Skip to main content

May 29, 2026

Gideon Build Log: The First Real Scene Workflow

This was the best Gideon run so far.

Not because everything was perfect.
Because it was real.

The first half of the day was infrastructure: clean up the repo, make Fresh Gideon the app that opens, repair the Desktop launcher, verify Forge/Draft Room save paths, migrate scene drafts to versioned paths, and make the Library stop showing old clutter.

The second half was the actual test:

Can a writer bring messy scene notes into Gideon, talk through the scene, answer questions, lock an outline, build a Scene Card, shape a T1 beat outline, make changes, and save the result into project memory?

For the first time, the answer felt like yes.

Build Thesis

The first real scene workflow held.

Fresh Gideon moved from app repair into actual writing work: notes, questions, outline, Scene Card, T1 beats, user corrections, and saved project memory.

#194-#203Recorded A01S02 run.
5 ClipsWorkflow, not demo theater.
1 SceneA01S02 entered the system.

Before The Recording, Trust Had To Be Repaired

The day started with a simple product rule:

Gideon should not ask the writer to remember what the app should already know.

That meant cleaning the working path before doing creative work.

Fresh Gideon had to open from the Desktop app. It had to land at 127.0.0.1:4327. The health check had to say gideon-fresh. The Library had to show the real A01S00 and A01S01 scene packs. Forge had to suggest the right -002 audit path. Draft Room had to suggest v2 when opening a generated draft alternative.

This was not glamorous work.
It was the work that makes the creative layer believable.

Pre-Recording Verification

The app had to earn the recording.

Desktop app opened Fresh Gideon at 127.0.0.1:4327.
Health confirmed gideon-fresh.
Library showed A01S00/A01S01 scene packs.
Forge and Draft Room save suggestions behaved.

The Storage Model Got Cleaner

We also made a structural decision before recording:

Current drafts should not live in current-draft/.

They should live in drafts/.

But Adam made the important call on version naming:

Do not start with v1.

The first draft is the clean name:

EP101_A01S01.md

The next version is:

EP101_A01S01-v2.md

That matters because creative versioning should feel like a writer’s file system, not a database artifact. Forge receipts can use -002, -003. Drafts use v2, v3.

Storage Rule

First draft is the clean name.

EP101_A01S01.md
EP101_A01S01-v2.md
EP101_A01S01-v3.md

Forge receipts:
EP101_A01S01-T3-forge-audit.md
EP101_A01S01-T3-forge-audit-002.md

The Test Scene: A01S02

The real test was EP101_A01S02.

This is the explosion / public event scene after the harbor and South 101 setup.
The scene’s job is not just spectacle.

The public explosion turns Gallows Bay toward smoke, sirens, injuries, calls, and official response.|
In that same window, the South 101 crew opens Martin Rusk’s storage unit and removes targeted records.

The scene has one core read:

The explosion is terrifying, but it is not the whole event.

That is exactly the kind of sequence Gideon has to support: part story, part continuity, part source memory, part writer instinct.

Scene Engine

Smoke pulls the town one way. Records move the other.

Public event

Explosion, smoke, injuries, sirens, calls, civic attention.

Private operation

South 101, Rusk's unit, red key tag, targeted records.

The Recorded Loop

The practical loop was:

input, output, input, output, save.

That is not because Gideon requires that exact pattern.
It happened because the writer asked for a working version first, made choices, then asked for a final or full version on the second input.

That is the key product lesson:

Gideon should not decide when something becomes official. The writer does.

The writer can keep iterating before saving, save over an artifact, save an alternate version for comparison, or come back later with a different pass.

Working Loop

The writer decides when output becomes memory.

Gideon can keep generating, but the save is deliberate. The user can revise, compare, overwrite, or hold the work before turning it into project memory.

Input Output Input Output Save

The Clips

These clips are not polished lab demos.

They are workflow evidence.

The useful part is watching the writer stay in authorship: giving Gideon raw material, judging the output, changing the prompt, editing notes, correcting canon boundaries, and deciding when to save.

That is where Gideon starts to feel less like a generator and more like a room you can work in.

Recording Session #194-#201

A real Gideon test, captured in clips.

Clips 1-2 / Sessions #194-#198

Talk the scene through, then correct it.

The working outline pass, the prompt drift, the corrected run, and the first proof that the scene could move from raw notes into useful structure.

Watch for: the user catching the artifact drift, adjusting the prompt, and getting the outline back on track without derailing the session.

Product note: correction has to be cheap enough that the writer stays inside the work.

Story work: boots before the blast, explosion as public event, South 101 moving calmly under cover, red marine key tag, and the Harbor Light/referral label.

Useful failure: the phrase "source for the Scene Card next" pulled Gideon toward the wrong artifact. The corrected prompt was smaller and cleaner: use my answers and lock the outline.

Why it matters: the output did not steal the scene. It gave the writer something to answer.

Clip 3 / Session #200

Scene Card save.

The locked outline becomes a source-of-truth Scene Card, saved through Draft Room with Update Canon Memory on.

Watch for: the input/output/save loop working cleanly once the artifact target is precise.

Receipt: Scene Card memory updated; no canon files updated; one skipped item noted.

Output shape: identity, scene pack, pulse, beats, source match table, visual iconography, emotional roles, threat presence, continuity flags, and missing inputs.

Path: Episodes/EP101/act-01/scene-cards/EP101_A01S02.md.

Lesson: writer-approved outline becomes structured project memory.

Clip 4 / Session #201

T1 beat outline.

Gideon builds the T1 spine with bodies in the beats, marks the artifact correctly, and leaves Mara as the key user correction.

Watch for: MAN IN BOOTS, WOMAN 1, Darla Pike, and the red marine key tag turning the outline from abstract plot into playable scene pressure.

User note: Mara is not in this scene. She moves to Scene 3.

Beat handles: MAN IN BOOTS, MAN 1, WOMAN 1, WOMAN 3, MAN 4. Bodies, jobs, movement, pressure.

Artifact status: Gideon marked the run correctly as T1 and scene-beat-outline.

Why it matters: the outline stayed usable for drafting instead of becoming abstract summary.

Clip 5

Lock the T1.

Writer notes go back through Writer's Room and Gideon locks the pass: no Mara, Darla at South 101, red tag beat, label button.

Watch for: Draft Room notes becoming Writer's Room input, then coming back as a locked T1 beat outline.

Final image: HARBOR LIGHT / REFERRALS / OLD disappears into the van under harbor sirens.

Locked choices: MAN 1 is blown back by the blast; WOMAN 1 carries radio/emergency response; Darla Pike works the South 101 office; Len Carver owns South 101 and lives in an RV as a camp-host / maintenance-man figure.

Key image: the red marine tag hangs below the key, swings through frame, then settles as the lock turns.

Next canon gap: Darla Pike and Len Carver need rebuilt profiles.

What The Test Proved

A01S02 proved that Gideon is not overfit to the first two scenes.

The project now has:

– a locked final outline;
– a Scene Card;
– a saved T1 beat outline;
– source-memory receipts;
– recorded evidence of user correction;
– next-session canon tasks.

The most important thing is not the number of files.
The important thing is that the writer could keep moving.
The wrong output was corrected.
The wrong path was corrected.
The raw Markdown friction was named.
The Mara boundary was corrected.
The Darla/Len canon gap was caught.

This is what a working creative tool has to allow: judgment, correction, and continuation.

What We Learned

Correction is part of the product.
Save receipts are trust surfaces.
Draft Room needs a better writing surface.
Needs Work canon memory has to catch named detail.

Next: Canon Recovery And The Network

The next session should go straight at canon.

Darla Pike and Len Carver need profile recovery or rebuild.
That one stings because the profiles may have existed already, and they were good.

The product lesson is clear:

When Gideon sees meaningful named-character detail and no matching canon file exists, it should probably stage that material as Needs Work.

The other big next step is the Network:

who is in it,
who is adjacent,
who is pressured,
who is exploited,
who wants what.

The Black Ledger needs its equivalent of a faction profile before too many more scenes get added.

Next Build Stack

Canon recovery, then the Network.

  1. Rebuild Darla Pike and Len Carver profiles.
  2. Add new character/business canon for fishing guide and jet boat company.
  3. Run a Network alignment canon check.
  4. Create a Black Ledger faction/alignment profile.
  5. Improve Needs Work staging for named details.

Leave a Reply

Discover more from Signal & Salt AI

Subscribe now to keep reading and get access to the full archive.

Continue reading