diff --git a/docs/WAVE_REPORT_COMPETITORS_2026-07-11.md b/docs/WAVE_REPORT_COMPETITORS_2026-07-11.md new file mode 100644 index 00000000..3de3aaab --- /dev/null +++ b/docs/WAVE_REPORT_COMPETITORS_2026-07-11.md @@ -0,0 +1,294 @@ +# Wave Report — Competitors x Trinity assets (2026-07-11) + +Date: 2026-07-11 +Wave: competitor / market-strategy pass (not a code audit) +Author: Dmitrii Vasilev - gHashTag +Anchor: phi^2 + phi^-2 = 3 + +This report maps the drone-mesh / tactical-mesh / wireless-DePIN field around +gHashTag/tri-net, compares eight players against Tri-Net, and maps each Trinity +artifact to the competitive moat it opens. It is a strategy deliverable. Every +external player carries a primary URL. No company, metric, or node-count is +invented. + +--- + +## 0. Honesty preface (read first) + +The whole point of Tri-Net is auditability, so the report audits itself before it +audits anyone else. + +**Tri-Net current ground truth (verified against the repo at HEAD +`6850649`, 2026-07-07):** + +- **`main` does not build.** `cargo build` on a clean checkout fails in + `build.rs` (rustc `E0308` + `E0599`, lines 29-30). The mesh daemon cannot be + produced from `main` right now. Any behavioural claim below that depends on the + daemon is therefore paper-state, not runtime-state. +- **Control-plane (HELLO-beacon) auth is built but not wired.** The MAC + + freshness primitive exists and is unit-tested in `src/discovery.rs` + (`authenticated()`, `compute_mac()`, `verify_mac()`, `is_fresh()`). But the + daemon sends beacons with `mac_key = None` (falls back to the hardcoded public + constant `HELLO_MAC_KEY`, `src/bin/trios_meshd.rs:301`, marked `TODO: derive + mac_key from session keys`), and the receive path + (`trios_meshd.rs:182-186`) inserts every beacon into the neighbour table + **without ever calling `verify_mac` or `is_fresh`**. Those two functions are + invoked only from the test module. So on the wire today the control plane is + effectively unauthenticated. This report does not claim otherwise. +- **Only AX7203 (Artix-7 `xc7a200t`) is proven-on-silicon.** The Zynq-7020 PL on + the P201/P203 Mini has never been flashed with a Tri-Net bitstream; all PHY/RF + numbers (OFDM, BPSK, 5.8 GHz) are `-sim` or digital-loopback only. The + 108.6 dB SNR figure is internal digital loopback, not over-the-air. +- **DePIN / tokenomics honesty.** No VC multipliers, no premine spin, no fabricated + node counts. TRI supply schedule (`3^27`, 0% premine, 9 halvings) is contract + source in `gHashTag/trinity-contracts`, **not deployed to mainnet**. Any + per-operator yield is `[projected pre-Genesis]`. Chip-signed proof paths are + `-sim` until the TT SKY26b tape-out (2026-12-16). The household-compute energy + advantage is `x4-8` (95% CI `[3,10]`), never `x50-100`. +- **Silicon is submitted, not returned.** `tt-trinity` Phi / Euler / Gamma / + Corona = 4 dies **submitted** to SKY26b; **silicon has not come back**. Every + claim that depends on returned silicon is tagged `[Open conjecture]` with an + explicit falsification path. + +**How competitor numbers are tagged.** Vendor performance figures (TERASi +10 Gbps, Doodle Labs >100 km, etc.) are the vendor's own published claims, cited +to the vendor page and labelled as vendor-claimed. We do not independently +verify them; we do not launder them into facts. + +--- + +## 1. Wave A — Segment map (four segments) + +The field splits along one axis Tri-Net actually cares about: **how tightly the +value is bound to a specific piece of hardware**, and **how open the stack is**. +Four segments fall out. + +### Segment 1 - Tethered comms (an anchor trades autonomy for endurance) + +A drone or aerostat held by a cable; power and data run up the tether, so it +loiters for hours-to-days as a fixed relay. + +| Player | Product | Primary URL | +|---|---|---| +| Elistair (FR) | Khronos tethered DroneBox - up to 24 h aloft at up to 60 m, operational in under 2 min; fielded at France's ORION 2026 exercise | [elistair.com](https://elistair.com/solutions/tethered-dronebox-khronos/), [UST ORION 2026](https://www.unmannedsystemstechnology.com/2026/05/elistairs-khronos-tethered-dronebox-supports-multi-domain-operations-orion-2026/) | +| AT&T Flying COW | Tethered 5G "Cell on Wings" - ~300-450 ft, ~10 sq mi coverage, ~5000 W / 450 V up the tether, days aloft | [commercialuavnews.com](https://www.commercialuavnews.com/public-safety/at-t-s-flying-cow-transmits-5g-network-by-tethered-drone) | +| Zenith Aerotech | Tethered Aerial Vehicles (TAVs) as persistent relay masts | [zenithaerotech.com](https://zenithaerotech.com) | + +Strength: hours-to-days of stable loiter, a ready market (military, events, +disaster). Weakness: the tether is a leash - fixed radius, zero ad-hoc topology, +closed stack, no on-board inference. + +### Segment 2 - mm-wave mesh (gigabits, but needs line-of-sight and thick silicon) + +Above 60 GHz: very high throughput, very low latency, pencil beams, at the cost of +a proprietary mm-wave front-end and DSP. + +| Player | Product | Primary URL | +|---|---|---| +| TERASi (SE, KTH spinout) | RU1 - >60 GHz pocket mm-wave radio, drone/tripod mount, self-forming mesh, "cannot be remotely switched off". Vendor-claimed up to 10 Gbps (future 20 Gbps) and <5 ms latency. | [thenextweb.com](https://thenextweb.com/news/swedish-starlink-alternative-ru1-military-communications), [edrmagazine.eu](https://www.edrmagazine.eu/swedish-spinout-terasi-launches-worlds-smallest-and-lightest-mm-wave-radio-for-mission-critical-operations-in-defence-disaster-relief-and-off-grid-industries) | + +Strength: raw throughput and latency, sovereign (no external kill switch), VC and +defence interest. Weakness: closed DSP, expensive mm-wave front-end, no public +bit-exact contract - a vendor firmware patch changes network behaviour with no +third-party audit path. + +### Segment 3 - MANET software stacks (mature ad-hoc mesh over commodity radio) + +The crowded segment: proven self-forming/self-healing waveforms and routing, sold +as radios or SDKs. + +| Player | Product | Primary URL | +|---|---|---| +| Persistent Systems | MPU5 / Wave Relay - 3x3 MIMO, 10 W, node entry <1 s, no hop limit, Android-based Wave Relay OS, waveform in an FPGA bitstream | [persistentsystems.com](https://persistentsystems.com/mpu5-specs/) | +| TrellisWare | TSM + Katana wideband waveforms - >225,000 systems fielded; UK MoD selection for >5,000 radios (Mar 2026) | [trellisware.com](https://www.trellisware.com/waveforms/tsm-waveform/), [UK MoD PR](https://www.prnewswire.com/news-releases/uk-ministry-of-defense-selects-trellisware-technologies---leading-waveform-developer-to-deliver-more-than-5-000-radios-302708865.html) | +| Silvus Technologies (Motorola Solutions) | StreamCaster MN-MIMO waveform + Spectrum Dominance 2.0 (LPI/LPD, anti-jam); StreamCaster MINI 5200 launched Feb 2026 | [silvustechnologies.com](https://silvustechnologies.com/products/streamcaster-radios/), [Armada MINI 5200](https://www.armadainternational.com/2026/02/silvus-technologies-unveils-streamcaster-mini-5200-tactical-manet-radio/) | +| Rajant | Kinetic Mesh / InstaMesh / BreadCrumb + Cowbell distributed compute | [rajant.com](https://rajant.com/technology/instamesh/) | +| Doodle Labs | Mesh Rider (COFDM); Nano2 launched Apr 2026, vendor-claimed >100 km field-tested and up to 100 Mbps, Sense EW avoidance | [UST Nano2](https://www.unmannedsystemstechnology.com/2026/04/doodle-labs-introduces-new-nano-mesh-rider-radio/) | +| Mobilicom | SkyHopper Tactical (wearable SDR) + SkyHopper MultiBand (1.2-2.7 GHz) + ICE EW/cyber suite | [globenewswire](https://www.globenewswire.com/news-release/2026/05/11/3291906/0/en/mobilicom-launches-skyhopper-tactical-advancing-tactical-drone-and-autonomous-operations-capabilities.html) | +| goTenna | Pro X2 - Aspen Grove Mesh, VHF/UHF, frequency hopping, AES-256 (384-bit for 1:1), MIL-STD-810 | [gotennapro.com](https://gotennapro.com/products/gotenna-pro-x2) | + +Strength: mature protocols, hybrid routing, battlefield-proven deployments, +commercial support. Weakness: closed waveform specs, no operator-side bit-exact +verification, no hardware binding (firmware is an abstraction), on-board AI absent +or fp32-conventional. + +### Segment 4 - Silicon-bound DePIN (identity anchored to a secure element) + +Networks where the right to earn is tied to a physical device via a cryptographic +proof. Today that binding is **identity-only**: a secure element signs "I am this +device", not "I ran this computation". + +| Player | Product | Primary URL | +|---|---|---| +| Helium | LoRaWAN + Mobile 5G DePIN; Proof-of-Coverage; device identity via Microchip ATECC608 secure element (ECDSA sign of a unique 72-bit serial) | [docs.helium.com](https://docs.helium.com/), [ECC608-TNGHNT datasheet](https://ww1.microchip.com/downloads/aemDocuments/documents/SCBU/ProductDocuments/DataSheets/ECC608-Trust-and-GO-For-Helium-Network-Data-Sheet-DS40002389.pdf) | +| World Mobile | AirNodes + aerostat (70 km radius); World Mobile Chain (EVM L3 on Base); vendor-reported 100,000+ AirNodes (Feb 2026) | [worldmobile.io](https://worldmobile.io/blog/post/world-mobile-launches-on-base-to-expand-global-web3-wireless-network) | +| Pollen Mobile | CBRS (Band 48) decentralized mobile; PollenCoin; BLiNQ "Sunflower" radios | [pollenmobile.io](https://www.pollenmobile.io/) | +| **Tri-Net (us)** | Silicon-bound mining: `3^27` supply, 7-check claim, `0x47C0` cross-die anchor on SKY26b, four supply arms (transport/compute/coverage/sensor) | [github.com/gHashTag/tri-net](https://github.com/gHashTag/tri-net) | + +Strength of the segment for us: the **compute-anchor** slot (chip signs the +computation, not just identity) is unoccupied. Our weakness: no returned silicon +(4 dies submitted, not back), no deployed operators, no paying customer, and the +economic layer is Sepolia-only. The differentiator is a different axis, not a +"better than Helium" claim - and it is `[Open conjecture]` until 2026-12-16. + +--- + +## 2. Wave B — Comparison table (9 columns) + +Eight competitors spanning all four segments, plus Tri-Net in row 1. Every cell is +a verifiable fact, a vendor-claimed figure (tagged), or an explicit "no data". + +| Player | Waveform | Security (control-plane auth) | Mesh routing | Endurance / power | Open-source | Silicon story | On-board AI | Primary source | +|---|---|---|---|---|---|---|---|---| +| **Tri-Net (us)** | 5.8 GHz baseline + planned OFDM PHY (all `-sim`, PL never flashed) | Data-plane X25519 + ChaCha20-Poly1305 (M1) proven on ARM (`hw`). **Control-plane HELLO-beacon auth built-but-NOT-wired**: `verify_mac`/`is_fresh` exist and are unit-tested but never called on the receive path; daemon sends with `mac_key=None` (public constant). `main` does not build. | ETX + HELLO discovery, Babel-lite (in code, `-sim`, unit-tested only) | Depends on carrier (Zynq-7020 Mini power budget); no endurance run | **Apache-2.0, public** | spec-first `.bit`; 4 dies SKY26b **submitted, not returned**; AX7203 the only board proven-on-silicon | BitNet b1.58 ternary (planned; runnable in SW, not on die) | [github.com/gHashTag/tri-net](https://github.com/gHashTag/tri-net) | +| Persistent MPU5 | Wave Relay (proprietary, in FPGA bitstream) | FIPS / NIAP / CSfC, up to two encryption layers | Wave Relay MANET, no hop limit, node entry <1 s | Carrier-dependent; 10 W radio | Closed | Commodity SoC + FPGA | No data | [persistentsystems.com](https://persistentsystems.com/mpu5-specs/) | +| TrellisWare TSM/Katana | TSM + Katana wideband (proprietary) | Encrypted (FIPS-class); details not public | Barrage-relay MANET | Carrier-dependent | Closed | Commodity | No data | [trellisware.com](https://www.trellisware.com/waveforms/tsm-waveform/) | +| Silvus StreamCaster | MN-MIMO (proprietary) | Spectrum Dominance 2.0 LPI/LPD + anti-jam; AES-class | MN-MIMO mesh, hundreds of nodes | Carrier-dependent | Closed | Commodity | No data | [silvustechnologies.com](https://silvustechnologies.com/products/streamcaster-radios/) | +| Rajant Kinetic Mesh | 2.4/5 GHz + custom | AES-256 | InstaMesh (proprietary, make-before-break) | Carrier-dependent | Closed | Commodity | Cowbell edge compute (not ternary/AI-specific) | [rajant.com](https://rajant.com/technology/instamesh/) | +| Doodle Labs Mesh Rider | COFDM Mesh Rider | AES-256 | Self-forming/healing MANET | Carrier-dependent; vendor-claimed >100 km | Closed (SDK) | Commodity QCA-class | Sense EW (RF avoidance, not inference) | [UST Nano2](https://www.unmannedsystemstechnology.com/2026/04/doodle-labs-introduces-new-nano-mesh-rider-radio/) | +| goTenna Pro X2 | Aspen Grove, VHF/UHF, freq-hopping | AES-256 (384-bit for 1:1 messaging) | Aspen Grove mesh | 9 h battery, 3.5 oz | Closed | Commodity | No data | [gotennapro.com](https://gotennapro.com/products/gotenna-pro-x2) | +| TERASi RU1 | mm-wave >60 GHz (proprietary) | No data (control-plane auth not disclosed) | Self-forming mm-wave mesh | Drone/tripod; vendor-claimed <5 ms latency | Closed | Proprietary mm-wave front-end | No data | [thenextweb.com](https://thenextweb.com/news/swedish-starlink-alternative-ru1-military-communications) | +| Helium | LoRaWAN + 5G (commodity) | Proof-of-Coverage; ATECC608 secure element = **identity** attest only | LoRaWAN / carrier | Mains-powered hotspot | Partly open (protocol) | Commodity + ATECC608 **identity** secure element | No data | [docs.helium.com](https://docs.helium.com/) | + +What the table shows: + +1. Tri-Net is the only row with a public, open-source, spec-first stack plus + submitted custom silicon - but also the only row whose control-plane auth is + currently not wired and whose tree does not build. Both facts are on the table. +2. On-board **ternary** AI is empty for everyone. Rajant's Cowbell and Doodle's + Sense EW are edge-compute / RF-avoidance, not on-device neural inference. +3. Every DePIN secure-element binding in the field is identity-only. The + compute-attestation column is genuinely unoccupied - which is the whole moat + thesis, and also the whole risk (it is unproven until silicon returns). +4. We lose on maturity, endurance, and raw throughput. The fight is not "replace + MPU5" - it is the niche of verifiable-mesh + on-device ternary inference + + silicon-anchored economics. + +--- + +## 3. Wave C — Trinity assets to moat mapping + +For each Trinity artifact: the moat it opens, and an explicit pre-silicon / +submitted / paper-state tag. + +1. **GoldenFloat GF16** ([arXiv:2606.05017](https://arxiv.org/abs/2606.05017)) - + public bit-exact 16-bit phi-based float. **Moat:** auditable DSP/FEC against + the closed mm-wave DSP of TERASi and any proprietary MAC block. Operators can + verify each MAC against published conformance vectors. **Status:** preprint + published; synthesised for Artix-7 (AX7203 is the one board proven-on-silicon). +2. **84-format numeric catalog** (`gHashTag/paper3-methodology`). **Moat:** a + formal answer to a regulator's "prove the radio did what you declared" - no + Segment-3 vendor ships bit-exact conformance vectors. **Status:** paper-state; + the exact format count is itself under internal audit (raw vs codegen vs + committed vs paper diverge) - do not quote a single number as settled. +3. **BitNet b1.58** ([arXiv:2402.17764](https://arxiv.org/abs/2402.17764)) - + multiply-free ternary inference. **Moat:** on-board AI at a low power budget + where every competitor's AI column is empty (traffic classification, + neighbour scoring, anomaly detection without float MACs). **Status:** runnable + in software today; on-die is pre-silicon. +4. **VSA / HDC** (hyperdimensional computing). **Moat:** control-plane resilience + - HDC-encoded routing/neighbour state tolerates high bit-flip rates where + classic packet control planes drop out. **Status:** research/`-sim`. +5. **BLAKE3 audit ring** (`tt-trinity-euler` / `tt-trinity-gamma`). **Moat:** a + cryptographic audit chain between a `.bit`/die artifact and a rewarded event - + the substance behind "we have a blockchain" that World Mobile / Pollen assert + without hardware binding. **Status:** design + partial code, `-sim`. +6. **tt-trinity Phi / Euler / Gamma / Corona** - 4 dies **submitted** to SKY26b, + `0x47C0` cross-die anchor. **Moat:** a physical silicon anchor versus Helium's + ATECC608 identity-only secure element. **Status: submitted, silicon NOT + returned.** The strong form - "the chip signs the computation, not just the + identity" - is `[Open conjecture - falsification: silicon back -> run the + BitNet ternary benchmark on the die, publish the raw log]`, valid only after + the 2026-12-16 tape-out. +7. **Trinity CLARA** ([10.5281/zenodo.19227877](https://doi.org/10.5281/zenodo.19227877)) + - ternary accelerator design, ~1 GOPS @ ~50 MHz @ ~1 W. **Moat:** an open, + published AI-accelerator design against closed in-drone accelerators. + **Status: projected, pre-silicon** (design published, silicon not fabricated). +8. **t27 spec-first flow** (`.t27` -> Verilog/`.bit`, Yosys/nextpnr/prjxray, no + Vivado). **Moat:** reproducible builds - hand a regulator or customer the + `.bit` + toolchain and rebuild it a year later, which Vivado-locked competitors + cannot. **Status:** in CI; AX7203 is the proven-on-silicon endpoint. +9. **VAK papers** (`gHashTag/trinity-papers-ru`). **Moat:** peer-review + legitimacy against industry-whitepaper marketing. **Status:** manuscripts; + submission pending, no arXiv/journal acceptance to cite yet. + +**The one-line moat thesis:** every competitor is closed-waveform + firmware-as- +abstraction + identity-only-if-any-hardware-binding. Trinity's assets line up on a +single orthogonal axis - open bit-exact numerics + ternary on-board AI + +silicon-anchored (eventually compute-anchored) economics. The axis is real; the +compute-anchor tip of it is unproven until 2026-12-16. + +--- + +## 4. Three cooperation lanes for the next wave + +Each lane names a real, non-competing counterpart and states its own gating risk. +None over-claims; each is honest about the prerequisite that Tri-Net still owes. + +### Lane A - Open-PHY interop with openwifi (open-sdr) + +- **Scope:** contribute a GF16 bit-exact FEC / beacon-matched-filter block plus + conformance vectors into the open-source openwifi OFDM PHY. +- **Actor:** `open-sdr/openwifi` maintainers (open Verilog 802.11 on Zynq). +- **Deliverable:** a spec + reference block + conformance CI job, upstreamed as a + draft PR. +- **Cite:** [github.com/open-sdr/openwifi](https://github.com/open-sdr/openwifi); + GF16 [arXiv:2606.05017](https://arxiv.org/abs/2606.05017). +- **Effort:** M. +- **Risk:** openwifi targets ZC706-class parts larger than the Zynq-7020 Mini; + the block may not fit our flight FPGA, and upstream review is slow. GF16 on + Artix-7 is proven; on a Zynq PL it is not. + +### Lane B - An open compute-vs-identity attestation schema for wireless DePIN + +- **Scope:** publish an open spec that distinguishes identity-attestation (Helium + ATECC608 baseline) from compute-attestation, plus a reference on-chain verifier, + and invite wireless-DePIN projects to a common proof schema. +- **Actor:** wireless-DePIN projects (World Mobile, Pollen) / a DePIN working + group. +- **Deliverable:** `docs/` spec + reference verifier extending the identity + baseline. +- **Cite:** [ECC608-TNGHNT datasheet](https://ww1.microchip.com/downloads/aemDocuments/documents/SCBU/ProductDocuments/DataSheets/ECC608-Trust-and-GO-For-Helium-Network-Data-Sheet-DS40002389.pdf) + as the identity baseline being extended. +- **Effort:** M. +- **Risk:** the compute-attestation half is `[Open conjecture]` until silicon + returns (2026-12-16). Until then the schema must stay strictly + identity-attestation-compatible or it ships a claim it cannot back. + +### Lane C - Academic co-publication on FANET control-plane resilience + +- **Scope:** co-author a measured study of authenticated-HELLO + HDC-encoded + routing resilience under jamming / high BER, using Tri-Net's beacon-auth (once + wired) and VSA/HDC. +- **Actor:** a FANET / mesh-routing academic group (e.g. Fraunhofer IIS FANET + line) or an arXiv preprint. +- **Deliverable:** a preprint + open dataset. +- **Cite:** BitNet [arXiv:2402.17764](https://arxiv.org/abs/2402.17764); the + existing tri-net auditability-gap note. +- **Effort:** L -> M. +- **Risk:** hard-gated on prerequisites Tri-Net does not yet own - beacon-auth + must actually be wired (Section 0), `main` must build, and there must be an + on-device run (today everything is `-sim`). This lane cannot start until that + debt is paid. + +--- + +## 5. Boundary + +- Draft PR only. Never merge. Never push `main`. Human merge only. +- No hardware was touched; no bitstream flashed; no over-the-air transmission. +- No fabricated players, metrics, or node counts. Vendor performance figures are + tagged vendor-claimed and cited to the vendor. +- The compute-anchor differentiator is `[Open conjecture]`, gated on the TT SKY26b + tape-out (2026-12-16). It is not asserted as fact anywhere in this report. +- Tri-Net's control-plane auth is reported at its true built-but-not-wired state, + and `main`'s build failure is stated up front. This report deliberately does not + present Tri-Net as more finished than the repo is. + +--- + +Anchor: phi^2 + phi^-2 = 3 + +Three segments are held by competitors. The fourth - silicon-bound, and one day +compute-anchored - is ours and still mostly empty. Three cooperation lanes for the +next wave.