Skip to content

P104: a class junk floor |J| ≥ 4166 at owned mass 56 and κ ≤ .0122445 for the C = 56 flag class (Lean + certificate, conditional, no κ claim) - #7689

Open
DreamingOfClouds wants to merge 1 commit into
CrocSwap:mainfrom
DreamingOfClouds:research/p104-c56-class-ceiling
Open

DreamingOfClouds wants to merge 1 commit into
CrocSwap:mainfrom
DreamingOfClouds:research/p104-c56-class-ceiling

Conversation

@DreamingOfClouds

Copy link
Copy Markdown
Contributor

Lean plus an external exact certificate. No κ claim.

Result. In F1's C = 56 flag model the minimum junk count |J| lies in [4166, 4216]. Lower end: a Lagrangian dual certificate (frozen multipliers, exact per-chain flag DP over Q, all 8 configurations; two independent checkers agree). Upper end: a witness of 4216 cells (two configurations; odd cells and diagonal constraints met, rank 55/56). With hJ : 4166 ≤ |J|, the ceiling functional of ceiling_native gives F(1 − t) < 0 on [.0122445, 1] (CongestionNative.ceiling_c56, kernel-only), so κ ≤ .0122445 for any word satisfying the named hypotheses. .0122444 fails at the point, so .0122445 is the tightest certified 7-digit value of this piece set. For comparison the old class bound |J| ≥ 1936 gave .0136564.

What this is not. The frontier words are NOT in this class. They have rank 11 and multi-dimensional lifetimes (their junk count, 4962, is computed, not a point of this model); they are covered by #7668 and #7678. This package is a structural statement about what C = 56 flag decoders must pay, not a bound on any current word.

Scope hypotheses (all in PROVED.md; none proved in Lean).

  • The certificate is computer-certified outside Lean. Lean takes hJ : 4166 ≤ J.card as a hypothesis; the DP is not formalised.
  • Rank-8 Y-plane rows with C = 56 in one of F1's 8 configurations; the rigidity of the row factorisation is numerical and quoted.
  • Monotone frame chains with any step sizes (multi-dimensional steps are dominated by a refinement; proved on paper in the audit, so not a hypothesis).
  • No split gathers and no split reads. One piece per gather and per read, per chain. This is a real hypothesis, not a technicality: with split gathers a non-forced chain contributes an unbounded prefix sum and the bound for multi-chain cells fails; only the single-chain 1936 survives splits, and no cost rule excludes splits.
  • Plane separation (no second owned plane carries plane-1 X content).
  • Not covered: decoders with 9 or more rows, network decoders, owned mass above 56, public-layer cancellation of arrival junk. The ceiling also inherits the premises of ceiling_native (hcong, hdist, hspMono, hcstMono, the spF / cst brackets).
  • The upper end 4216 is an exhibited witness that was re-verified exactly; its data is not in this package, so only the lower end is reproducible from the files here. The true minimum is not claimed.

Contents. 10 new Lean modules (CongestionC56*) on top of 36 unchanged ones (46 in the closure, rebuilt fresh, CHECK.txt; axioms propext, Classical.choice, Quot.sound; no sorry, no native_decide). certificate/: frozen multipliers with SHA-256, geometry data, two checkers (jr_lagx.py, aud_cert.py), their logs. The multiplier search and table-generation scripts are not included.

DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code.

🤖 Generated with Claude Code

…2445 for the C = 56 flag class (Lean + certificate, conditional, no kappa claim)

ceiling_c56 (kernel-only) from hJ : 4166 <= |J|; the junk floor is a Lagrangian-dual certificate checked by two independent exact checkers (frozen multipliers, SHA-256). The frontier words are not in this class.
DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant