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
169 changes: 169 additions & 0 deletions research/c6-t1-unequal-mask-chains/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,169 @@
# Unequal T1 masks are joined by a monotone chain

No new κ. In the unsigned six-source characteristic-6 floor model, every
legal endpoint pair whose masks differ is realized by a floor word that
keeps the multiset `(6,6,6,7,7,7)` at every diagonal index `s = 4,...,18`.
There are 2280 such pairs. The selected conditional κ stays
`781451/250000000`.

This is an exhaustive finite census of those pairs, with an explicit word
for each pair. It is not a search that failed, and it is not a statement
about weight-3 shapes outside this multiset. The equal-mask pairs are the
120 words already constructed in PR #7042. They are not recounted here.

## 1. Hypotheses carried from the endpoint lemmas

A floor word has length 36 and calls each of six sources exactly six times.
`q_t` is the count of the first `t` letters, and

```text
S(s) = q_s + q_{39-s}.
```

The one-step identity, for `s = 4,...,17`, is

```text
S(s+1) = S(s) + e_{w_s} - e_{w_{38-s}}.
```

PR #7042 proves two facts that this note uses and that `verify.py` rechecks
on their finite domains.

**Lemma 1.** If `S(18)` is a permutation of `(6,6,6,7,7,7)`, then
`q_18 = (3,3,3,3,3,3)` and the three coordinates equal to 7 are exactly the
three bridge letters in positions 18, 19 and 20. There are 20 such splits.

**Lemma 2.** If `S(4)` is a permutation of `(6,6,6,7,7,7)`, write
`M_4` for its mask and `D` for the final letter. Either `D` is outside
`M_4` and `q_4` is 1 on `M_4 ∪ {D}`, or `D` is in `M_4` and `q_4` is 2 at
`D` and 1 on the rest of `M_4`. There are 60 pairs of each type, and every
pair `(M_4, D)` occurs once.

The recheck visits every row assignment in the four-row table of Lemma 1
and every vector of sum 4 in `{0,...,6}^6` against every final letter. The
counts are 20 and 120, with no extra shape.

## 2. Budgets

Fix masks `M_4` and `M_18` and a final letter `D`. Let `p = q_4` be the
vector in Lemma 2, let `b` be the indicator of `M_18`, and let `d` be the
indicator of `D`. The letters still to place are 14 early letters, in
positions 4 through 17, and 14 late letters, in positions 21 through 34.
Their per-source budgets are

```text
e = (3,3,3,3,3,3) - p,
l = (3,3,3,3,3,3) - d - b.
```

Every coordinate of `p` is 0, 1 or 2, so `e` is positive. Every coordinate
of `d` and of `b` is 0 or 1, so `l` is positive: the smallest entry is 1,
when `D` lies in `M_18`. The sums are `∑ p = 4`, `∑ b = 3` and `∑ d = 1`,
hence `∑ e = ∑ l = 14`. These formulas are checked on every orbit
representative.

A chain step consumes one early letter `a` and one late letter `b`.

- If `a = b`, the mask is unchanged.
- If `a ≠ b`, the step is legal only when `a` is currently a 6 and `b` is
currently a 7. The mask then loses `b` and gains `a`.

Either step preserves the multiset `(6,6,6,7,7,7)`. After 14 steps the
budgets are exhausted.

## 3. From a path to a word

Given such a path `(a_t, b_t)` for `t = 0,...,13`, build

```text
prefix + (a_0,...,a_13) + sorted(M_18) + (b_13,...,b_0) + (D).
```

The prefix is the increasing list `M_4 ∪ {D}` when `D` is outside `M_4`,
and `(D, D, u, v)` with `u < v` the other two mask sources when `D` is in
`M_4`. Position `4+t` holds `a_t` and position `34-t` holds `b_t`. The
bridge is positions 18, 19 and 20. It lies in every late prefix used by
`s = 4,...,17`, so it cancels in the one-step difference. It contributes
to `S(18)` exactly once.

The call count of source `j` is `p_j + e_j + b_j + l_j + d_j = 6`. The
length is `4 + 14 + 3 + 14 + 1 = 36`. Starting from the Lemma-2 vector
`S(4)`, the path rule keeps the multiset through `s = 18`, and the final
mask is `M_18` because the bridge indicator is `b` and `q_18 = p + e` is
identically 3. The program checks `q_18`, both endpoint masks, and the
multiset at every `s = 4,...,18`, rather than trusting this paragraph alone.

## 4. The ten orbits

Two unequal triples `(M_4, D, M_18)` and `(M_4', D', M_18')` lie in one
orbit of the symmetric group on the six sources when they have the same
intersection size and `D` lies in the same combination of the two masks.
The sources then split into four regions whose sizes depend only on that
intersection size `k`:

| region | size |
|---|---:|
| `M_4 ∩ M_18` | `k` |
| `M_4 \ M_18` | `3-k` |
| `M_18 \ M_4` | `3-k` |
| neither | `k` |

For `k = 0` the regions `both` and `neither` are empty, so `D` lies in
exactly one mask. The other two intersection sizes admit all four
membership patterns. That is ten nonempty classes. Their sizes, from
listing every pair, are

| intersection | `D` in `M_4` | `D` in `M_18` | pairs |
|---:|---|---|---:|
| 0 | no | yes | 60 |
| 0 | yes | no | 60 |
| 1 | no | no | 180 |
| 1 | no | yes | 360 |
| 1 | yes | no | 360 |
| 1 | yes | yes | 180 |
| 2 | no | no | 360 |
| 2 | no | yes | 180 |
| 2 | yes | no | 180 |
| 2 | yes | yes | 360 |

The total is 2280. Equal masks are excluded: there are `20 · 6 = 120` of
them, already realized by pure slides.

A permutation that matches the four regions, and sends `D` to `D'`, sends
a realizing word to a realizing word. Masks and the final letter move with
the sources, and the multiset of values of `S(s)` does not change. The
program builds one such permutation for every pair in an orbit and checks
the image word.

## 5. The census

For each of the ten orbits the program searches the budget graph from the
Lemma-2 mask, with the budgets of Section 2. The state is the remaining
early budget, the remaining late budget, and the current mask. Each
coordinate is at most 3, and the mask has three members. Breadth-first
search returns a path of length 14 in every orbit. The assembled
representative keeps the multiset, has `q_18` identically 3, and has the
requested endpoint masks. Transporting that word along the region
bijection succeeds for all 2280 pairs.

The recorded moving word of PR #7042,

```text
012340011223344555014555443322110023
```

is one pair in the intersection-2 class with `D` in neither mask. It is
checked directly. Swapping its first two early letters leaves the
multiset. That is a failure control, not a second census.

No orbit is reported empty. An empty search is not used as a nonexistence
proof, because none occurred. The statement is existence for every pair in
a finite, fully listed domain.

## 6. What this does not say

The alphabet is unchanged. No signed bill is computed, and the selected
frontier is not changed. Lemma 1 still forces `q_18` to be identically 3,
so these words remain outside the `(0, 9)` shape that PR #7029 called
(H1). The other five separation-3 families are not searched. A word that
keeps this multiset is not a classification of every weight-3 diagonal.
87 changes: 87 additions & 0 deletions research/c6-t1-unequal-mask-chains/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,87 @@
# Every unequal T1 mask pair has a monotone chain

**Exhaustive finite census, with an explicit word for every pair. No new κ.**
In the unsigned six-source characteristic-6 floor model, all 2280 endpoint
pairs whose masks differ are realized by a floor word that keeps the
multiset `(6,6,6,7,7,7)` at every index `s = 4,...,18`. The equal-mask
pairs are the 120 pure slides of PR #7042 and are not rebuilt as the new
claim. The selected conditional κ remains `781451/250000000`.

The argument and the orbit table are in [PROOF.md](PROOF.md).

## Scope and evidence

Original card: `connector-alphabet-change`. The card asks whether one new
connector operation has a favorable budget. Scope status: **changed**. The
alphabet stays the unsigned dirty connector. The assigned question is open
question 1 of PR #7042: which unequal endpoint masks are joined by a
monotone chain of this multiset.

The predecessor already showed that every equal-mask pair works and that
one unequal pair works. The missing statement was the boundary of that
transport. This package lists every unequal pair, splits them into the ten
orbits of the symmetric group on the six sources, finds one 14-step path
in each orbit, and checks the transported word on all 2280 pairs.

Evidence class: **exhaustive finite census**. The domain is the finite set
of triples `(M_4, D, M_18)` with `M_4 ≠ M_18`. The enumeration is every
3-subset pair and every final letter. Each word is checked against the
diagonal multiset at all fifteen indices. A failed search is not used; every
orbit returned a witness. This is not a proof about shapes other than
`(6,6,6,7,7,7)`, and it is not unrestricted nonexistence or existence
outside this model.

`current-priorities.md` on this base still prints the older decimal from
#5599. The verifier reads `certificates/selected-result.json` and does not
edit it.

## Reproduce

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

```sh
python3 -B research/c6-t1-unequal-mask-chains/verify.py
python3 -O -B research/c6-t1-unequal-mask-chains/verify.py
python3 -B scripts/research_package.py --package research/c6-t1-unequal-mask-chains
```

The marker is

```text
PASS_C6_T1_UNEQUAL_MASK_CHAINS pairs=2280 orbits=10 witnesses=10 checked=2280 controls=7
```

The controls recheck the two endpoint domains (20 splits at `s = 18`, 120
splits at `s = 4`), the recorded moving word, a swapped-letter failure, the
ten orbit sizes, and the selected κ string. Checks raise `SystemExit`, so
`-O` does not strip them.

Not run: a proof assistant, a SAT solver, a supplier or native replay, and
`scripts/research_selected_baseline.py`. The selected record is provenance
and the κ guard only.

## Open next questions

- For one separation-3 family other than type T1, which unequal endpoint
masks are joined by a monotone chain of that family's multiset? *Why:*
T1 is the only family with no coordinate 12, and its transport is now
complete. A family that uses a 12 has a different legal step. *First
step:* fix that family's multiset, list its legal exchange, and search
one symmetry class of unequal masks. *Blocked by:* the step rule in this
package exchanges a 6 with a 7 only. It does not apply to a coordinate
12.

## Credit

The depth-two interface is Nick Petalas's (PR #4231, PR #4248, PR #6107),
with the Ross Richardson PR #934/PR #943 lineage. JosieElliston's PR #6141
records the c=6 solver verdict. PR #6955 classifies the empty centre.
PR #7029 isolates type T1 as the family with no frozen 12. PR #7042, head
`807149a85d185b8272aa1fda43ba6d8d7c541cfe`, constructs the 120 equal-mask
words, one moving word, and the two endpoint lemmas used here. Those lemmas
are rechecked on the same finite domains and are not a new classification.
The upstream finding that records the open question is discussion comment
19514157.

Prepared by unpaidmoon with substantial Cursor Grok assistance. Independent
human review and a proof-assistant proof are unclaimed. Apache-2.0.
18 changes: 18 additions & 0 deletions research/c6-t1-unequal-mask-chains/SOURCES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
{
"schema_version": 1,
"base_commit": "17469fff78c22941280f04d39dc5a1faee09c840",
"files": [
{
"path": "certificates/selected-result.json",
"usage": "input",
"git_blob": "d59b5035e7b4cbaa9d688a1d32dd0d6eb79c466a",
"sha256": "9b8d0eea95b3bf0dfcb7938a732a6439b87a2685dec2a5ab9e525b3b34debbfa"
},
{
"path": "docs/research/agent-task-catalog.json",
"usage": "provenance",
"git_blob": "3404fbdd5e7529a0674eef26114b5071e4586cec",
"sha256": "ead6cdc9d0af6d58e8a49744d8ec8c7c15e13d2ba93c276c434bcf9b51b93a81"
}
]
}
62 changes: 62 additions & 0 deletions research/c6-t1-unequal-mask-chains/TASK.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,62 @@
{
"schema_version": 1,
"task_id": "connector-alphabet-change",
"priority_revision": 8,
"base_commit": "17469fff78c22941280f04d39dc5a1faee09c840",
"original_card": {
"id": "connector-alphabet-change",
"question": "Does one explicit extension of the connector alphabet offer a favorable completed budget?",
"starting_points": [
"docs/research/research-family-status.md",
"research/universal-connector-boundary/README.md",
"certificates/selected-result.json"
],
"methods": [
"Construct and charge a legal extra operation on a minimal instance.",
"Prove a target-specific bound after adding one specified operation family."
],
"acceptance_gate": "State which fixed-alphabet premise changes and charge its implementation; do not declare it free. 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/universal-connector-boundary/verify.py"
],
"expected": "PASS_UNIVERSAL_CONNECTOR_BOUNDARY",
"inspect": "Reproduce the 945-pairing control, then read Exact interface to list the allowed calls.",
"scope": "Positive control / source-bound orientation only; this is not a new construction or full cold replay."
},
"first_experiment": "Add exactly one legally priced operation and test a tiny dirty-input instance before enumerating larger cases.",
"bottleneck_link": "A new legal connector operation could change recursive interface cost; a larger census in the same alphabet is a separate side result.",
"allowed_variations": [
"Choose one additional operation and a small source count for its first counterexample."
],
"scope_change_triggers": [
"Keeping the alphabet and only increasing the source count; record this as a changed assignment.",
"Extending a source-count census without naming a supplier/interface decision it could change; audit #2973 pruning concerns before relying on the affected predecessor exclusions."
],
"overlap_terms": [
"universal connector",
"source-toggle",
"connector alphabet"
]
},
"scope_status": "changed",
"actual_scope": "In the unsigned six-source characteristic-6 floor model, all 2280 endpoint pairs with unequal masks of the multiset (6,6,6,7,7,7) are realized by an explicit floor word that keeps that multiset at every s=4..18.",
"scope_change_reason": "The card asks whether one new connector operation has a favorable budget. This package keeps the unsigned dirty connector alphabet fixed and answers open question 1 of PR #7042: which unequal endpoint masks are joined by a monotone chain of this multiset. No signed bill is computed.",
"bottleneck_link": "A missing unequal-mask chain would have been a diagonal obstruction inside type T1 that uses neither the (0, 9) hypothesis nor a frozen 12. Every such chain exists, so this multiset adds no further obstruction of that kind. The selected frontier stays 781451/250000000.",
"budget_minutes": 60,
"stop_rule": "Stop after the ten orbit witnesses and the check of all 2280 transported words. Do not price a supplier and do not search the other five separation-3 families.",
"outcome": "completed",
"evidence": "verify.py rechecks the endpoint domains, finds a 14-step path in each of the ten orbits, and checks the transported floor word on all 2280 unequal pairs. It prints PASS_C6_T1_UNEQUAL_MASK_CHAINS. Python 3.12, standard library only.",
"author": "unpaidmoon",
"ai_assistance": "Substantial Cursor Grok assistance. Independent human review and a proof-assistant proof are unclaimed.",
"open_questions": [
{
"question": "For one separation-3 family other than type T1, which unequal endpoint masks are joined by a monotone chain of that family's multiset?",
"why_it_matters": "T1 is the only one of the six families with no coordinate 12, and every unequal T1 pair now has a chain. A family that uses a 12 has a different legal step.",
"first_step": "Fix that family's multiset, write the legal exchange, and search one symmetry class of unequal masks.",
"blocked_by": "The step rule used here only exchanges a 6 with a 7. It does not apply to a coordinate equal to 12."
}
]
}
30 changes: 30 additions & 0 deletions research/c6-t1-unequal-mask-chains/VERIFY.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
{
"schema_version": 1,
"status": "ready",
"python": "3.12",
"requirements": [],
"inputs": [
"certificates/selected-result.json"
],
"commands": [
{
"argv": [
"{python}",
"-B",
"{package}/verify.py"
],
"timeout_seconds": 120,
"checks": "Endpoint-domain recheck, ten orbit searches, explicit transport of all 2280 unequal pairs, moving-word control, swapped-letter failure, and the selected kappa string."
},
{
"argv": [
"{python}",
"-O",
"-B",
"{package}/verify.py"
],
"timeout_seconds": 120,
"checks": "The same census under python -O. Checks raise SystemExit, so optimized mode does not drop them."
}
]
}
Loading