the atlas

Existence-Leak: a ZK proof of feasibility leaks an upper bound on reconstruction difficulty; I(feasibility; method) > 0

first-disclosure cliff is the deep cut; the tail is diminishing returns, no recovery. The convex shape is a chosen teaching model — under an adversary who gains super-additively from correlated systems the tail changes shape.

Three-axis separation is multiplicative

the multiplicativity itself is the structural claim (C7, the V6 falsification frontier); the slider values are illustrative.

The Moving Ceiling: frontier capability growth raises C_S(t) + C_M(t) against fixed archives without raising H(X); R(t) drifts upward and every static reconstruction guarantee has a finite shelf life t*

the drift is coupled to frontier model capability, not to any action of the subject; the exponential is an illustrative growth law (mechanism strongly evidenced, n=2; functional form unparameterized — evidence of mechanism, not of rate). The bound R < 1 holds only while the capacity-deficit condition C_S + C_M < H(X) holds under non-collusion against the stated adversary class — that deficit is exactly what the drift erodes. Design for re-keying, not for permanence.

Compositional Leakage Amplification: policy-only separation compounds toward (2^N − 1)ε with chain depth; amnesia separation breaks the Markov chain and caps at Nε; the gap is exponential-to-linear

the exponent is the structural claim (edge C7 → C83 → C17): without amnesia, every added hop can correlate with every prior one, so the worst case is the full subset lattice 2^N. Amnesia removes the conditioning that lets hops compound, leaving the linear sum.

Trust half-life begins at inscription; decays until renewed

half-life differs by inscription register (C31), is higher for productive than transactional edges (C32), and composes multiplicatively across the three axes (C33). The exponential is the illustrative decay law; the half-life reading is the claim.

Amnesia-enforced separation tighter than policy-enforced

the inequality (amnesia tighter than policy) is the claim — made quantitative at V6 Run 2 and quantified by the compositional gap (C83). The exact curve shapes are illustrative. Same algebra as the completeness inversion, opposite polarity.

Triadic-Constraint Homology (the ARCH-1 bridge, promoted from CM-C47): the three Φ axes and the lattice's Datum·Stratum·Spectrum are one triadic primitive; candidate pair map Protection+Delegation→Σ, Memory+Value→Δ, Connection+Computation→Γ; the gap is β

ARCH-1 promoted from CM-C47; two named predictions (bnot-pairs invert all axes; stratum-3 is the no-dominant-axis seat). The pair map is a candidate, not proven — the correspondence is the conjecture. Lineage: the Gödel/Tarski limitative twin (provability ⊥ verification, diagonal lemma ⊥ ARCH-1 fixed point).

Content-addressed liveness leak: a live content-address is an existence claim about its content; deduplication/GUID liveness leaks existence, and existence bounds the search. The address does not leak content; its liveness leaks existence. Edges → C81, → C92

C93 (~55%, the Limitative Reading, Run 8): a live content-address is an existence claim about its content — dedup/GUID liveness leaks existence, and existence bounds the search (kin to C81). The sha256 is computed live in the browser, so the leak is shown, not asserted: the store never hands over content, it only confirms existence by address. The integrity face — re-derive to verify, tamper-evident — is the benefit the harness holon layer relies on (κ as state, the mesh auditor, did:key/VRC edges; see HOLONS.md, KAPPA_HOLON_INTEROP in agentprivacy-docs); C93 is the privacy COST on the same address, and the reason a published or queryable κ is a disclosure the door governs.