Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions research/bit-all-bundle-price-floor/.gitattributes
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
retained/** -text whitespace=cr-at-eol
35 changes: 35 additions & 0 deletions research/bit-all-bundle-price-floor/CREDITS.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,35 @@
# Credit and exact sources

Prepared by Ali Romman with substantial OpenAI Codex assistance in research,
implementation, verification and writing. New work is Apache-2.0.

- Nick Petalas supplied the full-donor-span reduction, nine exceptions,
two-delivery floor and higher-arity question in
[PR #7739](https://github.com/CrocSwap/integer-mult-bounds/pull/7739).
Its [proof and contract](https://github.com/petalas/integer-mult-bounds/blob/7e44f44b8078b3788256772533890fbf85503a0c/research/bit-causal-multi-delivery/README.md)
and [checker](https://github.com/petalas/integer-mult-bounds/blob/7e44f44b8078b3788256772533890fbf85503a0c/research/bit-causal-multi-delivery/verify.py)
are retained unchanged. The new verifier calls its rational geometry,
packed row interpreter and complete orientation checks.
- Ali Romman's [PR #7723](https://github.com/CrocSwap/integer-mult-bounds/pull/7723)
supplied the fixed-library model, single-delivery floor and multiple-delivery
question. Its [proof](https://github.com/aliromman/integer-mult-bounds/blob/a4fd22c08058c0fef830306b2d46e4a51f091939/research/bit-causal-delivery-rank-obstruction/PROOF.md)
and credit file are retained unchanged.
- Jerry Chen supplied the selected paired BIT caller in
[PR #7001](https://github.com/CrocSwap/integer-mult-bounds/pull/7001).
The [native word](https://github.com/CrocSwap/integer-mult-bounds/blob/e54523517d0f03e58352e647c403a5af3e525a57/research/paired-query-only-bit/retained/inputs/BIT-NATIVE-WORD.json.gz)
and [caller notice](https://github.com/CrocSwap/integer-mult-bounds/blob/e54523517d0f03e58352e647c403a5af3e525a57/research/paired-query-only-bit/NOTICE)
are pinned at canonical main. The retained notices credit zxv and earlier
native authors for the word and admission framework, DreamingOfClouds for
the chain-root caller, and Ali Romman and Cinna Davis for the root lineage.
Both source notices and the source license are retained unchanged.

The new work is the fixed-quotient support certificate, full coset distances,
all-arity conclusion and coordinate-set reconstruction. The predecessor's
5,733 exclusions and pair census are replayed dependencies, not new claims.
The geometry and part of the coverage logic are shared. A separate row
representation is not independent human or formal-kernel acceptance.

SOURCE-PINS.json binds eight retained files. SOURCES.json binds only the
base catalogue and selected record as provenance. No unmerged source ancestry
is asserted. The retained directory structure lets the unchanged checker
resolve its own native input; no path global or frozen byte is patched.
171 changes: 171 additions & 0 deletions research/bit-all-bundle-price-floor/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,171 @@
# Every fixed-library direct bundle retains the original BIT price

For the selected PR7001 native word, an arbitrary finite bundle of direct
nested deliveries cannot reduce the price of any original paid arrival.
The claim holds in both orientations, for every tariff `rank^t` with
`0<t<=1`. The original single delivery attains the floor. This is a
scoped finite-word theorem, supported by exact replay. It is not a
universal obstruction to changed BIT networks and gives no new kappa.

## 1. Fixed contract

Keep all 659 public inputs independent. For each original MOVE, keep its
destination chart, event deadline and required binary row. At a chart `b`,
grant the span `V_b` of its entrance rows and its original paid arrivals
strictly before that deadline. Grant retention and scalar XOR formation
for free. A direct delivery sends one row of `V_b` to the destination `d`
and pays `s_b^t`, where `s_b` is the positive dimension difference. The
two chart subspaces must be nested over the rationals. Grant `V_d` for free.

The source library is complete for this model: INIT creates zero, a
cocharted ADD stays in the span, SWAP carries a row with its chart, CLEAR
removes a row, and MOVE adds a row to one chart at later deadlines. This
is the fixed-library contract of Ali Romman's PR7723. Nick Petalas's
PR7739 permits a sum of direct deliveries and proves the one- and two-donor
floor. The new result closes its nine remaining higher-arity cases.

Repeated deliveries from one chart can be combined into one row there.
The combined row has the same delivery rank and no greater price. Thus
any feasible finite bundle reduces to distinct nonempty donors. The proof
below even grants repeated donors when it bounds support, so this reduction
is not needed for that upper bound.

## 2. Strict saving requires smaller total rank

If the original arrival rank is `r`, and a bundle has positive ranks
`s_1,...,s_k` with sum `R`, then

```text
sum_i s_i^t = R^t sum_i (s_i/R)^t >= R^t, 0<t<=1.
```

Each fraction lies in `[0,1]`, so its t-th power is at least itself.
Thus a strictly cheaper bundle must have `R<r`. Every donor in such a
bundle is individually lower rank. A target outside the sum of all those
donor libraries and `V_d` cannot be supplied by any cheaper bundle.

The unchanged PR7739 checker reconstructs all original libraries and
checks this full sum at all 5,742 arrivals. It excludes 5,733. Its nine
exceptions are forward arrivals into chart 0, from charts 510 through
518, at events `26861+8i`, `0<=i<9`. Each has old rank seven. The complete
donor sum contains these targets, but that sum ignores total rank.

We rerun the predecessor's exact source and all its pair checks. We do not
claim its 5,733 exclusions or its pair enumeration as new results.

## 3. Keep one quotient for all nine deadlines

Let `V_0` be the chart-0 library strictly before event 26861. Its dimension
is 524. Ordered binary elimination defines a linear projection `Q` from
the 659 independent-input coordinates to a systematic 135-dimensional
quotient. It clears each pivot in descending order. Its kernel is exactly
`V_0`: it kills that basis, leaves every nonpivot unit coordinate free,
and its rank is 135.

Use this same projection at all nine deadlines. A quotient rebuilt after
each arrival can turn one-coordinate donor rows into dense representatives;
that would lose the support bound needed here. This proof instead leaves
the later free destination rows explicit.

Write `g_i` for the projected target at event `26861+8i`. At deadline `i`,
the free destination library in this fixed quotient is

```text
F_i = span(g_0,...,g_(i-1)).
```

Exact replay checks this equality, including dimension `i`, and checks
that the current target is not inserted early. Each of the 514 eligible
lower-rank donor libraries is unchanged over the nine deadlines. The
checker compares all eight later snapshots with the first, for 4,112
library equality checks. Other intervening arrivals go to charts outside
this eligible set. The proof uses the checked equality, not that observation
as an unverified shortcut.

## 4. Every cheap bundle has support at most 30

For each eligible chart `b`, take the union of the supports of a basis
of `Q(V_b)`. Every row in that space has support within this union.
Cancellation cannot introduce a new coordinate. The exact inventory is
in [RESULT.json](RESULT.json), including every chart, rank, image dimension,
support mask and support size.

| Delivery rank | Eligible charts | Largest support union |
|---|---:|---:|
| 1 | 111 | 1 |
| 2 | 153 | 2 |
| 4 | 70 | 17 |
| 6 | 180 | 30 |

These are all nonempty eligible charts. No other positive rank below seven
occurs. A sum of delivered rows has Hamming weight at most the sum of their
support-union sizes. With total rank at most six:

- A rank-six donor uses the whole budget and has support at most 30.
- A rank-four donor leaves rank budget at most two. The remaining donors
have support at most two, so the total is at most 19.
- Without a rank-four or rank-six donor, support is at most total rank,
hence at most six.

Thus 30 is a valid upper bound, without an arity limit. A dynamic-program
relaxation that permits unlimited donors checks the same statement. Its
bounds at budgets zero through six are `0,1,2,3,17,18,30`. The value 30
is not claimed as the exact attainable weight or price for the target.

## 5. The target cosets stay at distance at least 48

A bundle reconstructs `g_i` with free destination XOR only if its sum is
in the coset `g_i+F_i`. Enumerate all `2^i` members of this coset. This is
only 511 exact vectors across the nine deadlines.

| Event | Free dimension | Coset vectors | Minimum Hamming weight |
|---|---:|---:|---:|
| 26861 | 0 | 1 | 56 |
| 26869 | 1 | 2 | 58 |
| 26877 | 2 | 4 | 52 |
| 26885 | 3 | 8 | 50 |
| 26893 | 4 | 16 | 51 |
| 26901 | 5 | 32 | 50 |
| 26909 | 6 | 64 | 50 |
| 26917 | 7 | 128 | 48 |
| 26925 | 8 | 256 | 48 |

The receipt gives each projected target, the full weight histogram and a
selector attaining the reported minimum. Every minimum exceeds 30.
Consequently, no bundle of total rank at most six can supply any exception.
This is an exact finite certificate, not a random search or a bound on
one, two or three donors only.

Together with Section 2 and the 5,733 inherited exclusions, this proves
the price floor at all 5,742 arrivals. The separate set-based interpreter
also checks that the original row belongs to its original source library
and that the original pair is nested, at every deadline in both orientations.
Thus the old rank tariff is attained, not merely a lower bound for an
empty feasible set.

## 6. Consequence, evidence and limits

With productive targets and all other caller charges fixed, replacing
original arrivals by these bundles cannot improve the paid moment for
`0<t<=1`. The compatible signed-margin gain is at most zero, attained
by the original deliveries. Additional nonnegative formation, storage or
restoration charges preserve the exclusion. This closes the arbitrary-arity
question for the stated fixed libraries; another subset-size census cannot
improve the conclusion.

The new interpreter represents rows as coordinate sets. It agrees with
the retained packed-integer interpreter on every paid row in both directions.
The rational nesting computation and full-donor reduction are called from
the unchanged predecessor; they are shared code, not independent checks.
All eight retained files have exact byte pins. Four reduced support-cap
mutants fail. Inserting each current target into its own free library makes
the separation check fail. Strict-budget and binary-sum controls also run.

The result does not permit new charts, intermediate relays, adaptive changes
to the libraries, helper promises, changed target rows or deadlines, or
matrix-valued reconstruction. A relay can change the available information
and needs a new charged contract. No physical construction for free storage
or formation is supplied. The selected cold replay, other predecessor full
suites, Lean and Docker were not run. No independent human or formal-kernel
acceptance is claimed. All five framework assumptions remain, and no new
kappa or selected-frontier change follows.
74 changes: 74 additions & 0 deletions research/bit-all-bundle-price-floor/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,74 @@
# An all-bundle price floor for the selected BIT word

Every original paid arrival retains its minimum price under arbitrary finite
bundles of direct nested deliveries from the original strictly earlier chart
libraries. The result covers both orientations, all independent inputs, free
retention and destination XOR, and every rank tariff s^t with 0<t<=1.
The original single delivery attains the floor. No new kappa follows.

This closes the nine higher-arity exceptions left by Nick Petalas's PR7739.
Its full-donor-span reduction already excludes the other 5,733 arrivals.
In one fixed 135-dimensional quotient, any bundle of total rank at most six
has support at most 30. Each of the nine target cosets has minimum Hamming
weight at least 48. Their complete enumeration has only 511 vectors.

Read [PROOF.md](PROOF.md), [RESULT.json](RESULT.json), [verify.py](verify.py),
[SOURCE-PINS.json](SOURCE-PINS.json) and [CREDITS.md](CREDITS.md). The receipt
lists all 514 eligible donor supports, all nine target coset histograms and
minimum witnesses. The proof explains why this excludes every arity.

## Evidence and reproduction

Use Python 3.12 and its standard library, from the repository root:

```sh
python3.12 -B scripts/research_package.py --package research/bit-all-bundle-price-floor
```

Both registered normal and optimized commands must print:

```text
PASS_BIT_ALL_BUNDLE_PRICE_FLOOR 5742 arrivals; 511 cosets; distance>=48>30
```

The unchanged predecessor reruns its rational nesting, full-span exclusions
and 4,626 singleton plus 478,764 pair checks. A new coordinate-set interpreter
agrees with its packed interpreter on all 5,742 paid rows and checks every
original-source attainment. The new certificate checks 4,112 later donor
library equalities, all 511 coset vectors, support caps and the strict budget.
Controls reject four false caps and nine arrivals used at their own deadline.
The declared source geometry remains shared; this is not blind human review.

The runner authenticates base provenance through Git. The direct verifier
runs from an exported package without Git or network access and checks all
eight retained source hashes. No source globals or frozen code are patched.
The selected cold replay, other predecessor full suites, Lean, Docker and
physical delivery or storage were not run. All five framework assumptions
remain. The selected frontier is unchanged.

## Selection and boundary

Base e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8. The selected
bindings pass; the catalogue assessment is older. TASK.json records the
changed bit-helper-contract scope. The parent handoff explicitly leaves
these nine cases open. Exact-source open/closed intake, current PR bodies
and live handoffs found no matching claim in their coverage. Sparse notes
and unpublished work remain coverage limits.

The contract fixes libraries, independent inputs, destinations, target rows
and deadlines. It permits any number of direct donors, but no new chart,
relay, helper relation or adaptive library. With productive targets and all
other caller terms fixed, its best compatible signed-margin gain is zero.

## Open next questions

Can one paid relay change the causal library enough to permit a complete
saving? Specify its source, intermediate chart, arrival deadline, formation,
retention and restoration first. Then compare the full added and removed
power tariff. The present proof supplies no relay price or physical program.
The arbitrary-arity search in the fixed original libraries is complete;
another donor-count census is not a useful continuation of this model.

Ali Romman, with substantial OpenAI Codex assistance. Apache-2.0. Nick Petalas's
reduction and checker, Jerry Chen's selected caller, and all native and caller
predecessors retain their credit and unchanged notices in CREDITS.md.
Loading
Loading