Perpenda · LLM Systems for PMs

Decision-grade fluency in LLM systems. One trade-off at a time.

A focused Android reader for senior product managers who need to make build / buy / skip calls on AI features — and back the call up with calibrated reasoning. Fifteen units, in order, each one a single trade-off. No streaks. No badges. No daily nudge.

View on GitHub Soon on Google Play.
v1.2 · June 2026 15 units published GPL-3.0
The thesis

A trade-off, not a topic.

Most AI-fluency content treats PMs like coding-bootcamp students. Perpenda treats you like the decision-maker you are. Every unit teaches one concept through the same three-question scaffold — because that's how product decisions actually get made.

§ When this matters

Which feature, which call, which Monday.

Each unit names the concrete decisions the concept changes — cost forecasts, latency budgets, vendor comparisons, scoping calls. Not "important to know," but "load-bearing for these decisions."

§ When this breaks

The way most teams get it wrong.

Failure modes, named. Estimating cost in words instead of tokens. Treating latency as one number. Shipping streaming UX where output is consumed atomically. The mistakes you can recognize after, but ought to recognize before.

§ What it costs

What you'd have to give up to be right.

The discipline the right call demands — naming three latency metrics instead of one, asking eng for benchmarks on your prompt shape, measuring before promising. Cheaper than the failure mode it prevents.

What you'll learn

The vocabulary, tied to the call.

Real terms, straight from the units — each one attached to a decision you actually have to make. Fluency you can use on Monday, not trivia.

Tokens · context windows · prompt caching
Forecast cost and capacity in the unit vendors actually bill — before finance asks.
TTFT · p50/p95 · streaming
Set a latency budget and decide where streaming actually earns its keep.
LLM-as-judge · golden sets · live A/B
Decide how you'll measure quality before you commit to a model.
Fine-tune vs. prompt vs. RAG
Match the fix to the failure mode — knowledge gap, behavior gap, or spec gap.
Recall@K · groundedness · faithfulness (RAGAS)
Scope and evaluate a RAG system on the three axes that actually break.
Agent decomposition · termination · tool schemas
Know when an agent loop earns its cost — and when it's expensive theater.
The path

Fifteen units. One curated order.

One path, in v1. LLM Systems for PMs. Fifteen units are published; more are shaped by the closed beta, rather than planned from the armchair.

§ 01
Tokenization
Settled
Published
§ 02
Context window
Settled
Published
§ 03
Latency
Settled
Published
§ 04
Evals
Settled
Published
§ 05
Model selection
Contested
Published
§ 06
Prompt design basics
Settled
Published
§ 07
Hallucination + reliability
Contested
Published
§ 08
Cost dynamics at scale
Contested
Published
§ 09
Fine-tuning vs. prompting vs. RAG
Contested
Published
§ 10
Vector search / RAG fundamentals
Contested
Published
§ 11
Streaming UX
Contested
Published
§ 12
Tool use · function calling
Contested
Published
§ 13
Multimodal · vision basics
Contested
Published
§ 14
Agents · multi-step reasoning
Contested
Published
§ 15
Safety + content moderation
Contested
Published
§ +
More units
Planned
Shaped by the closed beta
The app

A briefing, not a course app.

The whole loop, screen by screen — find your next unit, read the bite, decide, get per-criterion grading, see the sources, and return for spaced review. Editorial typography, hairline rules, no decorative imagery — the restraint makes the writing earn its space.

Perpenda path home
01 · Path home
Your next unit, one tap away.
Perpenda unit reader
02 · Unit reader
One scroll — definition, bite, depth.
Perpenda decision prompt
03 · Decision prompt
You decide — open-ended, not multiple choice.
Perpenda calibrated grade
04 · The grade
Per-criterion grading — with how sure it is.
Perpenda calibration and sources
05 · Calibration & sources
Every claim tagged and sourced — after you answer.
Perpenda spaced review
06 · Spaced review
Surfaces alongside the path — never a gate.
Calibration

Every claim, sourced and tagged.

Each unit's claims carry a tier: settled, contested, or unsettled. Sources sit after the decision prompt — never before — so the consensus doesn't prime your answer. The grader will tell you when it doesn't have enough signal to grade fairly.

Settled
Models bill and meter context windows in tokens, not characters or words.
Settled
BPE-style subword tokenization is the dominant scheme across frontier models.
Contested
Whether token-level pricing is the right unit of cost for product decisions long-term — vs. characters, semantic units, or compute-time billing — is unsettled as the market evolves.
Unsettled
Whether next-generation small-model latency improvements will obviate the streaming-vs-blocking distinction.
Confidence as a feature, not a softening.
Questions

Before you ask.

What is Perpenda?

Perpenda is a focused Android reader that teaches LLM systems to senior product managers — one trade-off at a time. It works through a curated path of single-trade-off units, in a fixed order, with calibrated grading against rubric criteria. No streaks, no badges, no daily nudge.

Who is Perpenda for?

Product professionals with stakes in AI-shaping decisions — product managers first, but also product marketers, founders, design leads, and product-adjacent execs, plus career-switchers moving into the field. Anyone who makes build / buy / skip calls on AI features and needs to back them with calibrated reasoning — people who'd rather decide right than read fast.

What's Perpenda's approach?

Perpenda teaches decisions before mechanism. Every unit is a single trade-off — when it matters, when it breaks, what it costs to get right — and every claim is sourced and tagged by confidence (settled, contested, unsettled). The aim is judgment you can act on, not theory you can recite.

What do I need to run it?

Perpenda is an open-source Android app, released under the GPL-3.0 license, with no ads and no behavioural tracking. All you need to run it is a modern Android phone and an account.

How many units are available?

Fifteen units are published today, each a single decision trade-off. The final units are deliberately left open — they're shaped during the closed beta rather than planned in advance.

What does "calibrated grading" mean?

Each unit's claims carry a calibration tier — settled, contested, or unsettled — and sources sit after the decision prompt, never before, so the consensus doesn't prime your answer. The grader will tell you when it doesn't have enough signal to grade fairly.

Built for the call you have to make on Monday.

Fifteen units, calibrated grading, the full loop. More units are shaped by real use, not the armchair.

Perpenda is in closed testing on Google Play — email to be added as a tester.
No streaks. No badges. No daily nudge.