MWG Advisory · Loop Audit

Three apps, eight dimensions, one rubric

pliability, GOWOD and Bend against the Loop Audit v2. Six weeks, fifteen recordings, one completed program, two deliberate lapses, one re-test.

01Scores

pliability 60 · GOWOD 53 · Bend 42

Eight dimensions, scored 0–5, weighted to 100. Only observed cells are firm; a cell scores 4+ only where I could see the link between my state and the next prescription.

DimensionWtpliabilityGOWODBend
Diagnosis104O3O1O
Assignment153O3O3O
Action enablement104O4O4O
Verification153O2O2O
Coaching & adaptation153O2O2O
Progression & mastery153O3O1O
Re-entry & cadence101O2O2O
Trust, agency & safety10—X—X—X
Earned / scored weight54.0 / 9048.0 / 9038.0 / 90
Normalized to 100605342
O  observed in use I  inferred X  not tested

Two land in Useful tool (50–69), Bend in Feature stack (30–49). Nobody reaches Strong but leaky (70–84). Trust is unscored for all three by decision, so all three are out of 90, normalized.

Three cells moved in the final week. All three moved up.

pliability verification and progression 2→3 on the re-test; GOWOD re-entry 1→2 on a push that prompted its own re-test. Ten corrections across the study, nine in the apps' favor.

02Profiles

Same band, different shape

pliability

60 · Useful tool · one broken link

Diagnosis4
Assignment3
Action4
Verification3
Coaching3
Progression3
Re-entry1
Trustuntested

GOWOD

53 · Useful tool · no broken link

Diagnosis3
Assignment3
Action4
Verification2
Coaching2
Progression3
Re-entry2
Trustuntested

Bend

42 · Feature stack · two broken links

Diagnosis1
Assignment3
Action4
Verification2
Coaching2
Progression1
Re-entry2
Trustuntested

Hatched bars are untested, not zero. A 60 built on a best-in-class diagnosis and no re-entry is a different product from a 53 built on an even spread.

03Detail

One line per cell

Dimension · testAppWhy
Diagnosis
Does it know anything about this body?
pliability4Camera, 3 min, movement compensations, a composite and a scored key area. Best instrument in the category.
GOWOD3Ten tests per side, plus sport and equipment on file — all of it alters the protocol. Held at 3: I grade myself on a slider.
Bend1Nine onboarding questions that personalize the paywall pitch. Nothing measured.
Assignment
Is today's work chosen for me?
pliability3A three-week day-by-day program written off the scan, with the reasoning carried through. It lives two taps from home.
GOWOD3Protocols built around time, equipment and the named workout I did — the best intake question in the set, asked every session.
Bend3A routine served without making me choose, and an AI that writes a rationale per exercise. Both discarded immediately after.
Action enablement
Can I actually do it?
pliability4Filmed coaching through the hold: instruction during the work, not instead of it.
GOWOD4Functional, familiar, adapted to the equipment on hand.
Bend4Lowest-friction session in the set. Tapping the live demo pauses the timer — a tell.
Verification
Does it know whether it worked?
pliability3Raised on the re-test. Run it twice and it grades against the baseline: 66→73, Back/Core 53→65, "you've improved by 12 points in a key area." Held at 3 — nothing triggers it, and between tests there is only attendance.
GOWOD2Checkmarks and cumulative hours. The monthly retest isn't tied to a target.
Bend2Logging works reliably. There is no target for it to be measured against.
Coaching & adaptation
Did yesterday change today?
pliability3"Your HRV is lower than average — let's down-regulate," with two named sessions attached. Real, and from a sensor it doesn't own. The program itself never bends.
GOWOD2Adapts to what I type, every time, from scratch. Nothing carries between sessions.
Bend2Setting difficulty to Expert returned near-identical content in a different wrapper. Difficulty is a label on one library.
Progression & mastery
What can I do now that I couldn't?
pliability3Raised on the re-test. A scored trajectory, a named "Most improved," +12 against a body area. Held below 4: Most improved and Needs attention are the same area with the same number, and the completion ceremony counts hours and days.
GOWOD3Six zone scores plotted over time — real history, no target on it. The surfaced milestone is 307 hours, trophy at 400.
Bend1The Bendometer reads 1% after one session and forecasts +2% tomorrow. A meter named after the body, computed from days elapsed.
Re-entry & cadence
What happens after missed days?
pliability1Lapsed twice; nothing happened. Missed routines stay available, but the day names aren't binding, so there was nothing to be late for.
GOWOD2Raised. It pushed me to re-test — the only notification in six weeks aimed at an instrument rather than a streak. Held at 2: on return after eighteen months it showed a March 2025 score as current.
Bend2Widget flipped to "Last chance!" before midnight; a Streak Restore covered the gap without scolding. A real mechanic, entirely about the counter.
Trust, agency & safetyUntested across all three, by decision. Noted for whoever runs it next: Bend ships a Caution Areas setting, the only explicit safety input I found.
04The cuts

Three tests that matter more than the total

Broken link

Any core stage below 2 caps the practical loop.

pliability — capped at re-entry (1).

Bend — capped twice, diagnosis (1) and progression (1). It can't start the loop or close it.

GOWOD — nothing below 2. The only intact chain of the three.

Causality

Can it explain why an input changed a recommendation?

pliability — and on the re-test it does what this cut actually asks: attributes a measured change to the work. Back/Core 53→65. The only causal claim about a body in the study.

GOWOD — shows its working: the protocol lists the time, equipment and workout it was built from.

Bend — per exercise, beautifully, for an input it immediately forgets.

Graduation

Capability and insight gained, or usage accumulated?

pliability — the only split verdict, and it falls on routing. The completion ceremony reports 5h 1m, 21 routines, 53 poses and a quote. Run the re-test yourself and you get 66→73 and a retired compensation pattern. The insight exists; the ceremony doesn't contain it.

Nobody graduates. On pliability's own screen, Most improved and Needs attention name the same area. Nothing closes, so no new goal opens.

GOWOD and Bend — usage only.

Friction, and the zero that flatters

Scored 0–3 per stage. One result matters more than the ledger: Bend scores zero friction on measurement because Bend has no measurement. An absent stage costs nothing, so absence reads as efficiency. Friction should only be scored within stages that exist — a note back to the rubric, not just the apps.

05Read

Everything is built. Almost nothing routes.

The headline changed in the last week of the study, and it changed in the apps' favor.

I started certain this category measures a body once and never checks. Wrong. Run pliability's test a second time and it produces a scored trajectory (66→73), a per-area delta with a tooltip (Back/Core 53→65), a plain-language claim, a dropped compensation pattern, and an offer to build the next program from the new numbers.

Measured, re-measured, delta rendered, next plan offered. The whole join is built.

And nothing routes a member to it. No prompt fired. The completion ceremony, ten days from the due date, never mentioned it. I went and got it. The most valuable screen in the category sits behind a door nobody knocks on. GOWOD holds the other half: it pushed me to re-test, the only notification in six weeks aimed at an instrument. Better reminder, weaker instrument. Between them they hold a complete product.

What's still genuinely absent is graduation — and underneath that progress screen, "we currently don't have any recommended videos that match your needs." An empty state at the exact moment the product knows the most it has ever known about this body.

Why the scores still sit where they do

Bend scores lowest and has the best engagement trend and the largest audience by an order of magnitude. Not a contradiction: these dimensions measure whether a product changes a user's state and makes it legible, not whether it can acquire users cheaply. Bend earns $4.54 per monthly active user. GOWOD and pliability earn $58 to $86.

This is a pricing-power instrument, not an engagement one.

So the opportunity is smaller units, not more features

The library, the camera, per-side measurement, AI rationale and a working re-test all already ship. The missing layer is small, frequent units of engagement that make a member feel attended to, each producing or spending a piece of state: a five-second check-in scoped to the diagnosed area, a fifteen-second single-movement spot-check, a one-question debrief, a weekly adjustment that cites last week, a notification that names the area rather than the streak.

The test for whether a unit earns its place: does it produce data or spend it? A unit that does neither is content.

Participation mechanics move frequency. Verification mechanics move price. Micro-units are how verification gets a cadence.

One tester, six weeks. Sharp for finding mechanisms, blunt for sizing them. No app is penalized here for something I didn't test, or credited for a dimension left out because it would have scored badly.

MWG Advisory · mwgadvisory.com Loop Audit v2 · final

Also: the full teardown this scoring is drawn from.

← Back to all research