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
19 changes: 18 additions & 1 deletion docs/ci-verification.md
Original file line number Diff line number Diff line change
@@ -1,8 +1,25 @@
# CI verification

## Current selected-family replay: #6030
## Current selected-family replay: #7000 with #7001

The [selected record](../certificates/selected-result.json) binds conditional
κ = 0.00363005817 to the [application review](research/pr-intake-evidence/pr7000-application/PROOF-REVIEW.md),
[validation receipts](research/pr-intake-evidence/pr7000-application/VALIDATION.json),
and [explicit source closure](../certificates/frontier7000-source.json).
`make frontier7000-verify` follows that exact-source selected replay. The review
reuses eleven authenticated preceding executions and records seven fresh serial
application, scalar and complete native-catalogue executions. The earlier
monolithic timeout is retained separately and is not completed coverage.
Consult the receipts for exact commands, source heads and limits.

The five inherited framework assumptions remain. No complete uniform word,
literal multiplier execution, independent human proof or formal-kernel proof
is asserted. Routine research submissions do not need to rerun this
selected-family replay. The sections below preserve historical checkpoints.

## Historical selected-family replay: #6030

The [historical checkpoint](../certificates/checkpoints/frontier6030.json) binds conditional
κ = 0.003125804 to the [proof review](research/pr-intake-evidence/pr6030-audit/PROOF-REVIEW.md),
[validation receipts](research/pr-intake-evidence/pr6030-audit/VALIDATION.json),
and [explicit source closure](../certificates/frontier6030-source.json).
Expand Down
138 changes: 138 additions & 0 deletions research/gaussian-outward-radius24/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,138 @@
# Exact forward Gaussian rounding threshold through radius 24

For the normalized four-coordinate Hadamard transform \(U=H/2\), put
\[
H=\begin{pmatrix}
1&1&1&1\\
1&-1&1&-1\\
1&1&-1&-1\\
1&-1&-1&1
\end{pmatrix},\qquad
S_R=\{x\in\mathbb Z^4:\|x\|_2\le R\}.
\]
An admissible map is an injection \(F:S_R\to\mathbb Z^4\) with
\(\|F(x)-Ux\|_2\le1\) at every input. Its forward slack is
\(\max_x(\|F(x)\|_2-\|x\|_2)\).

**New finite certificate.** On \(S_{24}\), an admissible map exists with
forward slack at most \(\beta=\sqrt{78}-\sqrt{73}\). The verifier constructs
and checks an integral flow that saturates all **1,638,241** inputs.

Jerry Chen's [PR #7571](https://github.com/CrocSwap/integer-mult-bounds/pull/7571)
proves the lower bound \(\beta\) at radius 9. With that lower bound, restriction
of the radius-24 injection gives the exact optimum \(\beta\) at every real
radius \(9\le R\le24\). In particular, the radii above 11 left open in that
work have the same optimum through 24. The new check establishes the
radius-24 upper bound. It uses the earlier lower bound as a cited premise.

For Ali Romman's [ideal Gaussian pair in #6509](https://github.com/CrocSwap/integer-mult-bounds/pull/6509),
write \(C_+(x,y)=((x+y+i(x-y))/2,(x+y-i(x-y))/2)\) in real-coordinate order
\((\Re x,\Im x,\Re y,\Im y)\). With \(S=\operatorname{diag}(1,-1,1,1)\)
and \(L(a,b,c,d)=(a,c,-d,b)\), the identity \(U=LC_+S\) transfers
the certificate as \(L^{-1}FS\). Both signed permutations are lattice
bijections that preserve Euclidean norms.

## Why the flow proves an injection

**Candidates.** Every coordinate of \(Hx\) has the same parity as the sum of
the coordinates of \(x\). When that sum is even, integer outputs within
distance one of \(Ux\) are its center and eight signed unit translations. When the
sum is odd, each coordinate is half-integral. The 16 corners from the two
nearest integers in each coordinate are all at distance one. Any other output
has some coordinate at distance at least \(3/2\).

**Orbits.** Use the even-sign signed permutations \(G\) of four coordinates.
They are generated by reflections in the normals \(e_i-e_j\) and \(e_i+e_j\).
The former swap two coordinates and the latter swap and negate them. Under
conjugation by \(U\), each of these normals maps to another vector with
exactly two nonzero entries, both \(\pm1\). This follows because distinct
Hadamard columns agree at two positions and disagree at two. Consequently
\(UGU^T=G\). The paired action \((x,y)\mapsto(Ax,UAU^Ty)\) preserves
the error and both radii.

An orbit is determined by its sorted absolute coordinates and, if none
are zero, its sign product. A zero allows an extra sign flip without moving
the point. For multiplicities \(m_a\) of each absolute coordinate, the orbit
size is
\[
\frac{4!}{\prod_a m_a!}\;2^{\min(3,4-m_0)}.
\]
Running the candidate rule on one representative gives the complete list
of incident output orbits. The graph has 11,383 input orbits and 11,661
output orbits incident to at least one allowed edge.

**Exact gap.** If \(m=\|y\|_2^2\le n=\|x\|_2^2\), the edge is
allowed. For \(m>n\), put \(d=m-n\) and \(t=\beta^2\). Because
\(0<\beta<1\), we have \(d>t\). Squaring the equivalent comparison
\(d-t\le2\sqrt{nt}\) gives
\[
t^2-(2d+4n)t+d^2\le0.
\]
Substituting \(t=151-2\sqrt{5694}\) gives \(p+q\sqrt{5694}\le0\)
for integer \(p,q\). Their signs determine when it is valid to compare the
squares. This preserves equality at the threshold.

**Lift.** The witness has a nonnegative integer amount for each used
orbit pair. The checker tests adjacency, outgoing sum equal to input orbit
size, and incoming sum at most output orbit size. For a pair of orbits
\(I,O\), distribute its amount uniformly over the actual edges from \(I\)
to \(O\). Transitivity makes that bipartite subgraph biregular. Hence
each input vertex gets total weight one and each output vertex gets
total weight at most one. For any input subset, its total outgoing weight
is its cardinality and fits within its neighborhood's cardinality.
Hall's theorem gives an injection on the full ball.

## Reproduction

From the repository root with CPython 3.12 and its standard library:

```sh
python3.12 -I -W error -B research/gaussian-outward-radius24/verify.py
python3.12 -I -O -W error -B research/gaussian-outward-radius24/verify.py
python3.12 -B scripts/research_package.py --package research/gaussian-outward-radius24
python3.12 -B scripts/check_research_commit.py --package research/gaussian-outward-radius24 --check-source-history
```

Both direct modes report `PASS_GAUSSIAN_OUTWARD_RADIUS24` with
21,458 positive orbit flows. The replay recomputes the witness and checks
it independently of the flow solver's saturation report. A separate
integer count slices the ball along the fourth coordinate. Controls
check equality and a larger gap at the threshold, reject invalid flows,
and exercise a residual path that reroutes an earlier assignment.

Replay is read-only and needs no external data. Its managed storage is
bounded by the fixed radius. Each of at most \(2\binom{28}{4}\) input orbit
keys has at most 16 candidate edges, and each output coordinate has
magnitude at most 25. The bounded residual search uses integer capacities.
The predecessor lower bound and signed-permutation transfer are written
proof dependencies. The replay does not run predecessor verifiers.

## Scope and credit

The finite certificate leaves all-lattice attainment, exact-sum and phase
identities, physical rounder costs, and selected caller admission open.
It yields no new \(\kappa\). The `uniform-transfer-audit` card changes from
an actual-caller implication to a finite precision-contract question. The
selected baseline check was metadata orientation.

This contribution was produced by Nick Petalas. Credit Jerry Chen for
[the finite Hall and orbit-flow result at exact head
`3c7c3e8000f592fd9adecb44008ff6d5ae5d453d`](https://github.com/CubeJerry/integer-mult-bounds/blob/3c7c3e8000f592fd9adecb44008ff6d5ae5d453d/research/sharp-finite-gaussian-outward-slack/orbit-extension/PROOF.md)
and for the larger-radius question. Credit Ali Romman for the Gaussian
pair and same-grid inverse in #6509, ywz65 for the
[finite-ball rounding boundary in #2589](https://github.com/CrocSwap/integer-mult-bounds/pull/2589),
and Nick Petalas for the
[zero-slack obstruction in #6619](https://github.com/CrocSwap/integer-mult-bounds/pull/6619).
The classical finite matching argument retains its usual attribution.
[Advance claim](https://github.com/CrocSwap/integer-mult-bounds/discussions/1166#discussioncomment-19693299).

## Open next questions

- Does the same threshold admit an injection on the entire lattice? A
radius-independent argument for Hall cuts would suffice by compactness
because every input has a finite neighbor set. The present result
controls only subsets inside the certified balls.
- Can an attained global forward bound coexist with exact sum or a
caller's phase identity? Begin by restricting the candidate edges to
a named caller's required invariants and identifying the resulting cuts.
The unrestricted graph does not encode these requirements.
18 changes: 18 additions & 0 deletions research/gaussian-outward-radius24/SOURCES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
{
"schema_version": 1,
"base_commit": "e54523517d0f03e58352e647c403a5af3e525a57",
"files": [
{
"path": "certificates/selected-result.json",
"usage": "provenance",
"git_blob": "75d77b328ea02ded535e7f23aee5b1e55d0bb1a8",
"sha256": "154c3d30455d778fd3343b6b4851f0fa4665f0f7625b0adb3675e2bafbbf6bf1"
},
{
"path": "docs/research/agent-task-catalog.json",
"usage": "provenance",
"git_blob": "3404fbdd5e7529a0674eef26114b5071e4586cec",
"sha256": "ead6cdc9d0af6d58e8a49744d8ec8c7c15e13d2ba93c276c434bcf9b51b93a81"
}
]
}
89 changes: 89 additions & 0 deletions research/gaussian-outward-radius24/TASK.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,89 @@
{
"schema_version": 1,
"task_id": "uniform-transfer-audit",
"priority_revision": 8,
"base_commit": "e54523517d0f03e58352e647c403a5af3e525a57",
"original_card": {
"id": "uniform-transfer-audit",
"question": "Can one of the selected record's five inherited assumptions be proved at its actual caller?",
"starting_points": [
"docs/research/current-priorities.md",
"docs/research/research-family-status.md",
"certificates/selected-result.json",
"docs/research/pr-intake-evidence/pr5599-completion/PROOF-REVIEW.md"
],
"methods": [
"Construct and independently check one source-bound instance of the stated first experiment.",
"Seek an exact counterexample or target-specific complete-budget obstruction to that same proposed change."
],
"acceptance_gate": "Identify the proposition and what would change downstream; do not equate a finite replay with a universal proof.",
"first_step": {
"command": [
"python3",
"-B",
"scripts/research_selected_baseline.py"
],
"expected": "PASS_RESEARCH_BASELINE_BINDINGS_ONLY",
"inspect": "Read revision 8, the selected review and live exact-source handoffs. Then follow this card's bounded first_experiment.",
"scope": "Source-bound orientation only, not a supplier replay or theorem proof."
},
"first_experiment": "Choose one assumption ID (native_all_columns, owned_erasure, compiler, precision_tape, analytic_recovery), one quantified implication and exact dependency closure. Check current handoffs, including #5625 row reentry and #5707 root-prefix lowering, without treating their partial scope as the whole theorem. Prove or falsify that implication with a caller-bound argument; stop at the chosen obligation. Finite tests or root digits do not discharge an all-size statement.",
"bottleneck_link": "Judge the actual current supplier contract and complete signed target/paid bill; frozen-profile ceilings require an applicability proof.",
"allowed_variations": [
"One exact source head, helper/cut region or quantified proposition and changed hypothesis; check live handoffs first. Different seeds are not different scopes."
],
"scope_change_triggers": [
"Changing source head, helper/cut domain, operation alphabet or caller boundary without recording the new scope.",
"Treating an archived proposal or local numerical pass as complete supplier admission."
],
"overlap_terms": [
"native_all_columns",
"owned_erasure",
"compiler",
"precision_tape",
"analytic_recovery"
],
"existing_checks": [
{
"scope": "PR #4100: proposed unconditional construction and finite checks, without independent acceptance.",
"evidence": "research/unconditional-batched-crt/README.md",
"remaining": "Independent proof audit of one caller-bound all-size implication; all other premises remain explicit.",
"status": "author-checked",
"source_commit": "337d04939571a637aeab54a0f58f54fde811663a",
"selected_record_sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
},
{
"scope": "Selected PR5599: complete helper-subspace construction and independent caller replay; five framework assumptions remain.",
"evidence": "docs/research/pr-intake-evidence/pr5599-completion/PROOF-REVIEW.md",
"remaining": "The next experiment must identify uncovered source-specific work. Successor archive intake does not transfer this admission.",
"status": "reviewed",
"source_commit": "09dd0965711ca7eeb169057a420abab1c9052137",
"selected_record_sha256": "2a0220e866417774606e0f991e41b9c1d26b1f43b97857238cf478dbc7d330b5"
}
]
},
"scope_status": "changed",
"actual_scope": "Certify attainment of the known forward slack threshold for injective error-one Hadamard/Gaussian rounding on the radius-24 lattice ball.",
"scope_change_reason": "PR #7571 leaves radii above 11 open. This finite rounding question isolates a precision contract with no selected-caller binding.",
"bottleneck_link": "Side result on outward norm slack under same-grid injectivity. Supplier cost and caller admission remain separate.",
"budget_minutes": 60,
"stop_rule": "Stop at the first stronger certified finite result and record the all-lattice question as open.",
"outcome": "completed",
"evidence": "Exact integral orbit flow saturates all 1638241 input points at sqrt(78)-sqrt(73). The cited PR #7571 lower bound gives equality for every radius from 9 through 24.",
"author": "Nick Petalas",
"ai_assistance": "Independent human and formal proof-kernel review are unclaimed.",
"open_questions": [
{
"question": "Does the same threshold admit an injection on the entire lattice?",
"why_it_matters": "It would close the gap between the finite forward lower bound and the global constructive upper bound.",
"first_step": "Find a radius-independent bound on the forward Hall cuts in this local graph.",
"blocked_by": "A uniform extension argument for the finite flow remains open."
},
{
"question": "Can a rounder at the optimum also preserve the invariants of a concrete Gaussian caller?",
"why_it_matters": "An injective rounding certificate alone is insufficient for a native caller.",
"first_step": "Restrict the candidate edges to one caller's exact sums or phase conditions and assess the new Hall cuts.",
"blocked_by": "The unrestricted graph omits caller constraints and restoration costs."
}
]
}
34 changes: 34 additions & 0 deletions research/gaussian-outward-radius24/VERIFY.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,34 @@
{
"schema_version": 1,
"status": "ready",
"python": "3.12",
"requirements": [],
"inputs": [],
"commands": [
{
"argv": [
"{python}",
"-I",
"-W",
"error",
"-B",
"{package}/verify.py"
],
"timeout_seconds": 60,
"checks": "Exact radius-24 orbit flow, witness constraints, complete input count and negative controls."
},
{
"argv": [
"{python}",
"-I",
"-O",
"-W",
"error",
"-B",
"{package}/verify.py"
],
"timeout_seconds": 60,
"checks": "Exact radius-24 orbit flow, witness constraints, complete input count and negative controls."
}
]
}
Loading