Building the Storyteller

Making narrative judgment auditable

The System

The newest specialist exists to fix a gap the other five leave behind

canvas-specialists is a bundle of six chained AI specialist agents for structured knowledge work. The Storyteller is the newest addition — added to close a gap the analytical five leave open.

First, what those five actually deliver.

1Researcher

2Formatter

3Data Analyzer

4Competitive Analyst

5Writer

6Storyteller the newest addition

The Proof

The analytical chain produces rigorous, well-cited documents

Researcher → Formatter → Analyzer → Writer is a genuinely working pipeline. What it hands you is rigorous and well-cited — the evidence is real and the citations hold.

The pipeline works. So where's the gap?

→Researcher gathers the evidence

→Formatter structures it

→Analyzer interprets it

→Writer turns it to prose

The Complication

But they read like what they are: structured data transformed into prose

The documents inform without moving. Rigor is not the same as resonance — you can be perfectly correct and still leave the reader unmoved.

Can't you just rewrite them by hand? Bruner says no.

Why It Can't Be Reformatted

Bruner: paradigmatic and narrative are two irreducible modes of thought

You can't reformat your way across. A hand-rewrite can't cross that gap — and it would destroy the traceability the pipeline earned. That's why a real specialist is needed, not an edit pass.

So the fix isn't rewriting. It's selecting.

"Two modes of cognitive functioning, two modes of thought, each providing distinctive ways of ordering experience." Jerome Bruner, Actual Minds, Possible Worlds, 1986

The Turn

The Writer covers everything. The Storyteller selects.

Selection — not summary — is the creative act. The fix isn't compressing a report into fewer words; it's choosing which findings become a story and which stay out.

And every one of those choices gets written down.

The Writer

Covers everything — thorough, complete, comprehensive.

The Storyteller

Selects — chooses what earns a place in the story.

The Mechanism

Every selection is logged in a NARRATIVE SELECTION trail

Included findings get narrative roles; omitted findings get explicit rationales. The editorial reasoning is written down, not hidden — the record is something a reviewer can actually read.

Which means nothing is silently dropped.

Includedrole: Hook
Includedrole: Complication
Includedroles: Evidence · Peak · Resolution
Omittedrationale: off-arc
Omittedrationales: Redundant · Undermines tone · Insufficient

The Payoff

Nothing is silently dropped — selection is no longer invisible

A reviewer can read exactly why each finding stayed or went. And when the input is too thin, the Storyteller fails loud instead of fabricating a story.

Narrative judgment becomes transparent, reviewable, and verifiable.

Every choice is accounted forIncluded findings carry roles; omitted findings carry rationales.
Fails loud, doesn't fabricateThin input triggers a loud failure instead of an invented story.
Judgment made auditableTransparent, reviewable, and verifiable — not invisible.

The Takeaway

A sibling deck put it on trial — and the Storyteller won

"How Do You Know?" validated the specialist with blind A/B/C scoring and 10/10 traps excluded. On narrative memorability the Storyteller beat the Writer, 4.83 vs 3.67 (p=.011).

Auditable selection isn't just principled — it was validated to move readers more.

4.83

Storyteller narrative memorability (of 5)

3.67

Writer narrative memorability (of 5)

p=.011

Blind A/B/C significance

10/10

Traps excluded

Sources

Research Methodology

Feature status: Active — narrative artifact / design case study

Data as of: 2026-07-14 (deck last regenerated spine-first, commit 955767a).

Repository: ramparte/amplifier-stories (checked out at ~/dev/ANext/amplifier-stories). Self-referential per ai_working/pipeline-batch/corpus-source-map.tsv.

Commands run:

git log --oneline --all --follow -- '*story-building-the-storyteller*' — commit trail 2333854 → 95376b8 → 955767a
git show -s --format='%ci %an %s' 2333854 955767a — first staged 2026-03-06 (cpark4x); regenerated 2026-07-14 (Sam Schillace)
git show -s --format='%ci %an %s' 816b6c4 cd2598f — narrative-spine (ABT) guidance added 2026-07-13 (Sam Schillace)
wc -l context/storyteller-instructions.md — 608 lines; Contract 1 = the narrative spine
grep -c 'NARRATIVE SELECTION' docs/story-building-the-storyteller.html — appears 6 times
tag-stripped docs/story-building-the-storyteller.html / docs/story-how-do-you-know.html — verified chain text, Bruner quote, validation figures (4.83 vs 3.67, p=.011, 10/10 traps)
git log --format='%an' -- context/storyteller-instructions.md agents/storyteller.md | sort | uniq -c | sort -rn — author counts

Primary contributors: Sam Schillace (14 commits to storyteller instruction/agent files; spine layer + regeneration); Michael J. Jabbour (12 commits); Chris Park / cpark4x (original author, staged 2026-03-06); sadlilas (promoted/scrubbed 2026-03-11).

Gaps: The canvas-specialists SOURCE definitions (the 6-stage pipeline, STORY OUTPUT header, framework-selection axes, and self-claimed counts) live in external repos (michaeljabbour/amplifier-bundle-graph-canvas, amplifier-bundle-research) not checked out locally, so those specifics are verified only from the deck artifacts, not from source. Bruner, Green & Brock (2000), and Slovic et al. (2007) are external academic citations. NARRATIVE SELECTION example content is illustrative sample data inside the deck.