Skip to content
Draft
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
97 changes: 97 additions & 0 deletions research/shared-port-raw-cut-invoice/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,97 @@
# The raw shared-port cut does not pay for its own cleanup

This is a Gaussian invoice for one changed child interface. It is not a new
decoder, a kappa, or a re-proof of the recording-only image bound.

## Source

Use `research/block3-shared-port-composition/WORD.json` and the operator
definitions in `inputs/pr4813-verify_schedule.py`. Products act on a column
from right to left. `A = C0`, `K = C0 C1 C2`, and the source identities

B = M C2 + 2 K a, D = Z1 V C2 W A

are the ones asserted by `load_operators`. Grades are `log2` of the uniform
support size. Every pure letter in either five-port schedule has grade 0, 1
or 3. The fused schedule's paid profile is two grade-1 calls and four grade-3
calls. The raw and fused schedules have the same final group-algebra state.

The predecessor moment charges an `Hr` call as `r^t` for real `t` in `[0, 1]`
and leaves monomials, field additions and the source's exact factor `1/2`
uncharged. That is the same exponent as `1-C` in the completion note's
`3^{1-C}` comparison. A positive gap means reward minus paid is positive.
Selected `κ` is not an input to this comparison.

## The exposed state

Index gates from zero. After raw gate 11, owner 1 is exactly

u = B x0 + D x1 = (M C2) x0 + 2 (K a) x0 + D x1,

and no other live owner depends on `x1`. The same glue, a grade-3 term on
`x0` together with a grade-2 term on `x1`, already holds after gates 9 and 10.
`D`, `K a` and `M C2` are invertible. Therefore, for every 8-by-8 matrix `L`,
the `x1` column of `L u` is `L D`, which vanishes if and only if `L = 0`. Any
left-linear image of the five live owners with zero `x1` column must put
`L = 0` on owner 1, and then that image does not contain the summand
`2 (K a) x0`.

Isolation needs a new carrier of `D x1`. Saved original inputs are the free
bypass already allowed by the completion argument. `H`, the source's monomial
multiple of `1/2`, sends `2 K a` to `K` and does not kill `D` or `M C2`.

## What it costs to continue and to clean

Raw gate 12 is one grade-1 `A`. The state after that gate equals the fused
state after `N`. Continuing with the remaining fused gates reproduces the
fused endpoint. Dropping gate 12 does not. The resulting hybrid pays

5 H1 + 3 H3.

That is the exposure-and-continuation bill: the fused `2 H1 + 4 H3`, plus
three rank-1 calls, minus the one rank-3 call that the raw prefix replaces.

`D = Z1 V C2 W A` uses exactly two grade-1 letters. No schedule letter has
grade 2, and a monomial preserves grade, so `D` is not one source paid letter.
A fresh copy on a saved `x1` therefore costs two `H1`. `M C2` is one `H1` on
a saved `x0`, followed by the free monomial `M`. One field subtraction for
each contaminant, then `H`, leaves the pure grade-3 unit `K x0`.

The full Gaussian bill is

8 H1 + 3 H3.

## Moment

Credit five rank-3 jobs: the four already available from the final image, plus
this new `K`. That credit is generous. It does not charge cleanup of the
off-diagonal grade-3 terms already present in the final rows.

gap(t) = 5 * 3^t - 8 - 3 * 3^t = 2 * 3^t - 8.

The derivative `2 ln(3) 3^t` is positive, so the maximum on `[0, 1]` is
`gap(1) = -2`. The gap is strictly negative on the whole interval.

If a new grade-2 letter equal to `D` is added to the alphabet, it replaces
only the two cleanup letters inside `D`. The bill becomes `6 H1 + 1 H2 + 3 H3`
and

f(t) = 2 * 3^t - 2^t - 6.

Now `f'(t) > 0` if and only if `(3/2)^t > ln 2 / (2 ln 3)`. The left side is
at least 1 for `t >= 0`. The right side is strictly less than 1 because
`ln 2 < ln 9`, that is `2 < 9`. Thus `f` is increasing and
`f(1) = -2 < 0`. Widening the alphabet by this one letter does not create a
positive gap.

Leaving both contaminants uncharged, and still crediting the fifth rank-3 job,
gives gap `2 * 3^t - 5`, which equals `+1` at `t = 1`. The three cleanup
rank-1 calls are what moves that endpoint from `+1` to `-2`.

## What this does not decide

Nonlinear maps, feedback into a live owner, precision, compiler reserve and
outer transfer are outside the invoice. Each of those, if charged, adds a
nonnegative cost on a gap that is already negative. A different cut, or a
consumer that accepts the glued sum `B x0 + D x1` as one legal child, is a
different contract.
91 changes: 91 additions & 0 deletions research/shared-port-raw-cut-invoice/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,91 @@
# The first raw shared-port cut loses its Gaussian moment

**Negative result. No new κ.** The selected conditional exponent stays
`605457647/200000000000`.

The recording-only theorem for the fused five-bank producer leaves one escape:
the raw state after gate 11, whose grade-3 channel has image dimension `5d`.
This package prices that state as a child that must be exposed, continued and
cleaned. On the predecessor moment, for every exponent `t` in `[0, 1]`,

```text
gap(t) = 2 * 3^t - 8 <= -2.
```

Replacing the two cleanup letters inside `D` by one new grade-2 letter still
gives `2 * 3^t - 2^t - 6 <= -2`.

## Scope and evidence

Original card: `complex-rank-shape-decoder`. Scope status: **refined**. The
card asks for a larger-rank decoder profile. The open request actually taken
is the changed-resource invoice after the fixed-producer completion ceiling
([discussion comment 19133451](https://github.com/CrocSwap/integer-mult-bounds/discussions/1166#discussioncomment-19133451)).
Open PR [#5491](https://github.com/CrocSwap/integer-mult-bounds/pull/5491)
already shows that this raw cut raises the grade-3 image dimension from `4d`
to `5d`, and it leaves the cleanup bill open. This package does not replay
that image proof. It prices the bill.

The exposed owner is `B x0 + D x1`, with `B = M C2 + 2 K a`. No left-linear
image of the live owners can drop `D x1` and keep `2 K a x0`. A fresh `D`
costs two source rank-1 calls, and `M C2` costs one more. Keeping the later
fused banks costs the hybrid profile `5 H1 + 3 H3` before that cleanup. The
proof, the derivative comparison, and the controls are in [PROOF.md](PROOF.md).

The moment does not charge monomials, field additions, the factor `1/2`, or a
field copy of the exposed owner. Charging any of them widens the deficit.
Precision, compiler reserve, outer transfer and feedback into a live owner are
not decided here. They are further nonnegative charges on a gap that is already
negative.

Source pins are in [SOURCES.json](SOURCES.json). The two live inputs are the
predecessor gate list and operator file, used at their bytes on this base.
They are not copied.

## Reproduce

From the repository root, Python 3.12, standard library only:

```sh
python3 -B scripts/research_package.py --package research/shared-port-raw-cut-invoice --check-only
python3 -B scripts/research_package.py --package research/shared-port-raw-cut-invoice
```

Direct checks, each under two seconds here:

```sh
python3 -B research/shared-port-raw-cut-invoice/verify.py
python3 -O -B research/shared-port-raw-cut-invoice/verify.py
```

Both print `PASS_SHARED_PORT_RAW_CUT_INVOICE` with `gap_t1=-2`. The verifier
rejects a missing aligning rank-1 call, an add-sign fault, a wrong pivot
operator, and a grade-2 letter hiding in the source schedule. No native suite,
Lean build, selected cold replay or full compiler/outer replay was run.

## Open next questions

1. Can a feedback write into the live owner, outside the recording-only
interface, separate `D x1` from `K a x0` with fewer than two extra rank-1
letters? This matters because that is the only Gaussian saving large enough
to flip `gap(1)`. The first step is one explicit feedback word and its paid
profile. The blocker is that every pure source letter is invertible, so a
left multiply cannot delete `D`.

2. Do the off-diagonal grade-3 terms in the final rows need their own cleanup
before those four jobs are legal children? This matters because the reward
used here already credits all four. The first step is to price one
off-diagonal class on row 3. The blocker is a consumer contract that says
whether an image-dimension credit is a legal child.

## Credit

The gate list and operators are the PR4813 / block3 schedule retained in
`research/block3-shared-port-composition`, including the hexapode, Dugongue,
Yiyang / Lyy-iiis, lee-phys, Jacob Alan Sussman and Cinna Davis / 1337isnot1337
lineage recorded there. The recording-only `4d` bound and the raw-cut `5d`
witness are Chris R / Clear1nsight and JoyboyBrian's open
[#5491](https://github.com/CrocSwap/integer-mult-bounds/pull/5491). This
package only prices the cleanup those results leave open.

Prepared by unpaidmoon with substantial Cursor Grok assistance. Apache-2.0.
30 changes: 30 additions & 0 deletions research/shared-port-raw-cut-invoice/SOURCES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
{
"schema_version": 1,
"base_commit": "32890bc5de034ab8fc93ac88ead7279a6a416d04",
"files": [
{
"path": "certificates/selected-result.json",
"usage": "provenance",
"git_blob": "fc3d4ba105e1f097d0d7d051a308281e86ab1446",
"sha256": "66974efdced4083c00025f97d95d83909a226cdf801edb9301c195c6641f6903"
},
{
"path": "docs/research/agent-task-catalog.json",
"usage": "provenance",
"git_blob": "f126dd2e2dad2b669e75d1d0d0bd05c600725602",
"sha256": "f7a48bfa2659b88378282a8eee7d9fe52cdf505594124c4d1686b58510fffd72"
},
{
"path": "research/block3-shared-port-composition/WORD.json",
"usage": "input",
"git_blob": "6126739ac5dce9122bc201c454e8213e8b1bac55",
"sha256": "05c81a7dfb37c240fdaef332474deb90f1b38e60dec392f3a23089f973a5f1ba"
},
{
"path": "research/block3-shared-port-composition/inputs/pr4813-verify_schedule.py",
"usage": "input",
"git_blob": "17a8439633effba74fff63a8fe4c332b8d709d80",
"sha256": "6216fd67f1a762303f0eed0e37b90ea0c72cf8e0cb7db70b1b27b2001ac20069"
}
]
}
78 changes: 78 additions & 0 deletions research/shared-port-raw-cut-invoice/TASK.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,78 @@
{
"schema_version": 1,
"task_id": "complex-rank-shape-decoder",
"priority_revision": 7,
"base_commit": "32890bc5de034ab8fc93ac88ead7279a6a416d04",
"original_card": {
"id": "complex-rank-shape-decoder",
"question": "Can a decoder deliver the local paid mass in fewer, larger-rank children (block frame climbs instead of rank-1 to rank-3 steps)?",
"starting_points": [
"docs/research/research-family-status.md",
"research/complex-local-word-ceiling/README.md",
"research/complex-decoder-field-extension-seed44/README.md",
"research/complex-frame-pair-decoders/README.md",
"research/selected-k8-unrestricted-ends/README.md",
"research/eisenstein-reduction-copy-floor/README.md"
],
"methods": [
"Find one exact small decoding identity whose frame path climbs in blocks, and price its complete per-call profile.",
"Bound the best effective child rank achievable for one fixed support alphabet, including copies, frames and restoration."
],
"acceptance_gate": "A per-call local profile, with all copied information, frames and restoration counted, whose ceiling exceeds the selected word's by at least 5%; masked or overcomplete supports must change an assumption of the eight-term all-pairs bound. A scoped local result may be retained without a complete supplier. Expanding it into further construction rounds requires the family-level budget and obligation gate in research-family-status.md; an optimistic passing relaxation alone is not acceptance.",
"first_step": {
"command": [
"python3",
"-B",
"research/complex-local-word-ceiling/verify.py"
],
"expected": "PASS_COMPLEX_LOCAL_WORD_CEILING",
"inspect": "Read the selected word's per-call local profile, its packing ceiling and the lever table; run levers.py for the ideal budget of the lever you target.",
"scope": "Positive control and planning budget only; this is not a new construction or full cold replay."
},
"first_experiment": "Before any synthesis, write the candidate per-call local profile (and centre charge / unpaid coordinates if changed) and screen it with ceiling.ceiling(); reject it if even the ideal ceiling does not beat the selected word's by your stated margin. Ideal budget: rank-16 delivery is 2.31x (levers.py).",
"bottleneck_link": "Rank-1 children are about half of recursive cost; the selected word's effective child rank is about 3.9. Larger-rank delivery lowers recursive cost per unit of paid mass.",
"allowed_variations": [
"One masked, overcomplete or block decoder family with exact labels and supports.",
"A scoped lower bound on effective rank for one alphabet.",
"If the coverage law, template or guard changes, derive an applicable fully charged replacement bound instead of applying the incumbent ceiling outside its scope."
],
"scope_change_triggers": [
"Repacking idle remainders of an unchanged word (covered by the packing ceiling).",
"Repeating the eight-term all-pairs search inside its proved scope.",
"Only changing helper terminal subspaces and monotone frames within the fixed scalar word, entrances and cleanup contract already covered by #1995/#2013.",
"Searching below C=40 in the #2017 Eisenstein binary scalar-rectangle or fixed-support block-linear additive pairing model; specify a changed premise instead. This bound does not cover every complex decoder.",
"Continuing a partial cost improvement while leaving the same unpriced consumer, restoration or central-readout obligation unresolved; refresh the family expansion decision first. Local product-retiming evidence has a narrow scope and is not a ban on changed graphs or contracts."
],
"overlap_terms": [
"rank shape",
"block climb",
"masked decoder",
"overcomplete",
"effective rank"
]
},
"scope_status": "refined",
"actual_scope": "Price the raw gate-11 shared-port cut as an exposed, continued and cleaned child, and decide whether its Gaussian moment is positive on the predecessor exponent interval.",
"scope_change_reason": "The catalogue card asks for a larger-rank decoder profile. The open handoff after the fixed-producer completion ceiling instead asks for one changed resource and a complete optimistic invoice before any synthesis. This package answers that invoice for the raw internal cut and stops because the gap is negative.",
"bottleneck_link": "This five-bank word spends its local mass on four rank-3 calls. The raw cut appears to add a fifth rank-3 job. The cleanup returns three extra rank-1 calls, so the route does not raise effective child rank.",
"budget_minutes": 60,
"stop_rule": "Stop when the Gaussian gap is nonpositive for every exponent t in [0, 1], including the one-letter grade-2 widening. Do not synthesize a decoder on this cut.",
"outcome": "negative-result",
"evidence": "python3 -B research/shared-port-raw-cut-invoice/verify.py prints PASS_SHARED_PORT_RAW_CUT_INVOICE with gap_t1=-2. The same command with -O passes. PROOF.md records the operator identity, the hybrid bill and the derivative comparison.",
"author": "unpaidmoon",
"ai_assistance": "Substantial Cursor Grok assistance.",
"open_questions": [
{
"question": "Can a feedback write into the live owner separate D x1 from K a x0 with fewer than two extra rank-1 letters?",
"why_it_matters": "That is the only Gaussian saving large enough to make gap(1) positive. Left multiplication cannot do it, because D is invertible.",
"first_step": "Write one explicit feedback word that uses the live owner and price its paid profile against 8 H1 + 3 H3.",
"blocked_by": "A legal feedback contract. The recording-only interface forbids writing into the producer, and every pure source letter is invertible."
},
{
"question": "Do the off-diagonal grade-3 terms in the final rows require more cleanup before those four jobs are legal children?",
"why_it_matters": "The reward credited here already counts all four. Further cleanup would deepen the deficit, not open a saving.",
"first_step": "Price one off-diagonal grade-3 class on final row 3 under the same source letters.",
"blocked_by": "A consumer contract that says whether image dimension 4d is itself a legal child or only a necessary condition."
}
]
}
31 changes: 31 additions & 0 deletions research/shared-port-raw-cut-invoice/VERIFY.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
{
"schema_version": 1,
"status": "ready",
"python": "3.12",
"requirements": [],
"inputs": [
"research/block3-shared-port-composition/WORD.json",
"research/block3-shared-port-composition/inputs/pr4813-verify_schedule.py"
],
"commands": [
{
"argv": [
"{python}",
"-B",
"{package}/verify.py"
],
"timeout_seconds": 60,
"checks": "Exact pivot identity, hybrid endpoint, paid profiles, unitarity, D factorization, and both negative endpoint gaps, plus aligning-call and ADD-sign rejection controls."
},
{
"argv": [
"{python}",
"-O",
"-B",
"{package}/verify.py"
],
"timeout_seconds": 60,
"checks": "The same certificate with assertions stripped. The checks use explicit exceptions, so the optimized run must still reject the controls."
}
]
}
Loading