Skip to content

Commit 794598a

Browse files
authored
Merge pull request #1012 from AdaWorldAPI/claude/step2-copula-and-mask-algebra
Step 2: ruling request + copula addendum — content never travels in classid
2 parents 4cb3661 + edf21ab commit 794598a

10 files changed

Lines changed: 3501 additions & 0 deletions

.claude/board/EPIPHANIES.md

Lines changed: 412 additions & 0 deletions
Large diffs are not rendered by default.

.claude/board/INTEGRATION_PLANS.md

Lines changed: 112 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,115 @@
1+
## 2026-08-23 — STYLE-MICROCODE FRONTIER, PHASE 1 (learner = revise + CHOICE; Phase 2 = R2IL, recorded)
2+
3+
`PROBE-STYLE-MICROCODE-FRONTIER-1` (9/9) maps the operator's intent —
4+
thinking styles as microcode with Autopoiesis-frontier reinforcement —
5+
onto shipped machinery and finds **no learner subsystem is needed**:
6+
styles are ordered groups of typed ops; frozen + explore coexist as a
7+
superposition over one op vocabulary; outcomes revise style-level NARS
8+
claims at runtime (stamped, double-count-inert); dispatch is
9+
expectation() + measured cost, flipping to the cheaper PROVEN style;
10+
freezing is the LearnedSurvivedTests admission predicate (#1011) at the
11+
style level; cheap-but-failing exploration is never reinforced (F6);
12+
evolution mints new frontier variants without mutating frozen microcode.
13+
Per-style learned state = one TruthValue + one Stamp, both shipped.
14+
**PHASE 2, recorded not built:** R2IL is the way richer op vocabulary
15+
(reconstructible typed behavior — typed drill, interventions,
16+
counterfactuals) and the identical loop lifts onto it once Phase 1's
17+
shape is ruled sound; V4 classid stays provisional, no R2IL type
18+
imported. **Widened synthesis recorded as hypothesis (three IFs, zero
19+
decisions): R2IL × BPE with OGAR-loco-shaped routing macros, V4 as the
20+
thinking-dynamic plane** — each conditional on measured recurrence per
21+
the root order's admission conditions; nothing built or reserved.
22+
Board: `E-THE-FRONTIER-LEARNER-IS-ALREADY-SHIPPED-1`.
23+
24+
## 2026-08-23 — MULTI-GROUP MEMBERSHIP PROBE (root order executed; carrier UNDECIDED)
25+
26+
`PROBE-MULTI-GROUP-MEMBERSHIP-1` (11/11) executes the operator's root
27+
order: the simple primitive FIRST. Multi-group membership is proven —
28+
one resident item in 2+ groups with a third addable (M1), one canonical
29+
`Membership` relation with inverse views (M2), masks as
30+
delete/rebuild-safe derived artifacts (M3), zero population movement
31+
across join/leave (M4), and scope inheriting down while cross-subtree
32+
membership stays pure relation topology (M5: GROUP MEMBERSHIP IS RELATION
33+
TOPOLOGY, NOT HHTL ANCESTRY). Behavioral half over #1001-shaped receipts:
34+
exact replay (B1), recurrence measured and honestly labelled as
35+
driver-recurrence (B2), ordered-group reconstruction with order preserved
36+
(B3), repeated-but-refused behavior structurally excluded from macro
37+
candidacy (B4/F6), and grouping currently ADDING bytes over raw receipts
38+
(B5). **BEHAVIORAL COMPRESSION CARRIER: UNDECIDED** — production-scale
39+
recurrence unmeasured; if too rare, the correct result is NO BPE. Board:
40+
`E-MEMBERSHIP-IS-PARTICIPATION-NOT-ANCESTRY-1`.
41+
42+
## 2026-08-23 — TOKEN-BPE GEOMETRY PROBE (bounded; verdict CAN-FIT, NOT YET BUY)
43+
44+
`PROBE-TOKEN-BPE-GEOMETRY-1` (8/8) answers the bounded operator question:
45+
can BPE act as a reconstructible intake tokenizer over the fixed `6×(8:8)`
46+
geometry without touching HHTL, classid semantics, or the resident ABI?
47+
**Yes it fits** (3.35× on the real in-tree KJV scene, byte-exact decode,
48+
`[u8;12]` Copy particles, zero classid involvement) — **nothing yet
49+
justifies a production carrier** (one fixture-scale corpus; the scale
50+
corpora are absent and reported absent). Reading C (BPE over lawful byte
51+
symbols) is the clean candidate; reading B is pair-encodable but its merge
52+
tree is MEASURABLY not a radix prefix partition (3 same-depth prefix
53+
collisions) — a merge tree is not the ontology tree, now as a fact.
54+
Surprises: scoped vocabularies LOST to global here (19% more tokens);
55+
vocab saturated at 180/255; EVERY verse overflows one particle (p50=4) so
56+
continuation rows are the norm. Board:
57+
`E-TOKEN-BPE-CAN-FIT-NOT-YET-BUY-1`. **Hard fence maintained:** this is
58+
TOKEN BPE; the behavioral-BPE / multi-group-membership investigation is a
59+
separate queued deliverable and inherits nothing from this result.
60+
61+
## 2026-08-23 — STEP 2: the open distributions MEASURED (addendum closed)
62+
63+
`PROBE-COPULA-DISTRIBUTION-1` (5/5) ran the two distributions the addendum
64+
named as open, and corrected the addendum twice. **`nars::facet_fold` (M26)
65+
already carries the copula losslessly in the M20 resident register** — a
66+
2-bit tag on rail 1 plus `Rel`'s u16 across rails 1+3, round-trip-exact on
67+
all 16 measured statements, **0 extra bytes, zero classid involvement**; the
68+
addendum's "sparse relation rows" was a hypothesis for shipped code.
69+
**"KJV Rel-heavy" was REFUTED** — real KJV through the real `stance`
70+
producer is Inh-dominated (Inh 13 vs Rel 2), so a Rel-heavy regime is
71+
UNDEMONSTRATED rather than unmeasured. **"tactics Impl" was a phantom**
72+
(tactics emits only Inh/Sim). Step 2's `cop` item now has a RECOMMENDATION:
73+
candidate **E — existing-tenant composition via `facet_fold`** — already
74+
shipped and costing nothing, so the still-blocked whole-KJV *scale*
75+
measurement does not gate adopting it.
76+
77+
## 2026-08-23 — STEP 2 ADDENDUM: the copula correction (retraction + measurement)
78+
79+
`.claude/plans/belief-abi-step2-addendum-copula-v1.md` — operator-directed
80+
correction of the ruling request's Item 1. **Retracts** `Copula → classid
81+
reference` (C1–C4 proved `COPULA ≠ RAIL PLACEMENT`, which never implied
82+
`COPULA = CLASSID`); **states the law** (CONTENT NEVER TRAVELS IN CLASSID —
83+
classid selects the reading); **measures** the Active-Directory
84+
group/membership interpretation on shipped operators
85+
(`PROBE-COPULA-GROUP-MASK-1`, 9/9: copula content resident in relation
86+
rows, groups as lossy-by-design ergonomics, members/memberOf as inverse
87+
views over one relation, no ancestry-faking, one classid across four
88+
copulas, one-pass group ∩ region ∩ condition composition); and **proposes
89+
NO mint** — candidate homes graded with masks-alone rejected as sole
90+
carriers and workload-scale distributions (KJV Rel-heavy, tactics Impl)
91+
named as the open measurements. Board entry:
92+
`E-CONTENT-NEVER-TRAVELS-IN-CLASSID-1`.
93+
94+
## 2026-08-23 — BELIEF-ABI-RESTORATION-1 STEP 2 (ruling request — awaiting operator decision)
95+
96+
`.claude/plans/belief-abi-step2-ruling-request-v1.md` — prepares the
97+
charter's Step 2 (*"operator ruling on the residue, per item, not
98+
wholesale"*). **Makes no ruling**; every entry is a labelled
99+
[RECOMMENDATION] with its own falsifier.
100+
101+
Headline: **five probes moved four of the six residue items since Step 1**,
102+
so the ruling should be made against the current state, not #1006's table.
103+
`cop` is SETTLED as not rail-expressible (this PR's C1–C4 — a rail is the
104+
taxonomy, a belief is a claim about it). `stamp` is **provably irreducible
105+
to geometry** (#1009 G3) and is the one genuine mint candidate. `rung`
106+
**conflates two independent axes** (#1011 E3) and both halves are
107+
ELIMINATION candidates, not mint candidates — though the depth-is-derivable
108+
result is one fixture and is flagged as the document's weakest claim.
109+
`contradiction` is the wrong SHAPE, not merely unwired (#1010 F1): a
110+
magnitude cannot express Auslöschung. `truth` is unchanged
111+
(compose-don't-mint); `premises` defers to the open address question.
112+
1113
## 2026-08-23 — TARSKI-MARKOV-HHTL (open-questions register — HELD, not a plan)
2114

3115
`.claude/plans/tarski-markov-hhtl-seam-v1.md`**holds open questions;
Lines changed: 249 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,249 @@
1+
# BELIEF-ABI-RESTORATION-1 — Step 2 ADDENDUM: the copula correction
2+
3+
> Status: ADDENDUM to `.claude/plans/belief-abi-step2-ruling-request-v1.md`,
4+
> operator-directed 2026-08-23. **Proposes NO mint.** Supersedes the ruling
5+
> request's Item 1 recommendation, which is retracted below.
6+
7+
## 1. The retraction
8+
9+
The ruling request's Item 1 recommended:
10+
11+
> ~~`Copula` → relation concept → classid reference~~ **RETRACTED.**
12+
13+
The C1–C4 probe result established only:
14+
15+
```
16+
COPULA ≠ RAIL PLACEMENT
17+
```
18+
19+
It did **not** establish:
20+
21+
```
22+
COPULA = CLASSID
23+
```
24+
25+
Those are unrelated conclusions, and the leap between them was the drift.
26+
Routing relation content through classid would smuggle instance/content
27+
semantics into the reading selector — the exact move the dock/route
28+
separation exists to prevent. The prior "a relation is a class" ruling
29+
addresses how OGAR *mints concepts*; it is not a licence to make `classid`
30+
a semantic payload field on relation rows.
31+
32+
## 2. The law
33+
34+
```
35+
CONTENT NEVER TRAVELS IN CLASSID.
36+
CLASSID SELECTS THE READING.
37+
38+
classid = HOW these bytes may be read
39+
HHTL = WHERE the resident thing lives
40+
mask = WHAT part / group / region conducts
41+
edges = HOW addressed things relate
42+
43+
THE POPULATION DOES NOT MOVE. THE VIEW DOES.
44+
SHARE THE HIERARCHY, NOT NECESSARILY THE PAYLOAD.
45+
AN INDEX OR MASK MAY ACCELERATE THE ABI.
46+
IT MUST NEVER BECOME A SECOND ABI.
47+
MEASURE THE DISTRIBUTION BEFORE BUYING THE REPRESENTATION.
48+
```
49+
50+
Do not mint one classid for Inh, another for Sim, another for Impl. Do not
51+
smuggle predicates, relation identity, group identity, or belief identity
52+
into classid — for copula or for anything else.
53+
54+
## 3. The machinery audit (what already exists for this job)
55+
56+
| machinery | what it is | role in the copula question |
57+
|---|---|---|
58+
| `WideFieldMask` (`class_view.rs`) | up-to-64+ field/group bit selection, `intersect`/`union`/`is_disjoint`, fail-closed `EMPTY` | broad group CLASSIFICATION of terms/rows — cannot carry pairwise topology |
59+
| `RowFocusMask` / `AttentionFocusFacet` | antichain of HHTL regions; `covers`/`common_prefix`/`intersect`/absorbing `union`/conservative `difference` | hierarchical region selection — WHERE a group applies, never WHAT relates to what |
60+
| relation rows (SPO store, `graph/spo/`; `EdgeBlock` per `ClassView::edge_codec_flavor`) | resident many-to-many topology between addressed things | the natural carrier for arbitrary relation topology |
61+
| `spo::truth::TruthValue` | per-edge (frequency, confidence), revision shipped | truth rides the CLAIM row, never the group |
62+
| CE64 / band readings | causal topology + reasoning lens registers | orthogonal planes; not copula carriers |
63+
64+
The rigid distinction, kept: **HHTL/V3 mask = hierarchical region
65+
selection; `WideFieldMask` = broad field/group selection; relation rows =
66+
arbitrary many-to-many topology.** A non-hierarchical relation is never
67+
forced into HHTL ancestry merely because HHTL is available.
68+
69+
## 4. What was measured (`PROBE-COPULA-GROUP-MASK-1`, 9/9)
70+
71+
Corpus: arena-closure output (Inh chain 1→5, 10 rows after closure) plus
72+
hand rows for Sim (cross-subtree), Impl, and two Rel verbs — 14 rows over
73+
8 terms across two HHTL subtrees.
74+
75+
**Distribution (G-DIST):** Inh=10, Sim=1, Impl=1, Rel=2; max fan-out 5,
76+
max fan-in 4; occupancy 14/256 possible cells = **5.5% — sparse**.
77+
*Fixture bias, stated:* closure only derives Inh/Sim, so this corpus is
78+
Inh-dominated. The KJV right-corner corpus (Rel-heavy) and tactics output
79+
(Impl) are the distributions a workload-scale measurement still needs.
80+
81+
**The Active-Directory shape holds** on shipped operators:
82+
83+
- **G-F1** — every copula reconstructs EXACTLY from resident row content
84+
`(tag, verb)`; the group reading is lossy BY DESIGN (`Rel(7)` and
85+
`Rel(12)` share one group, stay distinct copulas). Content lives in the
86+
row; the group is ergonomics.
87+
- **G-F2**`members(g)` and `memberOf(row)` are inverse VIEWS over one
88+
relation; the four group views partition all rows; resident bytes are
89+
untouched by both lookups. No duplicated canonical state.
90+
- **G-F4** — the cross-subtree Sim pair (0x40.\* ↔ 0x50.\*): neither
91+
address covers the other; the relation exists ONLY as a row. The
92+
hierarchy homes both ends; it does not pretend to BE the relation.
93+
- **G-F5** — ONE classid across all rows while four copulas differ;
94+
reconstruction never reads a classid.
95+
- **G-F6** — a 4-group and a 2-group reading coexist over the same bytes;
96+
insertion leaves prior rows byte-identical. Regrouping is view-only.
97+
- **G-F8** — reclassifying a row's group leaves its truth and stamp
98+
untouched: truth/provenance are properties of the CLAIM, never the
99+
classification.
100+
- **G-COMPOSE**`group ∩ HHTL region ∩ truth-condition` runs as chained
101+
predicates over borrowed rows in one pass; nothing materialized. The
102+
"brutal mask" composition works.
103+
104+
**G-F10 — the cost comparison, with its honest surprise.** At THIS
105+
fixture's scale the dense per-group t×t bitmap (32 B) is *cheaper* than
106+
sparse rows (784 B) — because t=8 is tiny. The scaling arithmetic inverts
107+
hard: dense grows as `groups × t²/8` (t=10⁴ ⇒ ~50 MB per group family
108+
regardless of content), sparse rows grow with actual relations. **Which
109+
wins is a property of the measured workload, not of the design** — which
110+
is exactly why the law says measure first. On these fixture numbers the
111+
addendum buys NOTHING.
112+
113+
## 5. Falsifier status (operator's F1–F10)
114+
115+
| falsifier | status |
116+
|---|---|
117+
| F1 exact copula reconstruction | **held** (G-F1) |
118+
| F2 member/memberOf without duplicated truth | **held** (G-F2) |
119+
| F3 mask forcing materialization sparse rows avoid | not triggered at fixture scale; re-test at workload scale |
120+
| F4 HHTL faking a many-to-many | **held** (G-F4 — the row carries it) |
121+
| F5 classid carrying content | **held** (G-F5) |
122+
| F6 group updates repacking the population | **held** (G-F6) |
123+
| F7 sidecar becoming a second object universe | not exercised (no sidecar built); fence stands |
124+
| F8 truth/provenance on the group instead of the claim | **held** (G-F8) |
125+
| F9 exact inverse lookup + provenance under grouping | **held** (G-F1+G-F2+G-F8 jointly) |
126+
| F10 mask denser than sparse rows for the workload | **measured both ways at fixture scale**; workload-scale open |
127+
128+
## 6. Up/down inheritance vs relation topology (deliverable point 8)
129+
130+
What CAN ride HHTL inheritance without confusing hierarchy with relation
131+
topology: **applicability and scope** — where a group's classification
132+
applies, where support generalizes (`common_prefix` up), where a falsifier
133+
propagates (`covers` down). What CANNOT: the pairwise relation itself.
134+
The Sim row is the measured witness: its endpoints share only the class
135+
root, and any attempt to express it as ancestry would misplace it. The
136+
hierarchy is shared; the payload is not necessarily.
137+
138+
## 7. V4 / BPE / OGAR-loco (deliverable point 7)
139+
140+
Left as MEASURED ALTERNATIVES, not adopted: a recurring
141+
`group ∩ region ∩ condition → behaviour` selection that survives
142+
falsification is a candidate for a learned routing particle. Per the
143+
standing law they remain addressed views/operators/sidecars over the same
144+
resident ABI — never another population owner. Nothing here builds one;
145+
recurrence has not been measured.
146+
147+
## 7b. ⊘ MEASURED — the three corrections that close this addendum
148+
149+
`PROBE-COPULA-DISTRIBUTION-1` (5/5) ran the two distributions §4 named as
150+
open. All three findings correct THIS document.
151+
152+
### Correction A — the shipped fold this addendum failed to audit
153+
154+
**`nars::facet_fold` (ENTROPY-MILESTONES M26) already carries the copula,
155+
losslessly, in the resident M20 register — with zero classid involvement.**
156+
§8 below listed "sparse relation rows with copula content resident in the
157+
row" as candidate C, a hypothesis to measure. It is not a hypothesis. A
158+
strictly cheaper form is shipped, tested, and green:
159+
160+
```
161+
CStmt {s, cop, p} ⟷ SpoFacet (the 12-byte content-blind register)
162+
rail 0 subject s as (lo, hi)
163+
rail 1 predicate (copula TAG, Rel lo) ← the copula lives HERE
164+
rail 2 object p as (lo, hi)
165+
rail 3 ew_subject (Rel hi, spare) ← Rel's u16 completes here
166+
```
167+
168+
Re-verified over the measured corpus, not trusted from unit tests: **16/16
169+
statements round-trip byte-exact** (D1), including `Rel(u16)` payloads
170+
spanning rails 1+3. Five copulas on one `(s,p)` yield five DISTINCT
171+
registers (D2) — the discriminating information is resident bytes, and
172+
nothing upstream is consulted.
173+
174+
This is the **"consult before you guess" tax** the repo's own CLAUDE.md
175+
warns about, paid in full: *"Proposing a type that already exists is a
176+
30-turn rediscovery tax — check first."* The probe-local `RelRow` in
177+
§4 was exactly that.
178+
179+
### Correction B — "KJV Rel-heavy" was REFUTED, not merely unmeasured
180+
181+
§4 named the KJV corpus as the **Rel-heavy** regime that would contrast
182+
with the Inh-dominated closure fixture. Measured, through the REAL
183+
`stance::stream` producer on REAL KJV Genesis 2–3:
184+
185+
| | Inh | Rel | Impl | Sim |
186+
|---|---|---|---|---|
187+
| closure fixture (§4) | 10 | 2 | 1 | 1 |
188+
| **real KJV narrative** | **13** | **2** | **1** | **0** |
189+
190+
**Inh 6× Rel.** The prediction is refuted. Both corpora now measured lean
191+
the SAME way, so **a Rel-heavy regime is UNDEMONSTRATED, not merely
192+
unmeasured** — a materially different status, and the reason for naming
193+
predictions in advance.
194+
195+
### Correction C — "tactics Impl" was a phantom
196+
197+
`nars::tactics` emits **only `Inh` and `Sim`** (every `Copula::` site
198+
verified). There is no tactics Impl distribution. The real producers are
199+
`nars::stance` (both `Impl` and `Rel(verb)`) and `reason_whole_book`
200+
(`Rel(pid)`).
201+
202+
### Cost, at the measured shape and extended
203+
204+
| representation | at measured shape | at t=10k |
205+
|---|---|---|
206+
| **`facet_fold`** | **0 extra bytes** (relabels an existing register) | **0** |
207+
| §4's `RelRow` | 896 B (n=16) | grows with relations |
208+
| dense 4-group bitmap | 164 B (t=18) | **50 MB** regardless of content |
209+
210+
§4's fixture-scale surprise (dense beating sparse) **inverts** at real term
211+
counts — and both lose to a fold that allocates nothing.
212+
213+
### Still BLOCKED, and not fabricated
214+
215+
The whole-KJV **scale** measurement cannot run: `data/coca/lexicon.tsv`
216+
(Release `coca-codebook-v2`) and `pg10.txt → kjv_spo.tsv` are both absent
217+
by design. A hand-written corpus would be a fabricated measurement, so none
218+
was produced. **Scale stays open; shape is now measured.**
219+
220+
## 8. What Step 2 now asks about `cop` (replacing Item 1's question)
221+
222+
Not *"where do we encode Copula?"* but:
223+
224+
> **What is the cheapest lawful resident relation + selection geometry
225+
> from which Copula is merely an ergonomic reading?**
226+
227+
**The answer is E, and it is already shipped.** Re-graded after §7b:
228+
229+
- **E. existing-tenant composition — `nars::facet_fold``SpoFacet`.**
230+
**RECOMMENDED.** The copula is a 2-bit tag on rail 1 (plus `Rel`'s u16
231+
across rails 1+3) of a 12-byte content-blind register the awareness
232+
plane already holds. Lossless, round-trip-gated, **0 extra bytes**, zero
233+
classid involvement. Verified on the measured corpus (D1/D2), not merely
234+
on its own unit tests.
235+
- **C/D. sparse relation rows (± group masks)** — SUPERSEDED by E for the
236+
copula question. The `PROBE-COPULA-GROUP-MASK-1` results still stand as
237+
the general **many-to-many topology** finding (G-F1/F2/F4/F5/F6/F8 held),
238+
and the group-mask ergonomics remain available as a SELECTION layer over
239+
whatever carries the relation — but the copula itself needs no new row.
240+
- **A/B. masks alone** — REJECTED as sole carriers: classification cannot
241+
carry pairwise topology; HHTL must not fake many-to-many (G-F4).
242+
- **F. a new tenant** — NOT proposed, and now clearly unnecessary.
243+
244+
**What the ruling can now decide, and what it cannot.** SHAPE is measured
245+
and points at E. SCALE is still open (whole-KJV blocked on uncommitted
246+
Release data). Since E allocates nothing and is already shipped, the scale
247+
question does not gate adopting it — it gates only any FUTURE proposal to
248+
replace it. If the ruling accepts E, `cop` leaves the residue list
249+
entirely: it has a home, and that home costs nothing.

0 commit comments

Comments
 (0)