ICNS — declare the numeral before you use it

tokens:   0.9    0.10    0.89    0.91

decimal-real          0.10 = 0.1  <  0.89  <  0.9   <  0.91
version-2-lex          (0,9)      <  (0,10) < (0,89) < (0,91)
version-packed-111     0.91  ->  (0, 9, 1)     not  (0, 91)

so:  succ(0.90) = 0.91        under decimal-10-p2   (resolution 1/100)
     succ(0.9)  = 1.0         under decimal-10-p1   (resolution 1/10)
     succ(x)    = undefined   under decimal-real    (no quantization)

and:  "V0.91"  ->  UNRESOLVED + W_PREFIX_ONLY
      a prefix is a hint, not a declaration

The last line is ICNS’s first proposition: string concatenation cannot make a numeral unambiguous. `V0.91` still admits (0,91), (0,9,1) and 0.91 — unless `V` resolves to a complete, retrievable schema, it carries context, not semantics.

N^  =  < s ; Δ , Π ; Ω >

Δ  =  < D, G, R, P, M, K >     declaration
      carrier, grammar, radix, position map, bounds, length

Π  =  < O, E, N, Q, C, S >     judgment
      order, equality, normalization, resolution, carry, successor

Ω  =  < ρ, ν, h, λ, μ >        envelope
      schema id, version, content hash, scope, provenance

meaning(s)          is not well-formed
meaning(s | Γ)      is                      Γ = (Δ, Π)

The envelope may never change what Δ and Π mean mathematically — it exists for resolution, verification, caching, tracing and cross-system exchange. And the same (ρ, ν) pair must never point at two different content hashes: that is E_SCHEMA_HASH_MISMATCH, refused rather than resolved.

A fourth distinction runs through the paper: the largest single digit, the digit-count limit, the largest representable value at that limit (1 − b⁻ⁿ), and the supremum or limit under unbounded expansion are four different things and must not share one word.

Resolution priority is fixed and silent overwrite is forbidden: inline ▷ anchor ▷ block ▷ sidecar ▷ document ▷ application ▷ registry. Two conflicting declarations at the same priority return E_SCHEMA_CONFLICT rather than letting the implementation pick.

Cross-schema comparison negotiates a common schema Γ* first and converts losslessly into it. If no common schema exists, or a conversion would lose scale or structure, the answer is INCOMPARABLE or a W_CONVERSION_LOSSY warning — not a number.

Six of the eight schemas frozen in 1.0 reach L4 — decimal-10-p1, decimal-10-p2, time-hms, version-2-lex, version-3-lex, version-packed-111 — while decimal-real and identifier-exact stop at L3, because a real number has no successor and an opaque identifier has no order.

Verified by 227 regression tests plus a normative conformance suite (icns-1.0-core, 18 cases, PASS). Within 1.x: no stable feature removed, no built-in schema semantics changed, no transport operation removed, no stable API path dropped without a replacement, and no new required field added to an existing stable request. Deprecation takes at least 2 minor releases and at least 180 days, whichever is later.

Otherwise the state is STABLE / MAINTENANCE. Restart conditions are explicit: a confirmed defect, a real use case, an external report, a publication need, a dependent product, formalization work that requires spec wording changes, or new theory strong enough to justify 2.0.

The one-line principle ICNS wants to outlive itself: before any symbol is executed, compared, sorted, exchanged, or handed to an artificial intelligence, it should first declare which semantic system it belongs to.

Like CAIR, ICNS is currently a research and engineering archive, so nothing here is a download link. Its release states the licensing position outright: 1.0 freezes technical contracts only, the bundled implementation keeps its research-artifact notice, and choosing MIT, Apache-2.0 or dual licensing remains a separate publisher decision. The compatibility promise above is a technical policy — not a commercial SLA, warranty, uptime or response-time commitment.

All related projects