TapeOut, up close
The public TapeOut protocol in enough detail to check the TapeID mapping against: the mechanism byte by byte, the published scoring rule, and the ecosystem numbers, each with its source and the date it was read. None of it is a TapeID achievement.
The mechanism
TapeOut puts logic on chain. Transistors of two kinds — NAND and LATCH — are laid out as a circuit, then taped out: the transistors are burned and the circuit becomes an immutable on-chain object with an id. A circuit is a small pure function — thousands of gates, not millions — and two read-only calls define everything a third party can do with it:
eval(uint256 id, bytes inputs) view returns (bytes) free to call, no wallet needed netlist(uint256 id) view returns (bytes) the bytes exactly as taped out
The netlist format is public, so the digest a record commits to is over bytes anyone can fetch and re-read the same way:
netlist records, big-endian:
NAND = 0x00 . u24 a . u24 b two input signals
LATCH = 0x01 . u24 d one input, clocked
REF = 0x02 ... calls another circuit;
a REF anywhere sets mining weight to 0
signal order: 0 = constant 0, 1 = constant 1, then the inputs,
then one signal per gate. A gate reads only lower-numbered
signals. The last nOut signals are the outputs.
input packing: pin i is bit i of the inputs word
(documented for up to 8 pins).Two limits shape real designs: a per-circuit size cap — about 25,000 gates by community figures, the exact contract cap unconfirmed — and a 12,000,000 gas ceiling per task run, read from the task registry.
How a design is scored
The task bank holds 306 published tasks, 267 of them on chain — 222 combinational, 84 sequential, 32 classed as trivial, with 15,381 reference NAND and 340 reference LATCH across the bank. The scoring rule below is the community-published reading of the mining formula; it has not been confirmed against a contract by us. The TapeID mapping does not depend on its details — only on the fact that the rule is published and reproducible:
b* = NAND count + LATCH count both weigh 1
A = g + 6s area: a latch costs six NAND
d = longest combinational path depth
C = A x max(d, 1)^3 cost: depth is punished cubically
q = clamp(C_ref / C, 0.25, 4) against the task's frozen reference
H = (b* + K_task x q) x P full premium only for the current
best-known implementation- Depth is cubed, so a shallow circuit scores well and a deep one is expensive.
- A latch costs six NAND of area and can deepen the path.
- The design premium pays the current best-known implementation of a task only; a verified-but-not-best circuit earns the base rate.
- Emissions split about 99% to the verified pool and 1% to the unverified pool, so a circuit that fails verification earns almost nothing.
The ecosystem in numbers
Read 2026-10-16 from the public research API and registry mirrors of the BNB Chain deployment. The “source” column says who published each figure; the reading is ours.
| Fact | Number | Source |
|---|---|---|
| Processors | 1,267 | registry |
| Processors that can mine | 2 | registry — 99.8% of what gets built earns nothing |
| Circuits taped out | 59,403 | registry — supply is not scarce |
| Circuit-market sales, all time | 2,969 | protocol count |
| Unique buyers, all time | 36 | registry — against 2,492 unique sellers |
| Miners | 20,013 | registry — but about 11 active creators on a given day |
| Largest single wallet | 1,026 circuits across 4 tasks | registry — farming, not designing |
| Verification pass rate | 96.7% of miners | registry — acceptance is close to a formality |
| Official transistors | 1,000,000 at 0.0005 BNB list — sold out | registry — the primary mint is closed |
What this means for an identity layer
Three of these numbers are the reason TapeID exists as a design:
- 96.7% pass rate. Acceptance alone does not filter. What is left to check is the author — which name did the work, and what else it has delivered.
- 36 buyers against 2,492 sellers. A buyer who cannot check a seller's history has nothing to price but the asking amount.
- 1,026 circuits across 4 tasks from one wallet. Volume is anonymous today. A named, anchored history is what a farm cannot produce cheaply, because each entry in it costs real work under a published rule.
This is also the limit of the claim: none of these numbers is TapeID's, TapeID does not fix any of them today, and the mapping that would let a record carry this kind of work is a format, not a pipeline.
Sources
- Public research API at
tapeout.work—/api/v1/processors,/api/v1/bem/tasks,/api/v1/bem/overview,/api/v1/bem/leaderboard,/api/v1/market-overview,/api/v1/daily-activity,/api/v1/protocol-pulse— read 2026-10-16. - Interface and netlist format: the community
tapemktrepository (README,NETLIST-FORMAT.md), including its X Layer deployment notes. - Official site data:
tapeout.net/processors.jsonandtapeout.net/pod/pod-taskbank.json. - The mining formula and the roughly 25,000-gate cap are community-published and are marked as such above.