zheng roadmap
what is built is specified in specs and measured in audit. this page holds the two things that are not: the status of the repair, phase by phase, and the open proposals. the design record is proof-system repair; its gates are fixtures and commands, so a phase below is closed only when its gate is green.
status of the repair (master, 2026-10-27)
the crate is 0.4.0 on master, not published (crates.io stops at 0.3.1, the retired fold). every figure below was measured on hemera 0.3.1, which the crate still pins; the frozen material is hemera 0.4, and moving zheng to it is the in-flight sweep below.
| phase (§5) |
what |
in master |
open |
| 0 |
relation shrink: linear forms, native constants, degree-7 S-box |
done (zheng#45): hash.tri certificate 294,861 → 15,608 B (audit/compact-relation-2026-10-09.md) |
— |
| 1 |
soundness floor |
public prefix and pinned constant placed by the verifier, vk digest absorbed, Fp3 challenges, one envelope ZHENGPF1 with a profile byte, the soundness ledger; fixtures forged-io, forged-program, zeroed-constant, meaningless-witness rejected (rs/tests/attack_fixtures.rs) |
the ledger still has conjectured rows (relation compiler, hash, IVC and wrap composition, program binding by digest), so the "no conjectured row" gate is not passed |
| 2 |
PCS bake-off |
WHIR ships in the succinct profile (envelope profile 1); TensorRs lost and stays decodable until phase 5 (audit/succinct-profile-2026-10.md); bit-flip scans 0 accepted |
small class: hash.tri 16,756 B meets the ≤ 20 KB gate, misses the ≤ 16 KB stretch; verify ≤ 1 ms missed (< 10 ms, soft3 audit/proof-latency-2026-10.md); large class: 71,081 B at 2^20, over 64 KB |
| 2h |
the hash row |
decided outside zheng: hemera 0.4 (hemera#15) frozen 2026-10-16, under the public cryptanalysis campaign |
zheng's gadgets, machine and fixtures still implement hemera 0.3.1: the sweep is #69/#73 below |
| 3 |
uniform step relation, accumulation, IVC |
the nox machine as one step relation (profile 4), hash-based accumulation of RS claims and its decider, IVC (profile 5), the wrap chain (profile 6): final proof ≤ 64,213 B over every transcript, 512 statements decided in one proof (accumulate::ccs) |
verify ≤ 1 ms missed (wrap final proof 5.3–5.7 ms on one thread, audit/wrap-fastverify-2026-10.md); the Trident second verifier (verify(verify(π)) across Rust and Trident) does not exist; on hemera 0.4 the 64 KiB bound is the L3b decision below |
| 4 |
zk profile |
veil (profile 2 scheme 2): 63.9 KB for one hash of a secret preimage, honest-verifier statistical zero knowledge in the ROM; MPC-in-the-head (scheme 1) as fallback |
a smaller zk opening (veil-W, #67) is research |
| P |
the prover everywhere |
lens backend trait with Metal and WebGPU backends and the CPU fallback, same bytes on every backend; profile 1 product statements ≤ 1 s on Metal up to Merkle depth 16 (soft3 audit/proof-latency-2026-10.md) |
profile 6 is a ~19 s wrap chain on Metal (audit/ivc-speed-2-2026-10.md), not seconds-class; no phone measured |
| 5 |
delete |
— |
the legacy feature (the 0.3 fold, universal CCS, φ* SpMV, the zheng CLI), TensorRs decoding and legacy formats are still in the tree |
in flight
| PR |
what |
state |
| #69, #73 |
zheng on hemera 0.4: gadgets, node jet, machine and circuit layouts, goldens re-pinned; #73 is #69 under a CI stand (train row L3) |
draft; one failing test, the 64 KiB final-proof bound, which waits for L3b |
| #78, #79 |
L3b, the owner's decision (due 2026-10-29 12:00 UTC): keep the 64 KiB gate with final level 10f:32:5 (#78), or raise it to 72 KiB with 9f:26:5 (#79) |
draft, stacked on #73 |
| #77 |
[profile.dev] opt-level = 3, so cargo test fits the train's stack-zheng gate (G10b) |
the change is already on master (taken with the promotion); the PR stays open only as part of the #73 stack |
| #67 |
veil-W (gate G1 of soft3 proposals/privacy-frozen.md): masked Spartan with a zero-knowledge WHIR opening, SPEND in ~46.5 KB, measured on hemera 0.3 |
draft, research; needs an independent review and a re-measure on hemera 0.4 |
| #70 |
block aggregation (gate G2): one proof per block |
draft, in flight |
proposals
| proposal |
status |
target |
| gpu-prover |
draft — the backends exist (Metal, WebGPU, CPU); the device-resident pipeline does not |
one pipeline from trace to proof without CPU phases |
| ring-aware-fhe |
draft — the ring PCS is specified in lens (ring-pcs); zheng has no ring-structured CCS or FHE jets |
TFHE bootstrapping as ring-structured CCS and dedicated jets |
| gravity-commitment |
draft — to be restated and measured on the shipped RS/WHIR commitment |
mass-weighted polynomial encoding: hot positions open cheaper |
the recursive-Brakedown opening (a constant-size opening that skipped the Merkle authentication of its queries) is not a proposal: it is unsound in a hash-only world and no version brings it back; its record is recursive-brakedown in docs/explanation.
lifecycle
| status |
meaning |
| in reference |
specified in specs/ and implemented |
| accepted |
approved, ready to implement |
| draft |
idea captured, open for discussion |
Homonyms
cyber/roadmap
roadmap the roadmap of cyber is the launch, in phases. the contract, the gates, the property registry and the calendar live on one page, cyber/launch; the engineering order under it lives in the [delivery roadmap](/cyber/roadmap/index). this page is the map between them and the record of what the…
soft3/roadmap
roadmap
soft3/glia/roadmap
glia roadmap done format & import canonical five encodings implemented: `u32` (16.16 fixed-point), `u16` (8.8), `q8` (32-val blocks), `q4` (32-val blocks), `ternary` (2-bit) import pipeline: all weights dequant→f32→canonical at pack time; config stored as integers (eps as 1/ε, sampling per-mille)…
neural/rs/roadmap
Rs development roadmap | Plan | Status | Scope | |---|---|---| | [Verified Bootstrap](/neural/rs/roadmap/verified-bootstrap) | accepted design; all implementation gates open | Reviewed nox seed, Trident interpreter of restricted Rs, own Rs compiler, native self-build and full Rust compatibility |…
neural/trident/roadmap
Design Proposals Proposals for language and VM design changes. Not spec — these are desires documented for future consideration. Each proposal is a standalone markdown file. Status is tracked in the frontmatter. Proposals | proposal | status | what | |----------|--------|------| | noun-types |…
soft3/nox/roadmap
roadmap
soft3/radio/roadmap
radio roadmap radio transmits. it dials, hole-punches, relays through a peer, gossips, and pipes verified streams; it holds no address, no identity, no store, no reconciliation, no format, no frame of its own. the ledger of what belongs to whom is `soft3/roadmap/component-boundaries.md`. | proposal…
soft3/hemera/roadmap
hemera roadmap hemera is specified: x⁻¹ partial S-box, 16 partial rounds, 32-byte output, ~736 constraints per permutation. these proposals are OPTIMIZATIONS on top of the current spec — reducing constraint count further and shifting hemera's role from "hash for everything" to "trust anchor."…
cyb/evy/roadmap
evy roadmap open proposals not yet executed. each is a draft → accepted → migrated-to-specs lifecycle. remaining proposals | proposal | status | what's missing | |----------|--------|----------------| | (none yet — first crates land via `bevy/roadmap/cyb-engine-shardstore` in bbg, which blocks…
soft3/tru/roadmap
roadmap the build plan and open proposals for tru. the current status of each spec is in [specs/](/soft3/tru/specs); this tree holds the forward-looking work. plan [implementation](/soft3/tru/roadmap/implementation) — the milestone plan: module layout, per-spec algorithm, predicate names, and…
cyb/seer/roadmap
roadmap open proposals for seer. the current contracts live in [specs/](/cyb/seer/specs/readme); this tree holds what is designed and not yet built. proposals | proposal | status | target | |---|---|---| | [learned policy](/cyb/seer/roadmap/learned-policy) | proposal | a GFlowNet mining policy that…
cyb/optica/ROADMAP
roadmap optica is a static-site publisher for markdown knowledge graphs. the rendering pipeline (markdown → HTML → templates → CSS + minimal JS) is constrained by the web platform — DOM, Canvas, CSS. that ceiling is acceptable for documents up to ~10⁴ nodes. high-performance interactive graph…
neural/eidos/roadmap
eidos v0.1.0 — first anchor release see also [replace-lean.md](/neural/eidos/roadmap/replace-lean) — the registry ladder E0–E5 (2026-10-27): the stack's assumptions registry is eidos's first real corpus; `Int` and the Goldilocks polynomial library move ahead of docs. Context eidos has solid bones:…
soft3/foculus/roadmap
foculus roadmap open problems, one file each. not phased milestones — specific unsolved questions, each with enough context to pick up independently. every item here originates from [specs/security-at-scale.md](/soft3/foculus/specs/security-at-scale)'s adversarial review, kept as a brief pointer…
warriors/trisha/roadmap
Trisha roadmap The release contract is the complete source → compile → run → prove → verify path, with explicitly selected Neptune transaction and mining operations. Current measurements and gate results live in [audit](/warriors/trisha/audit/readme), not in this specification. The coordinated…
cyb/product/roadmap
cyb roadmap architecture cyb is one robot with four implementations sharing one reference. bevy is the root runtime. all four implementations are accessible from bevy: bevy renders its own native UI (graph, 3D, effects) bevy embeds react via WebView bevy embeds leptos via WebView (WASM) bevy embeds…
soft3/mudra/roadmap
roadmap
cybics/crystal/roadmap
short term bugfixing cyb: clear a lot of technical debt and make it more reliable and usable offchain inference: generative model built from cybergraph skill for openclaw: agents can talk using decentralized social & cybergraph universal hash: agents can mine tokens on cpu learning rewards main…
cyberia/research/oxytocin/roadmap
neural/trident/reference/roadmap
Roadmap Current coordinated source candidate versions are Trident 0.3, Trisha 0.3 and Joy 0.5 with compiler API 3. Historical milestones below retain their original versions. Registry installation is not validated for this candidate; use the coordinated source archive while the [registry packaging…
cyberia/projects/oxytocin/roadmap
roadmap oxytocin opens in four stages. each stage is a complete, bookable experience before the next is built — protecting capital and letting each layer prove itself before the next is committed. stage-1 — the body the thermal anchor. the smallest operation that delivers a complete experience and…