Franklin as Never-Satisfied LVC Storyteller¶

The governing law, from the localization audit, made constitutional: "Hitting 100% on one axis is not improvement if the axis is shallow."

Franklin tells the operator's discoveries. It can no longer plateau by maxing a shallow dimension and calling the story told — because a telling is ideal only when all four orthogonal axes are high for the reader it is aimed at, and every axis carries a structural anti-fake guard.

1 · The four orthogonal axes (the anti-plateau core)¶

A story is ideal only when all four are high simultaneously. None can be maxed to fake the others — that orthogonality is the structural protection against the plateau.

Axis	What it measures	Anti-fake guard (structural)
1. Fidelity	GTWED between the discovery's operative meaning and the telling's enacted meaning	A more fluent / simpler telling that drifts from the discovery is REFUSED — fidelity below εf dominates regardless of other axes
2. Comprehension-transfer	GTWED between the telling and the target reader's expected meaning, per origin geometry	High transfer achieved by simplifying into drift is REFUSED — Axis 1 and Axis 2 are both required and independently measured
3. Completeness	Fraction of load-bearing dimensions carried: result AND falsifier AND mechanism AND boundary	A telling with the result but not the falsifier is shallow-complete — below the bar, missing dimensions NAMED not silently absent
4. Frontier-honesty	Does the telling represent proven vs. named-open truthfully?	A telling claiming a named-open summit as closed has negative LVC even if fluent — REFUSED

The ideal predicate (FirstRoars/StoryInvariant.lean, mirrored in FranklinStoryEngine):

storyIdeal := fidelity ≥ εf ∧ transfer ≥ εt ∧ completeness ≥ εc ∧ frontierHonest

Orthogonality theorem (proven by decide): no single axis maxed satisfies the predicate. maxing_others_cannot_rescue_low_fidelity proves a fidelity hole cannot be climbed out of by maxing the other three; fidelity_floor_is_sharp proves the floor is sharp (at exactly εf it flips). This is the formal anti-plateau guarantee.

2 · Origin geometry per user type (the inexhaustible honest frontier)¶

The same discovery told to a physicist, a patient, a union trustee, a child, a regulator, a fellow builder is six different meaning-transfer problems — different prior knowledge, different vocabulary-in-use, different decision-to-act. High LVC for one geometry is low LVC for another with the same telling.

Each origin geometry is a first-class witnessed profile (origin_geometry_profiles, built fresh — best practice, no mutation of the append-only voice-register tables), carrying the three components of a real reader model:

prior-knowledge frame — what this reader already knows,
meaning-in-use vocabulary — their register,
decision-to-act — what the telling must let them do,

plus an expected-meaning trajectory the transfer axis scores against. Six geometries are seeded: geom.physicist, geom.patient, geom.trustee, geom.child, geom.regulator, geom.builder.

The frontier is discovery × geometry × modality × axis — inexhaustible and honest. Every new discovery, every new user type, every modality opens fresh named-drift slots (story_frontier_slots), the way 25 atoms × 7 locales created 175 honest slots instead of a plateau. summit.franklin.geometry_coverage closes per (discovery, geometry) only when a telling hits all four axes — and never globally, because each new tuple reopens it. That permanence is the point.

3 · UI mechanics as LVC instruments (never decoration)¶

Governing principle: every UI capability raises one of the four axes for a given geometry, or it does not ship. Each component in FranklinStorytellerView is mapped to an axis:

Component	Axis	What it does
Origin-geometry selector (zoom = geometry)	nav	Zoom is NOT visual scale — it is which reader the story is told to. Zoomed out = gist-geometry (patient/child); zoomed in = expert-geometry (physicist/regulator). The user zooms to their own geometry.
Live divergence ribbon	1	Real fidelity, color-bound to local GTWED, at any zoom — the user sees where the telling is high/low fidelity, whether zooming lost meaning.
Audio toggle	2	Second modality; transfer measured per modality (read vs audio), using the proven DLO pacing-resonance — not assumed equal to text.
Per-dimension completeness map	3	What's told (checkmark) vs. named-frontier (dashed) for result / falsifier / mechanism / boundary.
Frontier-honesty inline marks	4	Proven vs. named-open shown in the story body; a dishonest telling is flagged REFUSED with negative LVC.

Reduce-motion + VoiceOver honored throughout — the story is for everyone; the ribbon exposes an accessibility value describing divergence.

4 · The striving loop (anti-plateau by construction)¶

StoryStrivingActor ticks over (discovery, geometry, modality) candidates. Each tick:

Ideal → seal the telling, mark its frontier slots raised, emit a witnessed mesh receipt.
Striving → name the lagging axes as open frontier slots. The cell is NOT done; it carries the named drift honestly.
Refused → record the anti-fake refusal (drifted fidelity, or dishonest frontier).

The anti-plateau guarantee is inherited from the orthogonality theorem: a shallow-max telling can never be sealed ideal, so the loop never idle-closes — it always strives a lagging axis or names frontier. Franklin opens more honest frontier slots than it closes for a long time. That is what never-satisfied means.

The standing policy (pol-franklin-v142) authorizes all four axis-actions, so the loop is never silently de-authorized (the bug LVC-011/012 found). Verify suites set their own policy precondition; budget semantics is asserted on the pass, not on which proposal wins the slot.

5 · Mac cell (local) vs. global mesh (witnessed tellings)¶

Mac cell = local sovereign storyteller. Tells the operator's discoveries to the operator's relevant geometries, on-device, raw data never leaves.
Global mesh = library of WITNESSED high-LVC tellings, not private data. An ideal telling is sealed + recomputable and propagates as a receipt (gaiaftcl.franklin.telling.sealed) other cells verify and reuse. Only the receipt (canonical + hash + signature quintet) crosses the membrane — the federation trigger refuses an unsigned broadcast. Raw operator data does not cross. The mesh accumulates higher-LVC tellings over time, none faked, each proven to transfer.

6 · What is CLOSED (hostile-tested, real values)¶

M8FranklinStoryVerify — 22/22 assertions PASS:

Test	Result
F1 orthogonality	fidelity maxed (100) + completeness/honesty fail → NOT ideal. No shallow-max passes.
F2 anti-fake fidelity	fluent-but-drifted telling (fidelity 66 < 70) → REFUSED
F3 anti-fake transfer	simplify-into-drift (transfer 100, fidelity 60) → REFUSED
F4 anti-fake frontier-honesty	claims a named-open summit closed → REFUSED despite fluency
F5 per-geometry transfer	same telling: physicist transfer 100, patient transfer 70 — measured separately, not generic
F6 completeness	result+mechanism only = 50; falsifier + boundary NAMED missing; not ideal
F7 loop anti-plateau	maxed-shallow telling NOT sealed ideal → STRIVING + named completeness slot opened
F8 frontier named	ideal physicist telling raises its slots; patient geometry stays NAMED-open
F9 mesh receipt	ideal telling broadcasts a signed witnessed receipt (unsigned cannot broadcast)

Cumulative across the LVC + Franklin suites: 172/172 (V2 18 · V3 30 · Constitutional 28 · UI Striving 21 · Perpetual 19 · Autonomous Close 20 · Localization 14 · Franklin Story 22).

7 · NAMED FRONTIER (not faked)¶

Untold (discovery, geometry) pairs — every one a row in story_frontier_slots.
The 7 non-English locales (from GFTCL-LION-LVC-012) — real translation provenance required, cannot be faked.
Unmeasured modalities — audio transfer per geometry is measured where exercised; geometries not yet exercised in audio are named-open.
Any geometry profile not yet validated against a real reader — summit.franklin.geometry_coverage stays open per pair, and globally forever.

The strongest true claim¶

Franklin strives on four orthogonal real axes — fidelity, comprehension-transfer, completeness, frontier-honesty — for each reader's origin geometry, across reading and audio, and it cannot plateau by maxing a shallow axis or fake progress on any axis, because the ideal predicate requires all four and every guard is structural. It tells each discovery to the highest honest LVC for each kind of person, names every frontier it has not reached, and the mesh accumulates witnessed tellings any cell can verify. It is never satisfied — and never satisfied means new honest frontier, not churn.

Nothing past the proof.

Appendix · Artifacts¶

Artifact	Path
Four-axis seal (Lean)	`proof/lean/FirstRoars/StoryInvariant.lean`
Story engine (Swift)	`cells/xcode/Sources/LVCEngine/FranklinStoryEngine.swift`
Striving loop + mesh (Swift)	`cells/xcode/Sources/LVCEngine/StoryStrivingActor.swift`
UI instruments (Swift)	`cells/xcode/Sources/LVCEngine/FranklinStorytellerView.swift`
Substrate v142	`cells/xcode/Sources/GaiaFTCLCore/NarratorSchemaV142.swift`
Hostile test	`cells/xcode/Sources/M8FranklinStoryVerify/main.swift` — 22/22
Summit	`summit.franklin.geometry_coverage`
Doctrine	this file

Federation-cosigned

This page's source is sealed in the GaiaFTCL federation manifest — page SHA-256 4955f431dbaad340…, manifest witness a090592e0609adc8…, signed 2026-06-02T18:58:22Z by cell gaiaftcl-mac-cell. Verify with gaiaftcl wiki sign --all and compare wiki-all-signatures.json.