One small kernel, one large ecosystem
Amplifier accumulated dozens of separate per-feature and per-project decks, each proud of its own numbers. Read one at a time they look like scattered wins. Read together they trace a single argument about how Amplifier is actually built.
First, does that corpus even hold up to inspection?
The story is self-referential and checkable. Every one of the 29 source decks it names is actually present in docs/: seven via .html links, twenty-two by slug mention.
The corpus is real. But real isn't the same as coherent.
Read one deck at a time and you get a pile of proud, disconnected metrics. Nothing in any single deck tells you why the whole ecosystem holds together. The wins are real; the throughline is missing until you go looking for the shared discipline underneath them.
So look underneath — at the thing every deck quietly depends on.
amplifier-core is the kernel every bundle in the ecosystem builds on — and the thing that could have sprawled but didn't. It's active and heavily maintained, with one dominant author keeping it coherent.
Its defining trait isn't how much it does. It's how little.
amplifier-core is just 18 source files, and its interface layer defines only 5 core protocols — Orchestrator, Provider, Tool, ContextManager, and HookHandler. A small, stable core is the single lever the whole ecosystem turns on.
Hold the core still, and watch what the edges can do.
A still center is exactly what buys the ecosystem its speed. 29 bundles compose on top of the kernel, and just one project — amplifier-tui — already carries an AI-built suite of 1,454 test functions across 38 test files.
That trade — small center, fast edges — is the whole point.
An 18-file, 5-protocol kernel sitting under 29 bundles. The whole ecosystem moves fast precisely because its core refuses to grow. That's the one discipline, held consistently, that every scattered deck was making the case for all along.
Which leaves one principle worth keeping.
The pattern generalizes past Amplifier: a small, stable core plus an auditable trail. Build so anyone can trace the story back to the source — the way all 29 cited decks resolve to real files in docs/.
The methodology behind every number here follows.
Feature status: Active
Data as of: July 21, 2026 (latest amplifier-stories commit 9230211, 2026-07-21).
Primary source repo: ramparte/amplifier-stories (confirmed via git remote -v). Cross-checked against ramparte/amplifier-core, ramparte/amplifier-foundation (upstream microsoft/amplifier-foundation), and ramparte/amplifier-tui. No anthropics/* forks used.
Commands run (independently reproduced):
find amplifier_core -name '*.py' | grep -vE 'test|validation' | wc -l → 18grep -nE 'class .*\bProtocol\b' amplifier_core/interfaces.py → Orchestrator, Provider, Tool, ContextManager, HookHandler (5)ls -d amplifier-bundle-* | wc -l → 29grep -rE '^\s*(async )?def test_' --include='*.py' . | wc -l → 1,454; find . -name 'test_*.py' | wc -l → 38git shortlog -sne → core 134/139 (96.4%), foundation 470/502 (93.6%), Brian Krabachgrep -oE '[a-z0-9-]+\.html' docs/the-story-of-all-stories.html | sort -u + per-slug [ -f docs/<slug>.html ] → 29 cited, 0 missingGaps: Several older per-feature deck figures (e.g. a "84 presentations" count, "~2,400 lines of kernel", "922 tests") are point-in-time or unverified snapshots and were deliberately excluded from this deck. Only independently reproduced numbers appear on the slides.
Primary contributor: Brian Krabach (96.4% of amplifier-core, 93.6% of amplifier-foundation commits).