
Extract claims, evidence, entities and contradictions.
Build/update idea graph.
Prioritize
Rank by information gain + trust + novelty + diversity − attention cost.
Return four, not forty.
Read & evolve
Open original, save, dismiss, annotate or follow creator.
Reader knowledge state changes.
05 / PRODUCT DECISION RECORDS
The product is the sum of choices and trade-offs.
Each decision includes an implicit reversal test: better evidence can change the choice.
Follow ideas, not feeds
The persistent object is a topic/claim thread rather than a chronological source stream.
Information gain over generic relevance
An item can be relevant yet useless if it repeats what the reader already knows.
AI should orient, not replace the creator
Summaries are short and cited; original-source clickthrough and creator attribution are first-class.
The queue should shrink
A successful system helps the user ignore more content with confidence.
06 / AI SYSTEM
The architecture separates AI judgment from exact rules, evidence, tools and human authority.
Every connector has a defined job. Animated dashed lines represent active evaluation/learning loops rather than decorative motion.
Primary decision/data flowContext/evidenceHuman/consequential pathContinuous evaluation loop
HOW I WOULD BUILD IT
Point by point, from state to operation.
Source normalization
Canonicalize URL, publisher, author, date, content type and access state.
Claim extraction
LLM extracts atomic claims, evidence links and entities into a structured idea graph.
Reader knowledge state
Track read/saved/dismissed claims and topics—not private raw history beyond what is needed.
Candidate generation
Retrieve by topic, entities, trusted sources and claim neighborhoods.
Trust + novelty features
Estimate source provenance, corroboration, novelty against prior reads and contradiction value.
Multi-objective rank
Balance relevance, insight, trust, diversity and attention cost.
Grounded orientation
Generate a short ‘why it matters / what it adds’ preview with source attribution.
Feedback + graph update
Reader action and new claims update ranking and the knowledge graph.
07 / EVALUATION ARCHITECTURE
How do I know the AI deserves to ship?
Component eval, system eval and product outcome are deliberately separated.
Claim atomicity
Claim atomicity, entity linking, citation alignment and duplicate/near-duplicate detection.
Relevance
Relevance, novelty, diversity, source concentration and information-gain judgments.
Faithfulness and source attribution; previ
Faithfulness and source attribution; preview must not introduce unsupported claims.
Original-source clickthrough and creator f
Original-source clickthrough and creator follows are monitored so summaries do not cannibalize creators.
Qualified read rate
Qualified read rate, save-to-read conversion, queue shrinkage and repeat thread return.
Source provenance
Source provenance, paywall/licensing flags and reader-history privacy.
Closed loop: trace → classify failure → add/refresh eval case → change source/retrieval/model/prompt/rule/tool → regression suite → controlled release → monitor outcomes/overrides → repeat.
08 / FEATURE PRD
A concrete example of how strategy becomes engineering work.
The PRD is intentionally feature-level and includes AI behavior, deterministic controls, telemetry and non-goals.
PRD example — Four-read knowledge queue
Problem
A reader follows a topic across fragmented sources and cannot distinguish genuinely additive work from repetitive commentary.
Outcome
Return a four-item queue where each item has a distinct reason for inclusion: new evidence, opposing argument, trusted update or high-novelty perspective.
User stories
- As a reader, I can see what an article adds relative to what I have already read.
- As a creator, my original work remains clearly attributed and one click away.
Functional + AI requirements
- Normalize and deduplicate candidate URLs before ranking.
- Extract claims and evidence links into an idea graph.
- Compute novelty against reader knowledge state, not only global corpus similarity.
- Apply source-diversity constraint so one publisher cannot dominate all four slots.
- Generate short preview with citation and explicit ‘what this adds’.
- Every card links to the original source; long-form content is not reproduced.
Acceptance criteria
- No duplicate/syndicated versions occupy multiple slots.
- At least three distinct sources when the candidate pool permits.
- Every generated preview is supported by the source.
- Dismiss/follow/save updates the next queue.
- A creator/source can be muted without deleting the topic thread.
Telemetry
thread_created · queue_generated · why_opened · original_opened · saved · dismissed · creator_followed · queue_refreshed
Non-goals
- Replacing original articles with full AI summaries
- Chronological infinite scroll
- Treating clicks as the only ranking objective
Engineering handoff
Typed state/schema · API/tool contracts · error states · permissions · eval fixtures · analytics events · rollout/rollback.
09 / BUILD VS BUY + RELIABILITY
Do not custom-build commodity infrastructure—and do not pretend every API always works.
BUILD DIFFERENTIATION
Claim graph + information-gain ranking + creator value
Use standard infrastructure for storage, models, tracing and connectors where it does not create strategic advantage.
BAD-DAY MODE
Fallback to trusted-source relevance queue
Timeouts, stale data, partial results, rate limits and provider failures are explicit product states—not invisible backend details.
10 / WORKING PRODUCT JOURNEY
Use the product from input to changed state.
Every click represents a user/product decision and explains why the information is needed.
THREADLINE · KNOWLEDGE QUEUE
The goal is not to give you more content.
Threadline succeeds when it helps you ignore repetitive content and identify what materially changes your understanding.
Four reads. Four distinct reasons.
NEW EVIDENCE
Workflow redesign before agents
8 min · institutional research
COUNTER-ARGUMENT
Where agent ROI breaks
11 min · practitioner essay
TRUSTED UPDATE
Autonomy needs risk tiers
6 min · enterprise research
NOVEL VIEW
Small teams, agent-heavy workflows
7 min · independent creator
Workflow redesign before agents
What this adds
Adds operating-model evidence you have not read yet.
PROVENANCE
Original source
Creator, publisher and canonical URL remain visible.
IDEA GRAPH
3 new claims
1 reinforces, 1 adds evidence, 1 challenges prior reading.
Your knowledge state changed.
Original-source read opened.
The next queue can reduce repetition, surface contradictions and preserve source diversity.
Was this case study useful?
A quick signal helps me understand what is useful to recruiters, founders and product teams.
11 / PRODUCT LIFECYCLE
Autonomy and market scope are earned in stages.
Research dossier
Validate reading overload and creator/source needs.
Topic queue
Manual/trusted-source corpus + four-read ranking.
Idea graph
Claim extraction, contradiction and information-gain features.
Creator ecosystem
Follows, attribution and source controls.
Professional intelligence
Team/shared threads, research workspaces and domain adapters.
GTM + ADOPTION
A credible route into the market.
Initial consumer/prosumer wedge; later professional intelligence/team use cases.
AI/product/technology professionals with high information volume.
Public topic pages, creator referrals and shareable research threads.
Individual threads → team research → specialized domain intelligence.
KILL / PIVOT CRITERION
What would make me stop?
Stop if readers mainly consume previews instead of original work, or if ‘information gain’ does not outperform simpler relevance/diversity ranking in meaningful read behavior.
12 / TOOL + MODEL DECISIONS
Tools are mapped to responsibility—not displayed as decoration.
Where the source material does not prove an implementation, the portfolio says proposed/candidate rather than “built with.”
Crawler / feed adapterssource ingestion
Embedding modeltopic/claim retrieval
Graph storeclaim/entity relationships
LLM structured extractionclaims + previews
Rankermulti-objective scoring
Analyticsqualified reading behavior
13 / SECONDARY RESEARCH
Evidence establishes the problem environment. It does not magically validate the solution.
Reuters Institute Digital News Report 2025
Across markets, 40% say they sometimes or often avoid news; selective avoiders cite overload among multiple reasons. This supports the broader attention environment, not the exact Threadline product.
Open source ↗Reuters Institute — notification overload
Reuters reports publishers use notifications to build habit while users can feel overwhelmed and disable them. Product implication: direct relationships need prioritization, not simply more alerts.
Open source ↗Portfolio research hypothesis
The stronger Threadline hypothesis is professional long-form fragmentation + information gain. That requires primary research with knowledge workers and creators before being treated as validated demand.
Internal/source-project basisEVIDENCE PACK
Download Threadline Product Evidence.xlsx ↗The working assumptions, PRD, eval suite and roadmap are inspectable.
No spreadsheet preview is embedded. Open the workbook only if you want the detail.