Framework

Loss Landscape Vocabulary Framework

v13 · April 2026 · Atlas Heritage Systems · Working document — not a finished product

A note before the math

You don't need to understand any of this to read the Framework. But if you want to know why the Framework is built the way it is, the math is where the answer lives.

When a language model trains, it moves through a mathematical landscape — hills, valleys, flat plains — searching for the lowest point. The vocabulary on this page describes the features of that terrain: what makes one region harder to cross than another, what gets preserved in the difficult parts, and what gets smoothed away in the easy ones.

The archaeological claim Atlas makes is simple: the hard parts leave marks. Those marks are readable. That's what the instruments are built to find.

Start with the plain language description of each term. Follow the math when you need it.

How it all fits together

The Framework names the terrain. The instruments measure behavior on it. The schema defines how measurements get recorded. The protocols govern how they're taken — CISP is the governance layer that sits above every active instrument run, enforcing isolation, sequencing, and the human-judgment boundary.

Below the protocols, the automation layer handles transcription: parsing raw model output, computing what can be computed, and leaving blank what requires a Technician's call. Below that is the data the instruments produce over time — the actual record Atlas is building.

The geometry sits at the end of the chain. PyHessian doesn't measure behavior; it measures the mathematical terrain the Framework describes. When there's enough data, the Hessian eigenvalue analysis will either confirm the Framework's terrain claims or force a revision. Working hypotheses stay hypotheses until the math has something to argue with.

Terrain Topology Global Geometry Navigator Flow & Resistance Structural Integrity Ablation Potential & Tension Architecture Summary Glossary

Potential & Tension — Revision Note

A post-adversarial-review revision to the framework's primary generative vocabulary. Resistance was identified as the wrong primary frame — it describes opposition to movement already occurring. Potential difference is upstream of movement. This section documents the revision without replacing the existing flow & resistance vocabulary.

Potential Difference (charged / discharged)

The gap between the navigator's current state and the terrain's local geometry that creates the condition for movement. In electromagnetism, current does not flow because of resistance — it flows because of potential difference. In the loss landscape: potential difference is the primary generative quantity. The gradient ∇L(θ) is its most direct expression.

ΔV(θ) = L(θ_terrain) − L(θ_navigator)

Tension (taut / slack)

The structural condition that holds potential difference stable without collapsing it. Slack tension means the navigator has decoupled from the terrain — not equilibrium, not stability, but the condition for slow invisible drift toward whichever end of the local topology has the lowest elevation. Slack in a valley is the precondition for undetected centerward drift.

Harmonics (resonant / dissonant)

The dynamic behavior that emerges when potential difference and tension interact over time. Further resolved via Skywork reading of Friston's active inference: free energy minimization produces oscillatory dynamics around posterior modes via the bidirectional prediction-error loop — an intrinsic oscillatory mechanism that is not regularization. Whether this bridges to loss landscape frequency-matching is the open question.

ω_landscape = eigenvalues of H(θ) ω_training = f(η, batch_size, curriculum) resonance condition: ω_training ≈ ω_landscape

Smith (2018) superconvergence; Smith (2015) cyclical learning rates

Revised Archaeological Claim

High-perplexity regions in a deployed model are not merely high-resistance territory. They are places of unresolved potential difference — where the gap between navigator state and terrain geometry never closed, where the system stayed charged without discharging. Via Song et al. (2024): weights encode predictions not inputs. High-perplexity regions are places where prediction error accumulated faster than inference could resolve it before the weight update discharged it. The retained potential difference framing now has a biological learning mechanism behind it.

Revision Status

Open problem — restoring force (partially resolved): Harmonics requires a restoring force. Curvature eigenvalues describe geometry but do not generate restoring force. Millidge (2023) provides volumetric basin dominance as partial resolution. Active inference provides intrinsic oscillatory mechanism. Whether either bridges to formal frequency-matching remains open. If the restoring force is always regularization — already in the toolkit — then harmonics is a redescription of known phenomena, not a new formal quantity. The tautology charge holds until a restoring force intrinsic to landscape geometry is identified that regularization doesn't already cover.