Port the capture loop, and make the per-press rate re-derivable
Fingerprintd:Engine is the policy the trustlet cannot supply. It never polls
for a finger: the normal world captures a frame, decides whether a finger is
there, and tells it. So finger detection, edge reporting and the accounting all
live out here, and they are the parts most easily got wrong in a way that reads
as bad hardware.
Baseline refuses to be a fixed threshold. The capture metric is per frame, so
it scales with how many frames a capture asks for, and it drifts upward while
idle -- an early session read "18 -> 24 with a finger" as weak detection when
the values were climbing regardless of what was on the sensor. The floor is the
maximum of the idle samples, and an uncalibrated Baseline calls nothing a
finger rather than inventing a threshold.
TouchTracker keeps the two modes apart. Enrolment reports touch on the rising
edge and release on the falling one and nothing while held, mirroring stock,
whose entire enrolment trace contains no image-ready event; emitting one per
held frame feeds the algorithm near-duplicate images from a single press.
Authentication does want it, because event 7 reaches the matcher
unconditionally.
AuthTally exists to stop two counting mistakes. Only a terminal verdict is an
attempt -- counting rescan frames as rejections is what turned an 8-for-8 run
into an apparent 8-of-39. And a press that ran out of frames without reaching a
verdict is UNDECIDED, not failed; treating it as a failure is the same error one
level up, which is why decided presses are counted separately.
The tests replay the three recorded runs in order rather than as totals,
because press structure only exists in the order. That makes the per-press
claim re-derivable here instead of quoted: the enrolled finger matched on all
five presses although five of its twenty frames did not, the wrong-finger
control matched nothing, and on the stock-budget run five of ten presses
reached a verdict and all five matched.
Verified by mutation: counting undecided presses as decided, using a baseline
before calibration, emitting image-ready during enrolment, and taking the first
idle sample as the floor each fail the suite.
2026-09-02 17:19:45 +02:00
|
|
|
// SPDX-License-Identifier: GPL-3.0-only
|
|
|
|
|
// SPDX-FileCopyrightText: Copyright (C) 2026 Catcrafts®
|
|
|
|
|
|
|
|
|
|
// lint-disable-file fixed-width-types
|
|
|
|
|
/*
|
|
|
|
|
Fingerprintd:Engine unit tests.
|
|
|
|
|
|
|
|
|
|
The accounting half replays the three recorded authentication runs as ordered
|
|
|
|
|
sequences, not just as totals, because press structure only exists in the
|
|
|
|
|
order. That is what lets the per-press claim be re-derived here rather than
|
|
|
|
|
taken from the journal: the enrolled finger matched on every press even though
|
|
|
|
|
five of its twenty frames did not.
|
|
|
|
|
|
|
|
|
|
It also pins the distinction one level up from the -11 mistake. At the stock
|
|
|
|
|
rescan budget a press whose frames all answered "not identified yet" ran out of
|
|
|
|
|
frames without reaching a verdict. It is undecided, not failed, and lumping it
|
|
|
|
|
in with failures is the same error in a new place.
|
|
|
|
|
*/
|
|
|
|
|
import std;
|
|
|
|
|
import Fingerprintd;
|
|
|
|
|
|
|
|
|
|
using namespace fingerprintd::engine;
|
|
|
|
|
using fingerprintd::ta::Verdict;
|
|
|
|
|
using fingerprintd::ta::Event;
|
|
|
|
|
|
|
|
|
|
namespace {
|
|
|
|
|
int Failures = 0;
|
|
|
|
|
void Check(bool cond, std::string_view msg) {
|
|
|
|
|
if (!cond) {
|
|
|
|
|
std::println(std::cerr, "FAIL: {}", msg);
|
|
|
|
|
++Failures;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// A recorded run, in order. The fixtures live with the Ta suite; these are
|
|
|
|
|
// the same three files.
|
|
|
|
|
std::vector<Verdict> LoadRun(std::string_view name) {
|
|
|
|
|
std::vector<Verdict> out;
|
|
|
|
|
std::string path = std::format("tests/Ta/fixtures/{}", name);
|
|
|
|
|
std::ifstream f(path);
|
|
|
|
|
if (!f) {
|
|
|
|
|
std::println(std::cerr, "FAIL: cannot open fixture {}", path);
|
|
|
|
|
++Failures;
|
|
|
|
|
return out;
|
|
|
|
|
}
|
|
|
|
|
std::string line;
|
|
|
|
|
while (std::getline(f, line)) {
|
|
|
|
|
if (line.starts_with("#") || !line.contains("AUTH ")) continue;
|
|
|
|
|
if (line.contains("*** MATCH ***")) out.push_back(Verdict::Match);
|
|
|
|
|
else if (line.contains("matcher never ran")) out.push_back(Verdict::MatcherNeverRan);
|
|
|
|
|
else if (line.contains("REJECTED")) out.push_back(Verdict::Rejected);
|
|
|
|
|
else if (line.contains("no match")) out.push_back(Verdict::NotIdentifiedYet);
|
|
|
|
|
}
|
|
|
|
|
return out;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// The matcher never running is exactly the released-finger frame, so it is
|
|
|
|
|
// also the finger-present signal for press accounting.
|
|
|
|
|
AuthTally Replay(const std::vector<Verdict>& run) {
|
|
|
|
|
AuthTally t;
|
|
|
|
|
for (Verdict v : run)
|
|
|
|
|
t.Observe(v, v != Verdict::MatcherNeverRan);
|
|
|
|
|
return t;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
int main() {
|
|
|
|
|
// ---- Baseline: never a fixed threshold
|
|
|
|
|
{
|
|
|
|
|
Baseline b;
|
|
|
|
|
Check(!b.Ready(), "uncalibrated");
|
|
|
|
|
Check(!b.IsFinger(1000000), "an uncalibrated baseline calls nothing a finger");
|
|
|
|
|
|
|
|
|
|
for (int i = 0; i < 5; i++) b.Observe(132);
|
|
|
|
|
Check(b.Ready(), "calibrated after five idle samples");
|
|
|
|
|
Check(b.Floor() == 132, "floor");
|
|
|
|
|
Check(b.Threshold() == 264, "threshold is 2x the floor");
|
|
|
|
|
|
|
|
|
|
// The real measurement: idle 132-133, finger 345-366.
|
|
|
|
|
Check(!b.IsFinger(133), "an idle frame is not a finger");
|
|
|
|
|
Check(b.IsFinger(345) && b.IsFinger(366), "a pressed frame is");
|
|
|
|
|
|
|
|
|
|
// The floor is the max of the idle samples. A drifting idle must not
|
|
|
|
|
// become a false finger.
|
|
|
|
|
Baseline drift;
|
|
|
|
|
for (std::int32_t m : {18, 20, 22, 24, 26}) drift.Observe(m);
|
|
|
|
|
Check(drift.Floor() == 26, "floor takes the maximum, not the first sample");
|
|
|
|
|
Check(!drift.IsFinger(24), "drift within the idle range is not a finger");
|
|
|
|
|
|
|
|
|
|
// Extra samples after calibration do not move it.
|
|
|
|
|
Baseline fixed;
|
|
|
|
|
for (int i = 0; i < 5; i++) fixed.Observe(100);
|
|
|
|
|
fixed.Observe(9999);
|
|
|
|
|
Check(fixed.Floor() == 100, "calibration closes after its sample count");
|
|
|
|
|
|
|
|
|
|
// A per-frame metric scales with the frame count, so a threshold from
|
|
|
|
|
// one configuration is meaningless in another. Two baselines, same
|
|
|
|
|
// sensor, different capture counts:
|
|
|
|
|
Baseline one, four;
|
|
|
|
|
for (int i = 0; i < 5; i++) { one.Observe(18); four.Observe(68); }
|
|
|
|
|
Check(one.Threshold() != four.Threshold(), "the threshold is not portable between configs");
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// ---- Edge detection
|
|
|
|
|
{
|
|
|
|
|
TouchTracker t;
|
|
|
|
|
// Enrolment: touch on the rising edge, release on the falling one,
|
|
|
|
|
// and nothing at all while held.
|
|
|
|
|
auto e1 = t.Observe(true, Mode::Enrol);
|
|
|
|
|
Check(e1.size() == 1 && e1[0] == Event::FingerTouched, "enrol: rising edge -> touched");
|
|
|
|
|
auto e2 = t.Observe(true, Mode::Enrol);
|
|
|
|
|
Check(e2.empty(), "enrol: a held frame reports nothing");
|
|
|
|
|
auto e3 = t.Observe(false, Mode::Enrol);
|
|
|
|
|
Check(e3.size() == 1 && e3[0] == Event::FingerReleased, "enrol: falling edge -> released");
|
|
|
|
|
auto e4 = t.Observe(false, Mode::Enrol);
|
|
|
|
|
Check(e4.empty(), "enrol: idle reports nothing");
|
|
|
|
|
|
|
|
|
|
// Authentication: every frame with a finger reaches the matcher.
|
|
|
|
|
TouchTracker a;
|
|
|
|
|
auto a1 = a.Observe(true, Mode::Authenticate);
|
|
|
|
|
Check(a1.size() == 2 && a1[0] == Event::FingerTouched && a1[1] == Event::ImageReady,
|
|
|
|
|
"auth: rising edge reports touched then image-ready");
|
|
|
|
|
auto a2 = a.Observe(true, Mode::Authenticate);
|
|
|
|
|
Check(a2.size() == 1 && a2[0] == Event::ImageReady, "auth: a held frame still reports image-ready");
|
|
|
|
|
auto a3 = a.Observe(false, Mode::Authenticate);
|
|
|
|
|
Check(a3.size() == 1 && a3[0] == Event::FingerReleased, "auth: release");
|
|
|
|
|
|
|
|
|
|
// The modes genuinely differ: enrolling every held frame is what feeds
|
|
|
|
|
// the algorithm near-duplicate images from one press.
|
|
|
|
|
TouchTracker x, y;
|
|
|
|
|
x.Observe(true, Mode::Enrol);
|
|
|
|
|
y.Observe(true, Mode::Authenticate);
|
|
|
|
|
Check(x.Observe(true, Mode::Enrol).empty(), "enrol emits nothing while held");
|
|
|
|
|
Check(!y.Observe(true, Mode::Authenticate).empty(), "auth emits while held");
|
|
|
|
|
|
|
|
|
|
Check(a.FingerDown() == false, "tracker reports lifted");
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// ---- Enrolment progress, read from the response
|
|
|
|
|
{
|
|
|
|
|
EnrolSession e;
|
|
|
|
|
Check(!e.Started(), "not started");
|
An enrolment cannot be ended by a finger release
A three-tap enrolment declared itself complete. The transcript says why:
frame 2: metric=308 FINGER ev5 rem=10
frame 3: metric=187 ev6 rem=0
samples: 10 of 10 accepted
The release event never enters do_enroll, so its response leaves
samples-remaining untouched at 0 -- which is indistinguishable from "none
remaining, you are finished". The session believed it, stopped after one press,
and called SAVE_DATA on an algorithm holding no template. That answered -1 and
wrote nothing, so the store was undamaged, but only by luck: the guard meant to
prevent a partial save was itself satisfied by the bogus count.
A reading is only meaningful when it came from the event that runs the enrol
path, and nothing about the value says so -- the caller has to. Observe now
takes that as an argument. Two further guards: a FIRST reading of 0 is an
unpopulated field rather than a finished enrolment, and the count only ever
falls, so an increase is noise.
Verified by mutation: trusting the release event's count, and accepting a
leading zero, each fail the suite.
2026-09-02 20:40:16 +02:00
|
|
|
e.Observe(-1, true);
|
Port the capture loop, and make the per-press rate re-derivable
Fingerprintd:Engine is the policy the trustlet cannot supply. It never polls
for a finger: the normal world captures a frame, decides whether a finger is
there, and tells it. So finger detection, edge reporting and the accounting all
live out here, and they are the parts most easily got wrong in a way that reads
as bad hardware.
Baseline refuses to be a fixed threshold. The capture metric is per frame, so
it scales with how many frames a capture asks for, and it drifts upward while
idle -- an early session read "18 -> 24 with a finger" as weak detection when
the values were climbing regardless of what was on the sensor. The floor is the
maximum of the idle samples, and an uncalibrated Baseline calls nothing a
finger rather than inventing a threshold.
TouchTracker keeps the two modes apart. Enrolment reports touch on the rising
edge and release on the falling one and nothing while held, mirroring stock,
whose entire enrolment trace contains no image-ready event; emitting one per
held frame feeds the algorithm near-duplicate images from a single press.
Authentication does want it, because event 7 reaches the matcher
unconditionally.
AuthTally exists to stop two counting mistakes. Only a terminal verdict is an
attempt -- counting rescan frames as rejections is what turned an 8-for-8 run
into an apparent 8-of-39. And a press that ran out of frames without reaching a
verdict is UNDECIDED, not failed; treating it as a failure is the same error one
level up, which is why decided presses are counted separately.
The tests replay the three recorded runs in order rather than as totals,
because press structure only exists in the order. That makes the per-press
claim re-derivable here instead of quoted: the enrolled finger matched on all
five presses although five of its twenty frames did not, the wrong-finger
control matched nothing, and on the stock-budget run five of ten presses
reached a verdict and all five matched.
Verified by mutation: counting undecided presses as decided, using a baseline
before calibration, emitting image-ready during enrolment, and taking the first
idle sample as the floor each fail the suite.
2026-09-02 17:19:45 +02:00
|
|
|
Check(!e.Started(), "an unpopulated field does not start the session");
|
An enrolment cannot be ended by a finger release
A three-tap enrolment declared itself complete. The transcript says why:
frame 2: metric=308 FINGER ev5 rem=10
frame 3: metric=187 ev6 rem=0
samples: 10 of 10 accepted
The release event never enters do_enroll, so its response leaves
samples-remaining untouched at 0 -- which is indistinguishable from "none
remaining, you are finished". The session believed it, stopped after one press,
and called SAVE_DATA on an algorithm holding no template. That answered -1 and
wrote nothing, so the store was undamaged, but only by luck: the guard meant to
prevent a partial save was itself satisfied by the bogus count.
A reading is only meaningful when it came from the event that runs the enrol
path, and nothing about the value says so -- the caller has to. Observe now
takes that as an argument. Two further guards: a FIRST reading of 0 is an
unpopulated field rather than a finished enrolment, and the count only ever
falls, so an increase is noise.
Verified by mutation: trusting the release event's count, and accepting a
leading zero, each fail the suite.
2026-09-02 20:40:16 +02:00
|
|
|
e.Observe(10, true);
|
Port the capture loop, and make the per-press rate re-derivable
Fingerprintd:Engine is the policy the trustlet cannot supply. It never polls
for a finger: the normal world captures a frame, decides whether a finger is
there, and tells it. So finger detection, edge reporting and the accounting all
live out here, and they are the parts most easily got wrong in a way that reads
as bad hardware.
Baseline refuses to be a fixed threshold. The capture metric is per frame, so
it scales with how many frames a capture asks for, and it drifts upward while
idle -- an early session read "18 -> 24 with a finger" as weak detection when
the values were climbing regardless of what was on the sensor. The floor is the
maximum of the idle samples, and an uncalibrated Baseline calls nothing a
finger rather than inventing a threshold.
TouchTracker keeps the two modes apart. Enrolment reports touch on the rising
edge and release on the falling one and nothing while held, mirroring stock,
whose entire enrolment trace contains no image-ready event; emitting one per
held frame feeds the algorithm near-duplicate images from a single press.
Authentication does want it, because event 7 reaches the matcher
unconditionally.
AuthTally exists to stop two counting mistakes. Only a terminal verdict is an
attempt -- counting rescan frames as rejections is what turned an 8-for-8 run
into an apparent 8-of-39. And a press that ran out of frames without reaching a
verdict is UNDECIDED, not failed; treating it as a failure is the same error one
level up, which is why decided presses are counted separately.
The tests replay the three recorded runs in order rather than as totals,
because press structure only exists in the order. That makes the per-press
claim re-derivable here instead of quoted: the enrolled finger matched on all
five presses although five of its twenty frames did not, the wrong-finger
control matched nothing, and on the stock-budget run five of ten presses
reached a verdict and all five matched.
Verified by mutation: counting undecided presses as decided, using a baseline
before calibration, emitting image-ready during enrolment, and taking the first
idle sample as the floor each fail the suite.
2026-09-02 17:19:45 +02:00
|
|
|
Check(e.Started() && e.Total() == 10 && e.Stages() == 10, "first response sets the total");
|
|
|
|
|
Check(e.Accepted() == 0 && !e.Complete(), "nothing accepted yet");
|
An enrolment cannot be ended by a finger release
A three-tap enrolment declared itself complete. The transcript says why:
frame 2: metric=308 FINGER ev5 rem=10
frame 3: metric=187 ev6 rem=0
samples: 10 of 10 accepted
The release event never enters do_enroll, so its response leaves
samples-remaining untouched at 0 -- which is indistinguishable from "none
remaining, you are finished". The session believed it, stopped after one press,
and called SAVE_DATA on an algorithm holding no template. That answered -1 and
wrote nothing, so the store was undamaged, but only by luck: the guard meant to
prevent a partial save was itself satisfied by the bogus count.
A reading is only meaningful when it came from the event that runs the enrol
path, and nothing about the value says so -- the caller has to. Observe now
takes that as an argument. Two further guards: a FIRST reading of 0 is an
unpopulated field rather than a finished enrolment, and the count only ever
falls, so an increase is noise.
Verified by mutation: trusting the release event's count, and accepting a
leading zero, each fail the suite.
2026-09-02 20:40:16 +02:00
|
|
|
e.Observe(9, true);
|
Port the capture loop, and make the per-press rate re-derivable
Fingerprintd:Engine is the policy the trustlet cannot supply. It never polls
for a finger: the normal world captures a frame, decides whether a finger is
there, and tells it. So finger detection, edge reporting and the accounting all
live out here, and they are the parts most easily got wrong in a way that reads
as bad hardware.
Baseline refuses to be a fixed threshold. The capture metric is per frame, so
it scales with how many frames a capture asks for, and it drifts upward while
idle -- an early session read "18 -> 24 with a finger" as weak detection when
the values were climbing regardless of what was on the sensor. The floor is the
maximum of the idle samples, and an uncalibrated Baseline calls nothing a
finger rather than inventing a threshold.
TouchTracker keeps the two modes apart. Enrolment reports touch on the rising
edge and release on the falling one and nothing while held, mirroring stock,
whose entire enrolment trace contains no image-ready event; emitting one per
held frame feeds the algorithm near-duplicate images from a single press.
Authentication does want it, because event 7 reaches the matcher
unconditionally.
AuthTally exists to stop two counting mistakes. Only a terminal verdict is an
attempt -- counting rescan frames as rejections is what turned an 8-for-8 run
into an apparent 8-of-39. And a press that ran out of frames without reaching a
verdict is UNDECIDED, not failed; treating it as a failure is the same error one
level up, which is why decided presses are counted separately.
The tests replay the three recorded runs in order rather than as totals,
because press structure only exists in the order. That makes the per-press
claim re-derivable here instead of quoted: the enrolled finger matched on all
five presses although five of its twenty frames did not, the wrong-finger
control matched nothing, and on the stock-budget run five of ten presses
reached a verdict and all five matched.
Verified by mutation: counting undecided presses as decided, using a baseline
before calibration, emitting image-ready during enrolment, and taking the first
idle sample as the floor each fail the suite.
2026-09-02 17:19:45 +02:00
|
|
|
Check(e.Accepted() == 1, "rem 10 -> 9 is one accepted sample");
|
An enrolment cannot be ended by a finger release
A three-tap enrolment declared itself complete. The transcript says why:
frame 2: metric=308 FINGER ev5 rem=10
frame 3: metric=187 ev6 rem=0
samples: 10 of 10 accepted
The release event never enters do_enroll, so its response leaves
samples-remaining untouched at 0 -- which is indistinguishable from "none
remaining, you are finished". The session believed it, stopped after one press,
and called SAVE_DATA on an algorithm holding no template. That answered -1 and
wrote nothing, so the store was undamaged, but only by luck: the guard meant to
prevent a partial save was itself satisfied by the bogus count.
A reading is only meaningful when it came from the event that runs the enrol
path, and nothing about the value says so -- the caller has to. Observe now
takes that as an argument. Two further guards: a FIRST reading of 0 is an
unpopulated field rather than a finished enrolment, and the count only ever
falls, so an increase is noise.
Verified by mutation: trusting the release event's count, and accepting a
leading zero, each fail the suite.
2026-09-02 20:40:16 +02:00
|
|
|
for (std::int32_t r : {8, 7, 6, 5, 4, 3, 2, 1}) e.Observe(r, true);
|
Port the capture loop, and make the per-press rate re-derivable
Fingerprintd:Engine is the policy the trustlet cannot supply. It never polls
for a finger: the normal world captures a frame, decides whether a finger is
there, and tells it. So finger detection, edge reporting and the accounting all
live out here, and they are the parts most easily got wrong in a way that reads
as bad hardware.
Baseline refuses to be a fixed threshold. The capture metric is per frame, so
it scales with how many frames a capture asks for, and it drifts upward while
idle -- an early session read "18 -> 24 with a finger" as weak detection when
the values were climbing regardless of what was on the sensor. The floor is the
maximum of the idle samples, and an uncalibrated Baseline calls nothing a
finger rather than inventing a threshold.
TouchTracker keeps the two modes apart. Enrolment reports touch on the rising
edge and release on the falling one and nothing while held, mirroring stock,
whose entire enrolment trace contains no image-ready event; emitting one per
held frame feeds the algorithm near-duplicate images from a single press.
Authentication does want it, because event 7 reaches the matcher
unconditionally.
AuthTally exists to stop two counting mistakes. Only a terminal verdict is an
attempt -- counting rescan frames as rejections is what turned an 8-for-8 run
into an apparent 8-of-39. And a press that ran out of frames without reaching a
verdict is UNDECIDED, not failed; treating it as a failure is the same error one
level up, which is why decided presses are counted separately.
The tests replay the three recorded runs in order rather than as totals,
because press structure only exists in the order. That makes the per-press
claim re-derivable here instead of quoted: the enrolled finger matched on all
five presses although five of its twenty frames did not, the wrong-finger
control matched nothing, and on the stock-budget run five of ten presses
reached a verdict and all five matched.
Verified by mutation: counting undecided presses as decided, using a baseline
before calibration, emitting image-ready during enrolment, and taking the first
idle sample as the floor each fail the suite.
2026-09-02 17:19:45 +02:00
|
|
|
Check(!e.Complete() && e.Remaining() == 1, "not complete at one remaining");
|
An enrolment cannot be ended by a finger release
A three-tap enrolment declared itself complete. The transcript says why:
frame 2: metric=308 FINGER ev5 rem=10
frame 3: metric=187 ev6 rem=0
samples: 10 of 10 accepted
The release event never enters do_enroll, so its response leaves
samples-remaining untouched at 0 -- which is indistinguishable from "none
remaining, you are finished". The session believed it, stopped after one press,
and called SAVE_DATA on an algorithm holding no template. That answered -1 and
wrote nothing, so the store was undamaged, but only by luck: the guard meant to
prevent a partial save was itself satisfied by the bogus count.
A reading is only meaningful when it came from the event that runs the enrol
path, and nothing about the value says so -- the caller has to. Observe now
takes that as an argument. Two further guards: a FIRST reading of 0 is an
unpopulated field rather than a finished enrolment, and the count only ever
falls, so an increase is noise.
Verified by mutation: trusting the release event's count, and accepting a
leading zero, each fail the suite.
2026-09-02 20:40:16 +02:00
|
|
|
e.Observe(0, true);
|
Port the capture loop, and make the per-press rate re-derivable
Fingerprintd:Engine is the policy the trustlet cannot supply. It never polls
for a finger: the normal world captures a frame, decides whether a finger is
there, and tells it. So finger detection, edge reporting and the accounting all
live out here, and they are the parts most easily got wrong in a way that reads
as bad hardware.
Baseline refuses to be a fixed threshold. The capture metric is per frame, so
it scales with how many frames a capture asks for, and it drifts upward while
idle -- an early session read "18 -> 24 with a finger" as weak detection when
the values were climbing regardless of what was on the sensor. The floor is the
maximum of the idle samples, and an uncalibrated Baseline calls nothing a
finger rather than inventing a threshold.
TouchTracker keeps the two modes apart. Enrolment reports touch on the rising
edge and release on the falling one and nothing while held, mirroring stock,
whose entire enrolment trace contains no image-ready event; emitting one per
held frame feeds the algorithm near-duplicate images from a single press.
Authentication does want it, because event 7 reaches the matcher
unconditionally.
AuthTally exists to stop two counting mistakes. Only a terminal verdict is an
attempt -- counting rescan frames as rejections is what turned an 8-for-8 run
into an apparent 8-of-39. And a press that ran out of frames without reaching a
verdict is UNDECIDED, not failed; treating it as a failure is the same error one
level up, which is why decided presses are counted separately.
The tests replay the three recorded runs in order rather than as totals,
because press structure only exists in the order. That makes the per-press
claim re-derivable here instead of quoted: the enrolled finger matched on all
five presses although five of its twenty frames did not, the wrong-finger
control matched nothing, and on the stock-budget run five of ten presses
reached a verdict and all five matched.
Verified by mutation: counting undecided presses as decided, using a baseline
before calibration, emitting image-ready during enrolment, and taking the first
idle sample as the floor each fail the suite.
2026-09-02 17:19:45 +02:00
|
|
|
Check(e.Complete() && e.Accepted() == 10, "complete at zero");
|
|
|
|
|
}
|
|
|
|
|
|
An enrolment cannot be ended by a finger release
A three-tap enrolment declared itself complete. The transcript says why:
frame 2: metric=308 FINGER ev5 rem=10
frame 3: metric=187 ev6 rem=0
samples: 10 of 10 accepted
The release event never enters do_enroll, so its response leaves
samples-remaining untouched at 0 -- which is indistinguishable from "none
remaining, you are finished". The session believed it, stopped after one press,
and called SAVE_DATA on an algorithm holding no template. That answered -1 and
wrote nothing, so the store was undamaged, but only by luck: the guard meant to
prevent a partial save was itself satisfied by the bogus count.
A reading is only meaningful when it came from the event that runs the enrol
path, and nothing about the value says so -- the caller has to. Observe now
takes that as an argument. Two further guards: a FIRST reading of 0 is an
unpopulated field rather than a finished enrolment, and the count only ever
falls, so an increase is noise.
Verified by mutation: trusting the release event's count, and accepting a
leading zero, each fail the suite.
2026-09-02 20:40:16 +02:00
|
|
|
// ---- A release event must not end the enrolment
|
|
|
|
|
//
|
|
|
|
|
// This shipped: the release event never enters do_enroll, so its response
|
|
|
|
|
// leaves samples-remaining at 0, and a session that trusts it declares
|
|
|
|
|
// itself finished after ONE press. SAVE_DATA is then called on an
|
|
|
|
|
// algorithm holding no template and answers -1.
|
|
|
|
|
{
|
|
|
|
|
EnrolSession e;
|
|
|
|
|
e.Observe(10, true); // touch: ten to go
|
|
|
|
|
e.Observe(0, false); // release: reads zero, means nothing
|
|
|
|
|
Check(!e.Complete(), "a release reading of 0 does NOT complete the enrolment");
|
|
|
|
|
Check(e.Remaining() == 10, "and does not move the count");
|
|
|
|
|
Check(e.Accepted() == 0, "nothing was accepted");
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// A first reading of 0 is an unpopulated field, not a finished enrolment.
|
|
|
|
|
{
|
|
|
|
|
EnrolSession e;
|
|
|
|
|
e.Observe(0, true);
|
|
|
|
|
Check(!e.Started(), "a leading zero does not start a session");
|
|
|
|
|
Check(!e.Complete(), "and certainly does not finish one");
|
|
|
|
|
e.Observe(10, true);
|
|
|
|
|
Check(e.Started() && e.Total() == 10, "a real count still starts it");
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// The count only falls; a jump back up is noise, not progress.
|
|
|
|
|
{
|
|
|
|
|
EnrolSession e;
|
|
|
|
|
e.Observe(10, true);
|
|
|
|
|
e.Observe(9, true);
|
|
|
|
|
e.Observe(7, true);
|
|
|
|
|
Check(e.Remaining() == 7, "decreases are taken");
|
|
|
|
|
e.Observe(9, true);
|
|
|
|
|
Check(e.Remaining() == 7, "an increase is ignored");
|
|
|
|
|
}
|
|
|
|
|
|
Port the capture loop, and make the per-press rate re-derivable
Fingerprintd:Engine is the policy the trustlet cannot supply. It never polls
for a finger: the normal world captures a frame, decides whether a finger is
there, and tells it. So finger detection, edge reporting and the accounting all
live out here, and they are the parts most easily got wrong in a way that reads
as bad hardware.
Baseline refuses to be a fixed threshold. The capture metric is per frame, so
it scales with how many frames a capture asks for, and it drifts upward while
idle -- an early session read "18 -> 24 with a finger" as weak detection when
the values were climbing regardless of what was on the sensor. The floor is the
maximum of the idle samples, and an uncalibrated Baseline calls nothing a
finger rather than inventing a threshold.
TouchTracker keeps the two modes apart. Enrolment reports touch on the rising
edge and release on the falling one and nothing while held, mirroring stock,
whose entire enrolment trace contains no image-ready event; emitting one per
held frame feeds the algorithm near-duplicate images from a single press.
Authentication does want it, because event 7 reaches the matcher
unconditionally.
AuthTally exists to stop two counting mistakes. Only a terminal verdict is an
attempt -- counting rescan frames as rejections is what turned an 8-for-8 run
into an apparent 8-of-39. And a press that ran out of frames without reaching a
verdict is UNDECIDED, not failed; treating it as a failure is the same error one
level up, which is why decided presses are counted separately.
The tests replay the three recorded runs in order rather than as totals,
because press structure only exists in the order. That makes the per-press
claim re-derivable here instead of quoted: the enrolled finger matched on all
five presses although five of its twenty frames did not, the wrong-finger
control matched nothing, and on the stock-budget run five of ten presses
reached a verdict and all five matched.
Verified by mutation: counting undecided presses as decided, using a baseline
before calibration, emitting image-ready during enrolment, and taking the first
idle sample as the floor each fail the suite.
2026-09-02 17:19:45 +02:00
|
|
|
// ---- The three recorded runs, replayed in order
|
|
|
|
|
{
|
|
|
|
|
auto enrolled = LoadRun("auth-enrolled-finger.txt");
|
|
|
|
|
auto wrong = LoadRun("auth-wrong-finger.txt");
|
|
|
|
|
auto stock = LoadRun("auth-stock-budget.txt");
|
|
|
|
|
Check(enrolled.size() == 25 && wrong.size() == 21 && stock.size() == 48,
|
|
|
|
|
"all three runs loaded in order");
|
|
|
|
|
|
|
|
|
|
// The enrolled finger, forced-terminal. 15 of 20 frames matched...
|
|
|
|
|
AuthTally e = Replay(enrolled);
|
|
|
|
|
Check(e.Matches() == 15 && e.Rejections() == 5, "enrolled: 15 match / 5 reject");
|
|
|
|
|
Check(e.TerminalFrames() == 20, "enrolled: 20 terminal frames");
|
|
|
|
|
// ...but every press did, which is the number a user experiences.
|
|
|
|
|
Check(e.Presses() == 5, "enrolled: five presses");
|
|
|
|
|
Check(e.PressesMatched() == 5, "enrolled: every press matched");
|
|
|
|
|
Check(e.PressesDecided() == 5, "enrolled: every press reached a verdict");
|
|
|
|
|
Check(e.Identified(), "enrolled: the finger was identified");
|
|
|
|
|
// The distinction the journal insists on.
|
|
|
|
|
Check(e.Matches() != e.TerminalFrames(), "the frame rate is not 100%");
|
|
|
|
|
Check(e.PressesMatched() == e.PressesDecided(), "the press rate is");
|
|
|
|
|
|
|
|
|
|
// The control. This is the claim that matters most about the device.
|
|
|
|
|
AuthTally w = Replay(wrong);
|
|
|
|
|
Check(w.Matches() == 0, "wrong finger: zero false accepts");
|
|
|
|
|
Check(w.Rejections() == 19, "wrong finger: 19 rejections");
|
|
|
|
|
Check(w.PressesMatched() == 0, "wrong finger: no press matched");
|
|
|
|
|
Check(w.PressesDecided() == 2 && w.Presses() == 2, "wrong finger: both presses decided");
|
|
|
|
|
Check(!w.Identified(), "wrong finger: not identified");
|
|
|
|
|
|
|
|
|
|
// The stock-budget run: most frames are "not identified yet".
|
|
|
|
|
AuthTally s = Replay(stock);
|
|
|
|
|
Check(s.Matches() == 8, "stock budget: 8 matches");
|
|
|
|
|
Check(s.NotIdentifiedYet() == 31, "stock budget: 31 rescan frames");
|
|
|
|
|
Check(s.Rejections() == 0, "stock budget: not one real rejection");
|
|
|
|
|
Check(s.TerminalFrames() == 8, "stock budget: 8 terminal frames, all matches");
|
|
|
|
|
|
|
|
|
|
// Five of its ten presses ran out of frames without a verdict. They
|
|
|
|
|
// are undecided, not failures -- and every press that DID reach a
|
|
|
|
|
// verdict matched.
|
|
|
|
|
Check(s.Presses() == 10, "stock budget: ten presses");
|
|
|
|
|
Check(s.PressesDecided() == 5, "stock budget: five reached a verdict");
|
|
|
|
|
Check(s.PressesUndecided() == 5, "stock budget: five ran out of frames");
|
|
|
|
|
Check(s.PressesMatched() == 5, "stock budget: every decided press matched");
|
|
|
|
|
Check(s.PressesMatched() == s.PressesDecided(),
|
|
|
|
|
"stock budget: the decided-press rate is 5/5, not 5/10");
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// ---- Undecided presses must not be counted as failures
|
|
|
|
|
{
|
|
|
|
|
AuthTally t;
|
|
|
|
|
// One press, all rescan frames, then a lift.
|
|
|
|
|
for (int i = 0; i < 4; i++) t.Observe(Verdict::NotIdentifiedYet, true);
|
|
|
|
|
t.Observe(Verdict::MatcherNeverRan, false);
|
|
|
|
|
Check(t.Presses() == 1, "one press");
|
|
|
|
|
Check(t.PressesDecided() == 0, "it reached no verdict");
|
|
|
|
|
Check(t.PressesMatched() == 0, "and matched nothing");
|
|
|
|
|
Check(t.Rejections() == 0, "but it produced no rejection either");
|
|
|
|
|
Check(t.TerminalFrames() == 0, "and no terminal frame");
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// ---- Session state: a failure needs a power cycle, not a retry
|
|
|
|
|
{
|
|
|
|
|
Check(CanInit(SessionState::Powered), "init from powered");
|
|
|
|
|
Check(!CanInit(SessionState::Ready), "no second init on a live session");
|
|
|
|
|
Check(!CanInit(SessionState::Cold), "no init before power");
|
|
|
|
|
Check(!CanInit(SessionState::Failed), "a failed session may not simply re-init");
|
|
|
|
|
Check(NeedsPowerCycle(SessionState::Failed), "it needs the rail cycled");
|
|
|
|
|
Check(!NeedsPowerCycle(SessionState::Ready), "a healthy session does not");
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (Failures == 0) std::println("Engine: all tests passed");
|
|
|
|
|
return Failures;
|
|
|
|
|
}
|