Skip to content

Commit eae4321

Browse files
committed
docs(blackboard): record HHTL fork-ladder decision + loose ends
Per the ndarray agent protocol (update blackboard after completing work): log the entropy_ladder fork-ladder addition, the four-vocabulary unification, the honest [S] joints (real CoarseResidue feed, orthogonality arbitration, Jirak σ), and that the driver-side wire merges with lance-graph materialize's from_live step. Co-Authored-By: Claude <noreply@anthropic.com>
1 parent fbdf6f5 commit eae4321

1 file changed

Lines changed: 37 additions & 0 deletions

File tree

.claude/blackboard.md

Lines changed: 37 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3,6 +3,43 @@
33
> **Read this first.** The "Polyglot Notebook" architecture below is a
44
> separate/older program, not the current epoch.
55
6+
## 2026-06-17 — DECISION: HHTL fork ladder coded in `hpc::entropy_ladder` (CONJECTURE)
7+
8+
Reified the operator's standing idea — *if the orthogonal (helix/CAM-PQ)
9+
leaf residue is strong enough, free energy forks into another domain
10+
(HHTL shift = new exploration)* — as pure functions beside the existing
11+
entropy/quadrant code. Unifies four vocabularies as one 2-axis structure:
12+
`entropy_ladder::Quadrant(entropy,energy)` ≡ `lance-graph-contract::mul::
13+
FlowState(challenge,skill)` (Csikszentmihalyi) ≡ Friston model-vs-surprise
14+
≡ Staunen↔Wisdom.
15+
16+
- `residue_surprise(mag, noise_floor, sigma_k) → [0,1]` — orthogonal residue
17+
magnitude (prediction error the in-domain centroid codebook fails to
18+
explain) → challenge axis. Below floor = quantization (≈0); linear ramp
19+
over `sigma_k·noise_floor`. Threshold provenance per `I-NOISE-FLOOR-JIRAK`
20+
(Berry-Esseen wrong under CAM-PQ weak dependence); ramp is an honest proxy
21+
pending Jirak calibration, **not** a claimed bound.
22+
- `ForkAction {Commit, DescendDeeper, ForkBasin, ForkDomain}` + `fork_decision`
23+
— bands challenge−skill exactly like the shipped `mul::flow_state_from`
24+
(Anxiety >0.2, Flow |δ|<0.15, Boredom <-0.2); HHTL depth decides descend-vs-
25+
fork. `ForkDomain` (mint a new classid = the Friston model-switch) requires
26+
BOTH leaf depth AND challenge≫skill — the "strong enough AT THE LEAF" invariant.
27+
28+
Layering kept honest: the `FlowState` *assessment* stays in lance-graph
29+
(thinking); the fork *math* lives here in ndarray (substrate, where residue +
30+
energy physically are) — per the Architecture Rule. Pure fns + one enum, no
31+
struct/layer, composes with `Quadrant`. 5 lib + 2 doctests, clippy clean.
32+
Branch `claude/jirak-math-theorems-harvest-rfii13`.
33+
34+
**Loose ends (CONJECTURE → gated):** (a) feed the *real* `edge_codec::
35+
CoarseResidue` magnitude from the live codec into `fork_decision` (currently
36+
caller-supplied); (b) `ForkDomain` vs `ForkBasin` should be arbitrated by
37+
residue *orthogonality* (⊥ all in-domain centroids = genuinely new), not the
38+
depth+delta proxy; (c) Jirak-derived σ threshold to replace the `sigma_k`
39+
proxy. This driver-side wire merges with lance-graph `materialize`'s
40+
`ThoughtCtx::from_live` step (same call-site).
41+
42+
643
## 2026-06-10 — DECISION: GUID prefix→shape routing crystallized (docs-only)
744

845
The operator-pinned canonical GUID (`OGAR/CLAUDE.md`: hex dash-groups =

0 commit comments

Comments
 (0)