signals-key-2026-10-17.md

signals key — dimension version 3 rule 7, launch row 9

Date: 2026-10-17. Scope: bbg launch/9-signals-key-v3 on origin/master 1db1db6 (dimension version 3). Re-implements the fix proposed in bbg#32 (branch launch/9-signal-record-key, written against the flat-commitment state before #43) on the keyed-trie state; bbg#31 holds the original demonstration (refs/pull/31/head). Decision: cyber/launch.md, 2026-10-17 — everything may change until 2026-11-02, the signals key lands in dimension version 3. Machine: M4 Max, shared with other agents' builds.

the defect

BbgState::signals was Keyed<u64, SignalRecord> keyed by the step a caller passed in:

Cybergraph::commit_signal passed the neuron's own signal-chain step. Every neuron's first signal is step 0, so two neurons' first signals shared one record; the later insert replaced the earlier one, and which record survived (and the signals leaf, and the root) depended on arrival order.
the native transition path (prepare_native_batch, cybergraph's durable node) passed a node-global acceptance counter. No record was lost, but the record set depended on the order this node accepted signals in, so two nodes holding the same signals held different signals dimensions.

Launch row 9 requires nodes to converge from different starting states; both keyings break it.

the rule

specs/state.md, dimension version 3, rule 7:

a signal record lives at signal_key(neuron, step) = H(neuron ‖ step) — hemera over the 32-byte neuron id and the step as 8 little-endian bytes (state::signal_key). The 40-byte preimage cannot coincide with the 64-byte preimages of axon_id / balance_key or the 72-byte intent key;
SignalRecord is [neuron, network, link_count, block_height, proof_hash, step] — step appended, so no earlier field offset moves (the primary cell of a signals entry stays at value offset 16, link_count).

Widths: trie leaf values and flat-table entries 28 → 30 limbs; native record (domain 10) 108 → 116 bytes, key 10 ‖ signal_key (was 10 ‖ position_be). TABLE_VERSION stays 2 (record-agnostic layout, rule 6); DIMENSION_VERSION stays 3.

API: apply_signal_record(record) -> key; prove_signal(state, neuron, step), prove_signal_key(state, key); bbg_query(Dim::Signals, key) takes the signal key; NativeChange::Signal { signal, header } and prepare_native(signal, Option<SignalRecord>, intent), a repeated (neuron, step) is Error::SignalExists (was PositionExists). The query authenticator no longer treats signals as a numeric dimension. CLI: bbg prove signals <neuron-hex>:<step>.

evidence

rs/tests/signal_key.rs:

test	property
`signal_key_binds_neuron_and_step`	key = hemera(neuron ‖ step_le); neurons and steps separate
`two_neurons_first_signals_both_persist`	both step-0 records survive, under v1 and v2
`arrival_order_does_not_move_the_root`	8 neurons × 6 steps (8 records per step value), 4 shuffles + reverse: identical root and signals leaf, v1 and v2
`every_record_is_under_the_root_and_provable`	trie path per record under the signals leaf; v1 entity read by (neuron, step) authenticates only that key; the step is under the root
`native_batch_keeps_both_step_zero_signals_and_rejects_a_repeat`	two native records in one batch, value carries the step; a dropped preparation restores; a repeat is `SignalExists`, state untouched; the next step is accepted
`replay_is_deterministic_and_recovery_matches_every_stored_root`	40 native batches over 8 interleaved chains: no record lost, the persisted change stream equals the state's records, a replay from genesis reproduces every snapshot, every stored root `time[h]` and the record stream

Also: root_vectors.rs computes the key from hemera directly in its independent reference derivation (all-dimension vector now has two step-0 records), and recovery_replay_reproduces_every_stored_root writes a signal record per signal (six chains from step 0; asserts no record lost).

Mutation check: with the neuron removed from the key preimage (the pre-fix collision), all six signal_key.rs tests fail and so do all_dimension_roots_match_the_reference and recovery_replay_reproduces_every_stored_root.

Gates (siblings: hemera origin/main fd55f99, lens origin/release/0.2 b1040c2, nox origin/release/0.4 12adaf8, zheng origin/release/0.4 8f20262, strata origin/main 12a7a6d, neuron origin/main 9df467d, tade origin/main 778dc6f):

rs/: cargo test --locked exit 0 (lib 81, every suite green), and cargo test --locked --all-features exit 0 (lib 108); zero warning: lines.
cli/: cargo test --locked exit 0, 16 passed.

roots

No version-3 genesis state carries a signal record: soft3's chain kernel (chain/bbg_state.rs) and genesis builder (genesis/) never write the signals dimension, so the empty signals trie (EMPTY) and every blessed root are unchanged. Rebuilt in the soft3 stand on this branch (hemera release/0.4 3cdaa7f; recorded in soft3 terms in cyber/launch.md row 14):

chain	state root	block-0 id	file
pussy	`a360798987fb821f3bb7e07c66947e91e1eacb4278a834972c32ea8b3054b8b2` (unchanged)	`bcb1e9b0…` (unchanged)	byte-identical, sha256 `ec7aee79…`; 69 MB peak
bostrom	`c50a6d8438ed8b0ef72450e7f83cab81e78bfd4e833106add9d62a0831bcdf13` (unchanged)	`9ab9fc49…` (unchanged)	byte-identical, sha256 `ffa3d6dd…`; 88 s wall, 107 s user, 7.3 GB peak

Both files compared with ~/.cache/cyber/genesis/ (built 2026-10-16): equal sha256, equal genesis particle, state root and block-0 id; only root_seconds in the side-car JSON differs. soft3 conformance check chain --genesis ~/.cache/cyber/genesis/pussy.tade: 30 entries, no drift. soft3 cargo test --locked --release --no-default-features --features chain-conformance in the stand: lib 252 passed, all suites green; one pre-existing warning (deprecated GenericArray::as_slice, chain/conformance.rs).

What does move: any state holding signal records — cybergraph's link/seal path and its durable native node (cybergraph launch/9-signals-key-v3), and soft3's stack conformance vector cybergraph::link@v1 "neuron-7/two-signals/bbg-root" (conformance/snapshots/mechanism.snap, blessed 2026-10-10 on hemera 0.3 and pre-v3 bbg; already due for re-bless with S2, not re-blessed here).