Skip to content

P104 t = −1 laws: charge deficiency, one-unit owned cuts, hub additivity, the per-J odd unit (no κ) - #7557

Open
DreamingOfClouds wants to merge 4 commits into
CrocSwap:mainfrom
DreamingOfClouds:research/p104-t-minus-one-laws
Open

DreamingOfClouds wants to merge 4 commits into
CrocSwap:mainfrom
DreamingOfClouds:research/p104-t-minus-one-laws

Conversation

@DreamingOfClouds

@DreamingOfClouds DreamingOfClouds commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

Retraction / addendum (Nov 13). Law 5 as stated below, and the "neutral period" wording of its premise PA (Lemma A), are retracted.

  • Counterexample (E8/P104) [C]. Take the σ-anchor design and change one thing per column: each Y_t first takes −Z_ts·x_s from a pure X_s at that partner's line, then is snapshotted and parked. That gives D = 7 with 0 neutral columns and 0 owned units, against the literal bound 96.
    • Sources spread over 5 or 9 partners give D = 5 and D = 9.
    • Controls give D = 96.
    • New checker: counterexample.py, standard library only, exact, cross-validated against a full register simulation.
    • Frame legality rests on the same unbuilt L side as the clean anchor design.
  • Corrected premise. A y-atom of column q is a frame-local time difference of Y_q or an owned unit. In the park family, every L-cell counted in Y_q^final arrives after the park snapshot, at frames ⊇ the export frame.
  • Proved [P].
    • For all programs, D ≥ (n − r) − dim Q_V, where Q_V is the y-classes of token-free arrival combinations at V.
    • In the park family, the 48-L-cells-per-column bill holds without neutrality.
  • Open.
    • Non-park removal: tokened public routes, the J gauge, carrier preloads.
    • The empirical law D ≤ ρ_K(sources).
  • Unchanged. Laws 1–4 are unaffected. No κ or price impact.
  • Details: research/p104-t-minus-one-laws/ADDENDUM.md.

Theory package research/p104-t-minus-one-laws/. It covers the t = −1 per-port contract on P104: M = [[1, −1], [1, 0]], X = x − y, Y = x. No κ claim, no word. The σ-rank o, o = rank(q-odd part), Lemma E and the hub bound come from #7555 and are cited, not repeated.

Tags: [P] proved in PROOF.md, [C] exact standard-library check, [cond] conditional.

Results

  1. Charge law [P, C]. In the relay family with charge-neutral owned deposits, settling every cell leaves the V-span deficiency at D = n − r: A3 3, A4 6, E8/P104 96.
    • Corrected anchored law: with early (anchored) cells E and settled cells S, D = dim{(k, k′) : k′ ∈ ker G, (I + Z^S)k + Z^E k′ = 0} − dim ker G. It matches the exact model on 90/90 random sets. The per-cell form dim(ker G ∩ ker Z^E) is wrong on 8/90.
    • σ-anchor schedule: one early cell per column on a perfect partner matching gives D = 0 on E8/P104 (3/3 matchings). The cost moves to an L-side schedule that is still open.
  2. Carried-row law [P, C]. One owned MOVE lowers the deficiency by at most 1, so D ≥ D₀ − #owned MOVEs ≥ D₀ − owned units. It is attained, and the s − 1 control fails.
  3. Hub additivity [P, C].
    • With nested hubs, the per-hub deficiencies telescope to dim K exactly; without forwarding they sum to ≥ dim K.
    • Checked on all 4,683 A3 hub sequences, 4,000 at A4 and all 510 E8/P104 hyperplane hubs (D₁ + D_V = 96 on 510/510).
    • Counting only the last hub gives as little as 36.
  4. Per-J odd unit [C; gauge ledger cond].
    • H = [[1, −q], [q⁻¹, 0]] has o(Hⁿ) = 2, 2, 0, 2, 2, 0. SW = J·L_q has o = 1, so the t = −1 native costs 17 more odd units per port per word.
    • The three-register lift is q-free only in a gauge that itself has o = 1.
    • σ-fixed automorphisms preserve o.
    • A gauge-shifted contract totals ≥ 20 odd units per port per word for every m, against 3 for the unshifted design.
  5. Y-atom conservation — RETRACTED as stated, see the addendum above. D ≥ dim{k ∈ ker G : k_N = 0} ≥ n − r − N_neutral − N_owned, which is 96 per call on E8/P104 with N = ∅. The inequality holds only under the corrected premise PA_gen.
    • Checked: the counting bound holds over all column sets at A3/A4 and is tight at every |N|.

Caveats

  • Laws 1, 3 and 5 are stated for the relay family.
  • The "neutral period" wording of Lemma A is false (addendum). Law 5 in its original form is withdrawn. The corrected premise PA_gen is proved in the park family and open outside it, and the below-re-add-frame case of law 3 is conditional on it.
  • The gauge ledger assumes boundary σ-odd units cost at least one odd op each.
  • Not packaged here:
    • the finite-order statement about the caller (fresh planes at every J);
    • the 20-odd-op caller floor with its exact callers.
  • The E8/P104 Gram is computed from the pinned PR968 decoder file. No data is vendored.
  • No human has reviewed the package.

Validator: python3.12 -B scripts/research_package.py --package research/p104-t-minus-one-laws → PASS_REGISTERED_RESEARCH_PACKAGE. The branch carries the two #7257 CI-prerequisite commits by Clear1nsight, cherry-picked with authorship kept.

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

🤖 Generated with Claude Code

Clear1nsight and others added 3 commits November 13, 2026 18:19
This is the exact documentation prerequisite retained by CrocSwap#7156. It points the onboarding check to the selected CrocSwap#7000/CrocSwap#7001 review. Credit to Rohan Arun and the source authors.
The PR CrocSwap#6292 rejection census and package plan are statements about 2485805. Load its exact sources and planner in a private sparse checkout, keeping all source hashes, 19 expected rejections, nine package selections and the empty-plan control. Subsequent edits to the live Makefile and CI guide no longer invalidate this historical check.

The original audit is by unpaidmoon, building on Dennis’s historical replay gate. The pinned repository view follows the precedent of Matarener’s PR CrocSwap#5434. Repair prepared with substantial OpenAI Codex assistance.
…itivity, the per-J odd unit (no kappa)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@DreamingOfClouds

Copy link
Copy Markdown
Contributor Author

lemma A is the enemy so like if you can break it please do

…example checker

Lemma A as worded ("a y-atom leaves Y_q only during Y_q's neutral period or
through an owned copy") is false on E8/P104: a sigma-anchor design whose every
Y_t first takes -Z_ts x_s from a pure X_s, then is snapshotted and parked, has
0 neutral columns and 0 owned units yet D = 7 (literal bound 96).
counterexample.py reproduces this exactly (standard library, cross-validated
against a full register simulation, with controls). ADDENDUM.md states the
retraction, the corrected premise, what is proved (D >= (n - r) - dim Q_V for
all programs; Lemma P in the park family) and what stays open. Laws 1-4 are
unaffected; no kappa claim.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@DreamingOfClouds

Copy link
Copy Markdown
Contributor Author

Retraction (pushed in 31842e1). Law 5 as stated is withdrawn, together with the "neutral period" wording of its premise (Lemma A).

The wording was "a y-atom of column q leaves Y_q only during Y_q's neutral period or through an owned copy". It is false on E8/P104.

  • Construction. Take the σ-anchor design and change one thing per column t: Y_t first takes −Z_ts·x_s from a pure X_s at a partner's line, and is then snapshotted and parked.
  • Result. That gives D = 7 with 0 neutral columns and 0 owned units, against the literal bound 96. Contamination is charged per distinct source, not per column.
  • Reproduce. python3.12 -I -B research/p104-t-minus-one-laws/counterexample.py (standard library, exact, about 11 s, also in VERIFY.json). It is cross-validated against a full register simulation, and its controls give D = 96. Frame legality rests on the same unbuilt L side as the clean anchor design.

ADDENDUM.md covers the rest:

  • the corrected premise;
  • what is proved: D ≥ (n − r) − dim Q_V for all programs, and Lemma P, which keeps the 48-L-cells-per-column bill without neutrality in the park family;
  • what stays open: non-park removal, and the empirical law D ≤ ρ_K(sources).

Laws 1–4 are unaffected. There is no κ or price impact. The validator passes.

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

@DreamingOfClouds

Copy link
Copy Markdown
Contributor Author

Follow-up to the addendum: the "one unit per source" bound is withdrawn, replaced by an exact rank formula

ADDENDUM.md lists D ≤ ρ_K(sources) (one owned unit per distinct contamination source) as empirical. It is false as an upper bound. The correct law is the following.

Theorem R (σ-anchor family) [proved].

D = rank(C′K) + res, where res = dim{u ∈ range G : (C′ − M0)ᵀu = 0} and M0 = Z_E⁻¹·G/2 − I.

  • C′ is the matrix of early (pre-snapshot) cell charges.
  • K is the kernel of G.

Three independent exact computations agree on 85 random A3/A4/D4 designs and on every counterexample below: a full register simulation with Y exact, the witness system, and the formula. On the 92 designs behind the original ρ_K observation it matches D in 92/92 with res = 0. The cases there with D < ρ_K are exactly those with rank(C′K) < ρ_K.

Counterexamples to D ≤ ρ_K, certified exactly on A4:

  • Singular clean anchors. 240 of the 30,240 matchings give D = 1 with no contamination, where the bound says D = 0.
  • Resonance. In the same frame-legal shape as the fixture (one source per column, s(t) ≠ σ(t)), 8 of the 256 orbits give D = ρ_K + 1 (exhaustive orbit search).
  • Not yet fully certified: on A4, 27 plane candidates were left unresolved. On D4, 124 of 150 random matchings are certified clean and 26 hit the time cap.

Consequences:

  • The only statement that survives unconditionally is the lower bound D ≥ rank(C′K).
  • For partition masks (early cells = the partner cells between the parts of a port partition, anchor source cell included), rank(C′K) = Σ_parts dim span(q_P) − r. This is checked exactly on random partitions in all four geometries, and on E8 two parts give 2,472 early cells at rank 8. The owned cost equals this rank only when res = 0; in general add res.
  • The counterexample.py fixture stays tight. For single-source contamination D ≥ ρ_K(S). The smallest E8 partner cover has 5 ports (exhaustive), so D ≥ 5 there, and the fixture's D = 5 (res = 0) attains it.

Also proved, for any legal program: every y-atom is an owned unit or a frame-local time difference of Y_q. Token-carrying public routes, the owned gauge J and carrier pre-loads therefore supply no further atom source.

Still open: the column-priced corrected premise (PA_gen) outside the σ-anchor family. That is, pricing deliveries into Y_t that ride inside the exported atom rather than arriving after the snapshot. Law 5 remains conditional on it.

No κ change.

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

@DreamingOfClouds

Copy link
Copy Markdown
Contributor Author

Second follow-up: a column-priced bound that holds for any program (Theorem M), plus a second geometry where the per-source bound fails

Theorem M [proved, any legal program].

D ≥ rank_K([v_x + v_y] over the arrivals at V) − (number of owned rows landing at V).

Here the bracket is the merged class of each arrival's x-part and y-part, and the rank is taken over K = ker G.

  • The proof is witness counting and nothing else. It uses no relay structure, no frames and no parks, so it covers token-carrying public routes, the owned gauge J and carrier pre-loads.
  • Checked on 400 random words that mix parks, time-difference restores, σ symbols and owned rows landing at V: 0 violations. The negative control, which uses all of Qⁿ instead of K, fails on 337 of the 400.
  • In the σ-anchor family the bound equals rank(C′K), so it recovers Theorem R's lower bound. It is tight at 5 on the counterexample.py fixture.

What it says about contamination. The blanket statement "contamination always costs owned units" is false:

  • A co-read strip can cancel the contamination before V at the cost of one extra read, and reach D = 0.
  • The correct price is the one in Theorem M: contamination that reaches V costs its merged K-rank, and contamination cancelled before V costs nothing in this rank (it costs reads instead).
  • The y-class form of the premise (PA_gen) cannot carry this price, because the class a snapshot restores ignores everything delivered before the snapshot.

Per-source upper bound, now certified on two geometries (single-source family):

  • A4, exhaustive. Two independent searches find the same 10 resonant orbits out of 256. In 2 of them the clean anchors are singular. The other 8 have clean D = 0 and D = ρ_K + 1.
  • D4. All 150 sampled matchings are now decided. Matching Reclaimed complex supplier: conditional kappa 8.0325171e-5 #112 has clean D = 0, one source per column, and D = 6 = ρ_K + 1.
  • A3. No resonance, exhaustive.
  • The multi-source family has certified counterexamples only; it has not been searched exhaustively.

Status of Law 5. It is now best read through Theorem M, which is proved for any program. The y-class premise is superseded rather than open. No κ change.

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

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.

2 participants