cone_size int64 6 16 ⌀ | expected stringclasses 2
values | explanation stringclasses 8
values | file stringlengths 7 23 | graph stringlengths 12 1.2k ⌀ | license stringclasses 2
values | n_edges int64 0 78 | n_paths int64 0 10 | n_signals int64 0 22 | observation stringclasses 1
value | paths stringclasses 8
values | reaching_secrets listlengths 0 2 ⌀ | refusal_reason stringclasses 3
values | scored bool 2
classes | secrets listlengths 0 3 | shortest_path_length int64 2 5 ⌀ | source stringlengths 634 4.51k | verdict stringclasses 3
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
6 | CONSTANT_TIME | null | barrett_ct.v | {"a_reg": ["a", "rst", "start"], "cadd": ["Q", "sub"], "cnt": ["W", "rst", "running", "start"], "csub": ["Q", "cadd"], "done": ["W", "cnt", "running"], "prod": ["V", "a_reg"], "r": ["DW", "csub"], "running": ["rst", "start"], "sub": ["a_reg", "tq"], "t": ["K", "prod"], "tq": ["Q", "t"]} | CC-BY-4.0 | 26 | 0 | 19 | done | null | [] | null | true | [
"a"
] | null | // barrett_ct — CONSTANT-TIME Barrett modular reduction for Kyber / ML-KEM (q = 3329). This is the REAL,
// value-correct Barrett reduce (the same construction as the reference `barrett_reduce`), not a placeholder:
//
// t = round(v * a / 2^K) with v = round(2^K / q) = 20159, K = 26 (the canonical Kyber constan... | CONSTANT_TIME |
7 | LEAKY | LEAKY — 'done' depends on a.
How each secret reaches it (shortest path first):
a
├─ acc
└─ done
Each arrow is a dependency edge: an assignment, or a condition guarding
one. The analysis over-approximates, so a path may be unreachable at run
time — read it and decide. That is why it is printed. | barrett_leaky.v | {"acc": ["Q", "a", "rst", "running", "start"], "done": ["Q", "acc", "running"], "r": ["DW", "acc"], "running": ["rst", "start"]} | CC-BY-4.0 | 12 | 1 | 9 | done | [{"secret": "a", "observation": "done", "signals": ["a", "acc", "done"], "length": 2}] | [
"a"
] | null | true | [
"a"
] | 2 | // barrett_leaky — the NON-constant-time counterpart of `barrett_ct`: a naive "reduce by repeated subtraction"
// normalization that LOOPS `acc -= Q` until `acc < Q`. This is a real, tempting implementation (no multiplier,
// tiny area — a common way people "just reduce mod q") but the NUMBER of iterations, hence the c... | LEAKY |
6 | CONSTANT_TIME | null | barrett_buggy.v | {"a_reg": ["a", "rst", "start"], "cadd": ["Q", "sub"], "cnt": ["W", "rst", "running", "start"], "csub": ["Q", "cadd"], "done": ["W", "cnt", "running"], "prod": ["V", "a_reg"], "r": ["DW", "csub"], "running": ["rst", "start"], "sub": ["a_reg", "tq"], "t": ["K", "prod"], "tq": ["Q", "t"]} | CC-BY-4.0 | 26 | 0 | 19 | done | null | [] | null | true | [
"a"
] | null | // barrett_buggy — a BUGGY Barrett reduction (the WS-B teeth): identical to `barrett_ct.v` except the Barrett
// SHIFT is WRONG (K = 25 instead of the canonical 26). The reciprocal constant V = round(2^26/3329) is
// calibrated for a 2^26 shift; using 2^25 scales the quotient estimate `t` by ~2x, an error far outside w... | CONSTANT_TIME |
6 | CONSTANT_TIME | null | ct_cmp.v | {"cnt": ["W", "rst", "running", "start"], "diff": ["W", "cnt", "rst", "running", "start", "xr", "yr"], "done": ["W", "cnt", "running"], "equal": ["diff"], "running": ["rst", "start"], "xr": ["W", "cnt", "rst", "running", "start", "x"], "yr": ["W", "cnt", "rst", "running", "start", "y"]} | CC-BY-4.0 | 29 | 0 | 12 | done | null | [] | null | true | [
"x",
"y"
] | null | // ct_cmp — CONSTANT-TIME equality/compare that ALWAYS scans all W bits (the memcmp-hardening pattern used
// against timing side channels in MAC/tag comparison). It accumulates a difference flag every cycle but never
// exits early, so `done = running & (cnt == W)` is a data-oblivious counter — completion timing is in... | CONSTANT_TIME |
9 | LEAKY | LEAKY — 'done' depends on x, y.
How each secret reaches it (shortest path first):
x
├─ xr
└─ done
y
├─ yr
└─ done
Each arrow is a dependency edge: an assignment, or a condition guarding
one. The analysis over-approximates, so a path may be unreachable at run
time — read it and decide. That is why it is ... | cmp_leaky.v | {"diff": ["rst", "running", "start", "xr", "yr"], "done": ["diff", "running", "xr", "yr"], "equal": ["diff"], "running": ["rst", "start"], "xr": ["diff", "rst", "running", "start", "x", "yr"], "yr": ["diff", "rst", "running", "start", "xr", "y"]} | CC-BY-4.0 | 24 | 10 | 10 | done | [{"secret": "x", "observation": "done", "signals": ["x", "xr", "done"], "length": 2}, {"secret": "x", "observation": "done", "signals": ["x", "xr", "diff", "done"], "length": 3}, {"secret": "x", "observation": "done", "signals": ["x", "xr", "yr", "done"], "length": 3}, {"secret": "x", "observation": "done", "signals": ... | [
"x",
"y"
] | null | true | [
"x",
"y"
] | 2 | // cmp_leaky — the LEAKY counterpart of `ct_cmp`: the classic early-exit memcmp that stops at the first
// differing bit, so `done` timing leaks the position of the first mismatch (a textbook tag/MAC timing oracle).
// Its completion cone contains operand bits, so no operand-free inductive invariant exists and the
// s... | LEAKY |
8 | CONSTANT_TIME | null | ct_div_wide.v | {"a_shift": ["WIDTH", "a", "done_now", "resetn", "running", "start"], "b_reg": ["b", "resetn", "running", "start"], "cnt": ["done_now", "resetn", "running", "start"], "done": ["done_now", "resetn", "running", "start"], "done_now": ["CW", "WIDTH", "cnt", "running"], "ge": ["b_reg", "rem_shifted"], "q": ["done_now", "q_r... | CC-BY-4.0 | 54 | 0 | 19 | done | null | [] | null | true | [
"a",
"b"
] | null | // ct_div_wide — a parameterized CONSTANT-TIME restoring divider with wipe-on-done, in clean RTL.
//
// Purpose (SCALE demonstration): the completion signal `done` is a pure function of the iteration COUNTER
// (cnt == W), never of the operand data — so the divider runs EXACTLY W fixed cycles regardless of a/b, i.e.
//... | CONSTANT_TIME |
16 | LEAKY | LEAKY — 'done' depends on a, b.
How each secret reaches it (shortest path first):
a
├─ a_shift
├─ rem_shifted
├─ rem_next
├─ rem_reg
├─ done_now
└─ done
b
├─ b_reg
├─ rem_next
├─ rem_reg
├─ done_now
└─ done
Each arrow is a dependency edge: an assignment, or a condition guarding
one. The an... | ct_div_leaky.v | {"a_shift": ["WIDTH", "a", "done_now", "resetn", "running", "start"], "b_reg": ["b", "resetn", "running", "start"], "cnt": ["done_now", "resetn", "running", "start"], "done": ["done_now", "resetn", "running", "start"], "done_now": ["CW", "WIDTH", "cnt", "rem_reg", "running"], "ge": ["b_reg", "rem_shifted"], "q": ["done... | CC-BY-4.0 | 55 | 8 | 19 | done | [{"secret": "a", "observation": "done", "signals": ["a", "a_shift", "rem_shifted", "rem_next", "rem_reg", "done_now", "done"], "length": 6}, {"secret": "a", "observation": "done", "signals": ["a", "a_shift", "rem_shifted", "ge", "rem_next", "rem_reg", "done_now", "done"], "length": 7}, {"secret": "a", "observation": "d... | [
"a",
"b"
] | null | true | [
"a",
"b"
] | 5 | // ct_div_leaky — the LEAKY TWIN of ct_div_wide (the teeth for the constant-time scaling proof).
//
// IDENTICAL to ct_div_wide EXCEPT the completion `done_now` also fires on an EARLY-EXIT when the running
// remainder reaches zero (`rem_reg == 0`). That makes the completion cycle DATA-DEPENDENT: for operands whose
// ... | LEAKY |
6 | CONSTANT_TIME | null | ct_gcd.v | {"a": ["K", "a_in", "b", "iter", "rst", "running", "start"], "b": ["K", "a", "b_in", "iter", "rst", "running", "start"], "done": ["K", "iter", "running"], "gcd_out": ["a"], "iter": ["K", "rst", "running", "start"], "running": ["rst", "start"]} | CC-BY-4.0 | 24 | 0 | 11 | done | null | [] | null | true | [
"a_in",
"b_in"
] | null | // ct_gcd — fixed-iteration (constant-time) GCD. `done` is a function ONLY of a cycle counter `iter`
// that advances unconditionally while running, so completion time is FIXED (K cycles) regardless of the
// secret operands — i.e. the timing observable is data-oblivious. The datapath still reduces the secret
// operan... | CONSTANT_TIME |
9 | LEAKY | LEAKY — 'done' depends on a_in, b_in.
How each secret reaches it (shortest path first):
a_in
├─ a
├─ eq
└─ done
b_in
├─ b
├─ eq
└─ done
Each arrow is a dependency edge: an assignment, or a condition guarding
one. The analysis over-approximates, so a path may be unreachable at run
time — read it and ... | euclid_gcd.v | {"a": ["a_in", "b", "eq", "rst", "running", "start"], "b": ["a", "b_in", "eq", "rst", "running", "start"], "done": ["eq", "running"], "eq": ["a", "b"], "gcd_out": ["a"], "running": ["rst", "start"]} | CC-BY-4.0 | 19 | 4 | 10 | done | [{"secret": "a_in", "observation": "done", "signals": ["a_in", "a", "eq", "done"], "length": 3}, {"secret": "a_in", "observation": "done", "signals": ["a_in", "a", "b", "eq", "done"], "length": 4}, {"secret": "b_in", "observation": "done", "signals": ["b_in", "b", "eq", "done"], "length": 3}, {"secret": "b_in", "observ... | [
"a_in",
"b_in"
] | null | true | [
"a_in",
"b_in"
] | 3 | // euclid_gcd — subtractive Euclidean GCD. The canonical NON-constant-time datapath: the number of
// reduction steps (hence the cycle at which `done` asserts) depends on the SECRET operands, so the
// completion time leaks operand structure. This is the negative / found-bug design: a self-composition
// proof of timin... | LEAKY |
6 | CONSTANT_TIME | null | euclid_gcd_repaired.v | {"a": ["K", "a_in", "b", "iter", "rst", "running", "start"], "b": ["K", "a", "b_in", "iter", "rst", "running", "start"], "done": ["K", "iter", "running"], "gcd_out": ["a"], "iter": ["K", "rst", "running", "start"], "running": ["rst", "start"]} | CC-BY-4.0 | 24 | 0 | 11 | done | null | [] | null | true | [
"a_in",
"b_in"
] | null | // euclid_gcd_repaired — the CERTIFIED REPAIR of `euclid_gcd.v` (WS-C). Generated by applying ONE known
// countermeasure class (fixed-trip-count / data-oblivious control) to the leaky design, in response to its
// timing-leak counterexample. This is NOT a hand-written ct_gcd: it PRESERVES euclid's own subtractive
// r... | CONSTANT_TIME |
6 | CONSTANT_TIME | null | ct_modmul.v | {"acc": ["W", "ar", "br", "cnt", "rst", "running", "start"], "ar": ["a", "rst", "start"], "br": ["W", "b", "cnt", "rst", "running", "start"], "cnt": ["W", "rst", "running", "start"], "done": ["W", "cnt", "running"], "mr": ["m", "rst", "start"], "result": ["W", "acc"], "running": ["rst", "start"]} | CC-BY-4.0 | 30 | 0 | 14 | done | null | [] | null | true | [
"a",
"b",
"m"
] | null | // ct_modmul — CONSTANT-TIME double-and-add modular multiplication `a*b mod m` (a building block of RSA/ECC
// scalar routines). Every cycle it doubles the accumulator and conditionally adds `a`, with a conditional
// modular reduction — but the multiplier bit only SELECTS values; it never gates control. Completion
// ... | CONSTANT_TIME |
6 | LEAKY | LEAKY — 'done' depends on b.
How each secret reaches it (shortest path first):
b
├─ br
└─ done
Each arrow is a dependency edge: an assignment, or a condition guarding
one. The analysis over-approximates, so a path may be unreachable at run
time — read it and decide. That is why it is printed. | modmul_leaky.v | {"acc": ["W", "ar", "br", "rst", "running", "start"], "ar": ["a", "rst", "start"], "br": ["b", "rst", "running", "start"], "done": ["br", "running"], "mr": ["m", "rst", "start"], "result": ["W", "acc"], "running": ["rst", "start"]} | CC-BY-4.0 | 22 | 1 | 13 | done | [{"secret": "b", "observation": "done", "signals": ["b", "br", "done"], "length": 2}] | [
"b"
] | null | true | [
"a",
"b",
"m"
] | 2 | // modmul_leaky — the LEAKY counterpart of `ct_modmul`: it early-exits once the residual multiplier `br` is
// zero (skipping the remaining doublings), so `done` timing leaks the multiplier's MSB position. The
// completion cone contains operand bits, so the self-composition proof correctly REFUSES to certify it.
modul... | LEAKY |
6 | CONSTANT_TIME | null | ct_mul.v | {"acc": ["W", "cnt", "mcand", "mplier", "rst", "running", "start"], "cnt": ["W", "rst", "running", "start"], "done": ["W", "cnt", "running"], "mcand": ["W", "a", "cnt", "rst", "running", "start"], "mplier": ["W", "b", "cnt", "rst", "running", "start"], "prod": ["acc"], "running": ["rst", "start"]} | CC-BY-4.0 | 29 | 0 | 12 | done | null | [] | null | true | [
"a",
"b"
] | null | // ct_mul — CONSTANT-TIME shift-add multiplier. Runs EXACTLY W iterations regardless of the operands:
// every cycle it conditionally adds the (shifted) multiplicand based on the current multiplier bit, but the
// bit only SELECTS a value — it never gates the control. Completion `done = running & (cnt == W)` is a plain... | CONSTANT_TIME |
6 | LEAKY | LEAKY — 'done' depends on b.
How each secret reaches it (shortest path first):
b
├─ mplier
└─ done
Each arrow is a dependency edge: an assignment, or a condition guarding
one. The analysis over-approximates, so a path may be unreachable at run
time — read it and decide. That is why it is printed. | mul_leaky.v | {"acc": ["mcand", "mplier", "rst", "running", "start"], "done": ["mplier", "running"], "mcand": ["W", "a", "mplier", "rst", "running", "start"], "mplier": ["b", "rst", "running", "start"], "prod": ["acc"], "running": ["rst", "start"]} | CC-BY-4.0 | 20 | 1 | 11 | done | [{"secret": "b", "observation": "done", "signals": ["b", "mplier", "done"], "length": 2}] | [
"b"
] | null | true | [
"a",
"b"
] | 2 | // mul_leaky — the LEAKY counterpart of `ct_mul`: it early-exits as soon as the residual multiplier becomes
// zero, so `done` fires after a number of cycles equal to the multiplier's MSB position — a timing channel
// that leaks the secret operand's bit-length. The completion cone contains operand bits (via `mplier==0... | LEAKY |
6 | CONSTANT_TIME | null | modexp_ct.v | {"acc": ["W", "cnt", "e", "rst", "running", "sq", "start"], "cnt": ["W", "rst", "running", "start"], "done": ["W", "cnt", "running"], "e": ["W", "cnt", "exp", "rst", "running", "start"], "result": ["acc"], "running": ["rst", "start"], "sq": ["W", "base", "cnt", "rst", "running", "start"]} | CC-BY-4.0 | 29 | 0 | 12 | done | null | [] | null | true | [
"base",
"exp",
"modulus"
] | null | // modexp_ct — CONSTANT-TIME square-and-multiply-ALWAYS modular exponentiation control, the standard RSA
// timing-attack countermeasure (the always-multiply / constant-time-exponentiation defense — the exponent
// bit never gates control, only selects a value; this is the always-multiply class, distinct from but in th... | CONSTANT_TIME |
8 | LEAKY | LEAKY — 'done' depends on exp.
How each secret reaches it (shortest path first):
exp
├─ e
└─ done
Each arrow is a dependency edge: an assignment, or a condition guarding
one. The analysis over-approximates, so a path may be unreachable at run
time — read it and decide. That is why it is printed. | modexp_leaky.v | {"acc": ["W", "cnt", "e", "rst", "running", "sq", "start"], "cnt": ["W", "e", "rst", "running", "start"], "done": ["W", "cnt", "e", "running"], "e": ["W", "cnt", "exp", "rst", "running", "start"], "result": ["acc"], "running": ["rst", "start"], "sq": ["W", "base", "cnt", "e", "rst", "running", "start"]} | CC-BY-4.0 | 32 | 2 | 12 | done | [{"secret": "exp", "observation": "done", "signals": ["exp", "e", "done"], "length": 2}, {"secret": "exp", "observation": "done", "signals": ["exp", "e", "cnt", "done"], "length": 3}] | [
"exp"
] | null | true | [
"base",
"exp",
"modulus"
] | 2 | // modexp_leaky — the NON-constant-time counterpart of `modexp_ct`: a square-and-multiply modular
// exponentiation that EARLY-EXITS when the remaining exponent register is all zeros. That is a real,
// tempting optimization (skip the trailing zero exponent bits), but it makes the completion cycle — hence
// `done` — d... | LEAKY |
6 | CONSTANT_TIME | null | x25519_fieldmul.v | {"acc": ["W", "ar", "br", "cnt", "rst", "running", "start"], "ar": ["opa", "rst", "start"], "br": ["W", "cnt", "opb", "rst", "running", "start"], "cnt": ["W", "rst", "running", "start"], "done": ["W", "cnt", "running"], "mr": ["modulus", "rst", "start"], "result": ["W", "acc"], "running": ["rst", "start"]} | CC-BY-4.0 | 30 | 0 | 14 | done | null | [] | null | true | [
"opa",
"opb",
"modulus"
] | null | // x25519_fieldmul — CONSTANT-TIME field multiplication `opa * opb mod modulus`, the inner primitive of an
// X25519 / Curve25519 Montgomery-ladder step (the field GF(2^255-19), modeled here at a tractable width W).
// The completion-channel CT ARGUMENT is width-parametric in structure, BUT this committed RTL is fixed ... | CONSTANT_TIME |
6 | LEAKY | LEAKY — 'done' depends on opb.
How each secret reaches it (shortest path first):
opb
├─ br
└─ done
Each arrow is a dependency edge: an assignment, or a condition guarding
one. The analysis over-approximates, so a path may be unreachable at run
time — read it and decide. That is why it is printed. | x25519_fieldmul_leaky.v | {"acc": ["W", "ar", "br", "rst", "running", "start"], "ar": ["opa", "rst", "start"], "br": ["opb", "rst", "running", "start"], "done": ["br", "running"], "mr": ["modulus", "rst", "start"], "result": ["W", "acc"], "running": ["rst", "start"]} | CC-BY-4.0 | 22 | 1 | 13 | done | [{"secret": "opb", "observation": "done", "signals": ["opb", "br", "done"], "length": 2}] | [
"opb"
] | null | true | [
"opa",
"opb",
"modulus"
] | 2 | // x25519_fieldmul_leaky — the LEAKY counterpart of `x25519_fieldmul`: it early-exits once the residual
// multiplier `br` is zero, skipping the remaining doublings, so `done` timing leaks the secret multiplier's
// most-significant set-bit position (its bit-length). The completion cone therefore contains operand bits,... | LEAKY |
null | null | null | barrett_spec.v | {"r": ["a"]} | CC-BY-4.0 | 1 | 0 | 2 | null | null | null | null | false | [] | null | // barrett_spec — the PUBLIC golden specification for the Kyber / ML-KEM Barrett reduction (WS-B): the plain
// arithmetic contract `r == a mod 3329` over the full 16-bit coefficient domain. This is the FIPS/spec-level
// reference the `barrett_ct` datapath is bit-exact to, over ALL 2^16 inputs. Purely combinational — ... | null |
null | null | null | alu_unprotected.v | {"_unused": ["a2", "b2", "op2"], "error_flag": [], "out": ["r0"], "r0": ["a", "b", "op"]} | CC-BY-4.0 | 7 | 0 | 10 | null | null | null | null | false | [] | null | // alu_unprotected — the UNPROTECTED ALU: same datapath as `parity_alu.v`, but NO fault detection
// (`error_flag` hardwired 0). The teeth for WS-A: a single modeled stuck-at in the OUT datapath corrupts `out`
// with error_flag never rising, so the golden_faulted 2-safety miter finds an UNDETECTED fault (`differ` SAT)... | null |
null | null | null | parity_alu.v | {"error_flag": ["r0", "r1"], "out": ["r0"], "r0": ["a", "b", "op"], "r1": ["a2", "b2", "op2"]} | CC-BY-4.0 | 9 | 0 | 10 | null | null | null | null | false | [] | null | // parity_alu — a fault-RESISTANT 4-bit ALU (the PROVEN side of the fault-injection twin, WS-A).
//
// The datapath `r0 = op(a,b)` is computed TWICE by two structurally-distinct copies (r0 from a/b/op, r1 from
// a2/b2/op2) and compared: `error_flag = |(r0 ^ r1)` — a concurrent-error-detection (CED) duplicate-and-compa... | null |
null | null | null | parity_alu_singlebit.v | {"error_flag": ["r0", "r1"], "out": ["r0"], "r0": ["a", "b", "op"], "r1": ["a2", "b2", "op2"]} | CC-BY-4.0 | 9 | 0 | 10 | null | null | null | null | false | [] | null | // parity_alu_singlebit — a PARTIAL countermeasure (the second WS-A teeth): the duplicate-and-compare covers
// only output bits [2:0] and IGNORES bit 3, mirroring `pcpi_div_halfwipe.v` (a countermeasure that protects
// only part of the state). A single modeled stuck-at on the bit-3 datapath corrupts `out[3]` while `e... | null |
null | null | null | pcpi_div.v | null | ISC | 0 | 0 | 0 | null | null | null | line 70: preprocessor directive is outside the supported Verilog subset ('`ifdef'). Dependencies created by it would be invisible to the cone analysis, so no verdict is returned. Flatten the design to a single module of assign/always statements, or analyse the submodule directly with --module. | false | [] | null | // Vendored standalone copy of picorv32_pcpi_div (the RISC-V DIV/REM co-processor) from
// the upstream picorv32 project (ISC license) -- see LICENSE-FIXTURES for full attribution.
// Completion is (!quotient_msk && running); quotient_msk is a pure 1<<31 >>1 shift register — so the
// completion cycle is data-oblivious... | UNKNOWN |
null | null | null | pcpi_div_wiped.v | {"dividend": ["divisor", "instr_div", "instr_rem", "pcpi_rs1", "quotient_msk", "resetn", "running", "start"], "divisor": ["instr_div", "instr_rem", "pcpi_rs2", "quotient_msk", "resetn", "running", "start"], "instr_any_div_rem": ["instr_div", "instr_divu", "instr_rem", "instr_remu"], "instr_div": ["pcpi_insn", "pcpi_rea... | ISC | 78 | 0 | 22 | null | null | null | null | false | [] | null | // WS-7 — no-secret-residue repair of picorv32_pcpi_div. Byte-identical to rtl_ct/pcpi_div.v EXCEPT that on the
// completion cycle (`!quotient_msk && running`) the operand-derived scratch registers are ZEROED, so no operand
// residue survives past `done`. pcpi_rd is a non-blocking assignment from the PRE-clock quotie... | null |
null | null | null | pcpi_div_halfwipe.v | {"dividend": ["divisor", "instr_div", "instr_rem", "pcpi_rs1", "quotient_msk", "resetn", "running", "start"], "divisor": ["instr_div", "instr_rem", "pcpi_rs2", "quotient_msk", "resetn", "running", "start"], "instr_any_div_rem": ["instr_div", "instr_divu", "instr_rem", "instr_remu"], "instr_div": ["pcpi_insn", "pcpi_rea... | ISC | 76 | 0 | 22 | null | null | null | null | false | [] | null | // WS-7 TEETH — a DELIBERATELY INCOMPLETE wipe. Identical to pcpi_div_wiped.v except it clears only `dividend`
// and leaves `divisor` holding operand-derived residue at completion. residue_check MUST still REFUSE this
// variant: a partial wipe is not no-residue. If this were accepted, the property would be vacuous.
m... | null |
null | null | null | pcpi_mul.v | null | ISC | 0 | 0 | 0 | null | null | null | line 67: for loop is outside the supported Verilog subset ('for ('). Dependencies created by it would be invisible to the cone analysis, so no verdict is returned. Flatten the design to a single module of assign/always statements, or analyse the submodule directly with --module. | false | [] | null | // Vendored standalone copy of picorv32_pcpi_mul (the RISC-V MUL/MULH co-processor) from
// the upstream picorv32 project (ISC license) -- see LICENSE-FIXTURES for full attribution.
// Completion is `pcpi_ready <= mul_finish`, and `mul_finish <= mul_counter[6]` where `mul_counter`
// is loaded from a fixed constant (63... | UNKNOWN |
null | null | null | tb_ct.v | null | CC-BY-4.0 | 0 | 0 | 0 | null | null | null | line 12: module instantiation is outside the supported Verilog subset ('euclid_gcd EU ('). Dependencies created by it would be invisible to the cone analysis, so no verdict is returned. Flatten the design to a single module of assign/always statements, or analyse the submodule directly with --module. | false | [] | null | // Cross-check testbench: does the ACTUAL RTL behave as the hand-authored z3 relation claims?
// - euclid_gcd: two secret operand pairs must complete at DIFFERENT cycles (the leak is real in RTL).
// - ct_gcd: any secret operand pair must complete at the SAME fixed cycle (constant-time in RTL).
// Prints "DONE ... | UNKNOWN |
hw-verify-paths
Dependency graphs and witness paths for constant-time RTL analysis — the reasoning, not just the label.
The companion dataset records
what each design is: CONSTANT_TIME or LEAKY. This one records why. For every
fixture it carries the full signal dependency graph, and for every leaky one the
concrete chains of signals that carry a secret to the observation.
Why witness paths and not just verdicts
A verdict is a label. It is enough to measure a classifier and not enough to build
one that can be trusted, because the underlying analysis is a syntactic
over-approximation: it follows every dependency edge whether or not that path can
be taken at run time. Some LEAKY verdicts are therefore paths that never execute.
A user staring at a bare verdict has no way to tell a real finding from a false one,
and learns to distrust the tool. A user given key → key_r → cmp_eq → running → done
can look at four signal names and decide. So the path is not decoration — it is what
makes an over-approximate analysis usable by someone entitled to disagree with it.
That makes this a different artefact for a different purpose: reasoning traces for training or evaluating models that must explain a hardware-security finding rather than merely emit one.
Install
pip install datasets
Quickstart
from datasets import load_dataset
import json
ds = load_dataset("nickh007/hw-verify-paths", split="test")
leaky = [r for r in ds if r["verdict"] == "LEAKY"]
r = leaky[0]
print(r["file"], "->", r["reaching_secrets"])
for p in json.loads(r["paths"]):
print(f" {p['secret']}: {' -> '.join(p['signals'])} ({p['length']} edges)")
graph and paths are JSON strings, because Arrow has no good column type for a
ragged adjacency map. json.loads them.
Worked example — what a record actually looks like
$ python -c "
from datasets import load_dataset; import json
ds = load_dataset('nickh007/hw-verify-paths', split='test')
r = [x for x in ds if x['verdict'] == 'LEAKY'][0]
print(r['file'], '->', r['reaching_secrets'])
for p in json.loads(r['paths'])[:1]:
print(' ', ' -> '.join(p['signals']), f\"({p['length']} edges)\")"
barrett_leaky.v -> ['a']
a -> acc -> done (2 edges)
That is the whole point of this dataset: barrett_leaky.v is not merely labelled
LEAKY, it carries the chain — the secret a reaches acc, which reaches the
completion signal done.
Fields
| Field | Meaning |
|---|---|
file |
fixture name |
source |
the full Verilog source |
scored / expected |
whether it is part of the scored benchmark, and its label |
observation / secrets |
the completion signal, and the declared secret inputs |
verdict |
CONSTANT_TIME, LEAKY, UNKNOWN, or null when unscored |
refusal_reason |
why no verdict was reached; null unless verdict is UNKNOWN |
graph |
JSON: signal → the signals it depends on (assignments and guards) |
n_signals / n_edges |
size of that graph |
reaching_secrets |
the secrets found in the observation's fan-in |
cone_size |
signals in the fan-in cone |
paths |
JSON list of {secret, observation, signals, length} |
n_paths / shortest_path_length |
path statistics |
explanation |
the rendered human-readable explanation |
license |
per record — four fixtures are ISC, the rest CC-BY-4.0 |
Honest scope
UNKNOWNrecords carry no graph. Three fixtures use constructs the analysis cannot read, so there is nothing to trace. Theirgraphis null and theirrefusal_reasonsays which construct stopped it. Emitting an empty graph would make an unreadable design look like one with no dependencies.- Paths are capped at 8 per secret. The number of paths through a dependency
graph is exponential;
n_pathsis a bounded sample, shortest first, not a total. - A path is syntactic. It shows a route the value could take, not one it necessarily does. That is the whole reason it is worth reading.
- Verdicts cover completion timing against the declared secrets — not power, EM, cache, or microarchitectural channels.
Reproducing it
Every record is computed from ctbench
at build time:
pip install git+https://github.com/nickharris808/ctbench@main
python build.py --check # fails if the committed data differs from a fresh build
The data cannot drift from the code that produced it, because the check is a test.
Licence
Records are CC-BY-4.0, except four fixtures derived from
picorv32 by Claire Wolf, which remain ISC.
The license field is per record rather than flattened, because flattening it would
misstate the terms on those four.
Part of the hw-verify toolkit
- Live demo — the checker in your browser
- hw-verify dataset — verdicts, masking probes, patch certificates
ctbench·ct-mask·patchproof·patchproof-verify
Citation
Every record here is generated by ctbench;
cite that, using the CITATION.cff in its repository (GitHub renders a "Cite this
repository" button from it).
Contributing
The most valuable contribution is a design whose witness path is wrong — a reported chain that is not a real dependency. See ctbench's CONTRIBUTING.
Part of the hw-verify toolkit
| Project | What it does |
|---|---|
| ▶ Live demo | Constant-time checker in your browser |
| Docs & overview | What the toolkit proves, and what it refuses |
ctbench |
The checker that generated every record here |
hw-verify |
One install, all three checkers |
| hw-verify dataset | The companion: verdicts rather than reasoning |
The commercial boundary. Everything open analyses a design disclosed in full. Proving a property to a third party who never receives the design is a different problem and a commercial one.
- Downloads last month
- 22