True cold boot
VDD and VBAT fall. C statics restart, fallback seeded becomes false, public Yasmarang returns to its constants and the call history begins at zero.
INTERACTIVE SECURITY EXPLAINER / FIRMWARE 4.1.9
See the weak Mk3 seed path, one state at a time.
Configure the wafer-coordinate UID, SysTick phase, RTC registers and every RNG-consuming action. Then watch the exact fallback and public Yasmarang streams produce the 32-byte pre-mnemonic value.
DEFAULT REACHABLE PAD SUPPORT224under the 256-Y model and 80,000 SysTick states
KNOWN UID80,000initial pads
STOCK COLD RTC0 + 0TR / SSR
SUPPORT ≠ SECURE ENTROPY
01 / DEVICE LAB
Operate the vulnerable path
Every control updates one deterministic execution. There is deliberately no address matching, batch scan or private-key derivation.
COLDCARDMK3 · LAB UNIT
BOOT ROM / MK3
Cold boot complete
v4.1.9 · RTC static
Press OK to continue
GENUINE LIGHTCold boot #1
REGISTER CONSOLELIVE
01
Device identitySTM32 UID low word = Y:16 | X:16
02
SysTick phase80 MHz down-counter, 1 ms reload
03
Cold RTC profileTR and SSR feed n and d
PAD0x004A5B50UID ⊕ SysTick
HISTORY0 F / 0 P0 rejections
SEED DRAW+8 F / +8 P32 bytes
02 / CALL HISTORY COMPOSER
History advances state. It does not reseed.
The exact trace has two dimensions: fallback calls and public-mixer calls. Rejection sampling makes them diverge.
Load a path
ADD AN OPERATIONone click = one real firmware action
#OperationΔ fallbackΔ publicState after
h = 0No RNG-consuming action since cold boot.
→Generate 24-word entropyrandom.bytes(32), then SHA-256+8+88 F / 8 P
03 / ENTROPY GEOMETRY
Configure the population model
This calculator counts distinct initial pads exactly for contiguous X/Y coordinate ranges and SysTick values from zero.
Model boundary
ST specifies two 16-bit coordinate fields. A uniform full 256 × 256 production grid is an assumption, not a measured Coldcard sales dataset.
DISTINCT REACHABLE PADS16,777,216
224.000support-size bits, not cryptographic entropy
Input tuples5,242,880,000UID × timer combinations
Mean preimages312.50tuples per distinct pad
One UID coverage0.4768%80,000 pads
Known-UID ceiling216.288candidate-count ceiling
ONE DEVICE, 1,000 HISTORIES80,000,000
GLOBAL MODEL, 1,000 HISTORIES16,777,216,000
=
0.4768%same coverage fraction
History adds rows to both spaces. It does not let a fixed UID resample the missing cold-boot pads. A genuinely different-history collision against the exact 256-bit seed remains a separate, negligible event under the random-output model.
04 / RESET SEMANTICS
Where does history go?
→
Startup actions
Settings filler, saves, keypad shuffles, paper wallets or USB crypto can advance state before New Wallet is selected.
HISTORY → h→
Warm Reset
MicroPython restarts without reinitializing the C statics. Both deterministic streams continue from their current state.
HISTORY → h + …05 / CLAIMS & PROVENANCE
“Could generate” is not “did generate.”
A post-hoc device demonstration establishes compatibility only. The Mk3 has no signed generation record or authenticated history log.
Importing a known seed
Works on any UIDIt proves present knowledge of the seed, not which device generated it or who historically owned the funds.
Compatible native generation
Possible, then timing-boundA matching UID/timer/history trace shows that the stock algorithm could produce the entropy. It can still be constructed after disclosure.
Historical attribution
Requires earlier evidenceUse pre-dispute descriptors, xpubs, multiple derived addresses, purchase records, funding records and timestamped commitments.
FORENSIC RULE
Never accept a single address, current signature or physical device as proof of seed origin. Verify a complete descriptor and contemporaneous evidence predating compromise.
METHODOLOGY
What is exact—and what is a model
The fallback initializer, both Yasmarang update functions, byte packing, rejection sampling and final SHA-256 match the disclosed v4.1.9 source. Device visuals and menu choreography are an educational reconstruction, not QEMU or a complete firmware emulator.