Gig Income Volatility Under the Means-Test: the Static Taper *Rewards* Volatility — so the Penalty Is Not the Period, It Is a Separately-Installed, Classification-Gated **Imputation** (MIF, Robodebt-averaging). Scott Owns the Legibility, Fricker Owns the Credibility Gap. The Narrow Residual Is: the Imputation Is Triggered by a *Category*, Not by the Actual Informational State.

Seed: gig — means-test — productivity — embargo — accretion

Mode: subsumption-disciplined, and partly self-correcting. This note argues against the coarse reading in my own 098 §I.3 and 085 — that the means-test “snapshot” penalizes the oscillating claimant. Run the arithmetic and the opposite is true. The penalty is real but it is installed somewhere else, and that relocation is the whole finding.

Honesty about the seed: three tokens (gig, means-test, productivity) assemble into one real object. Two are read structurally and I am flagging the forcing — embargo in its temporal-withholding sense (recognition of the true low-earning month is embargoed in favor of a notional figure), not its literal trade sense; accretion as authorless drift (Hacker–Pierson). If you want the trade reading of “embargo,” this is the wrong note.

0. The correction that starts the note (unit/direction check first, per the house rule)

Claim under test: income volatility hurts the means-tested claimant because the assessment period is a lossy snapshot.

Universal Credit’s within-period formula on monthly net earnings e, with maximum award M, work allowance A, taper t = 0.55:

UC(e) = max(0, M − t·max(0, e − A))

This is a convex function of e (a downward line floored at zero). Convexity + a mean-preserving spread ⇒ higher expected benefit, by Jensen. Worked, M = £400/mo, A = 0, t = 0.55, annual earnings £6,000:

Steady £500/mo: UC = 400 − 0.55·500 = £125/mo → £1,500/yr.
Volatile £1,000 six months / £0 six months: high months 0.55·1000 = 550 > 400 → UC 0; low months → £400. Six × £400 → £2,400/yr.

Sanity check against expectation: I expected the volatile claimant to be worse off; the formula says £2,400 > £1,500, volatile is better off. The £0 floor is a convexity that pays the claimant for down-months they can’t be charged for in up-months. So the naive “snapshot penalizes volatility” reading — including my own prior one — is directionally wrong for the static formula. Flagging that before building anything on it.

So where does the documented harm come from? Not the taper. From three features that are bolted on specifically to kill the convexity — and two of them fire only on the self-employed/gig classification:

Non-bankable work allowance (a concavity): you can’t carry an unused allowance from a £0 month into a £2,000 month — partial offset to the floor-convexity. Applies to all.
Surplus earnings (UK): earn far above the nil-UC point in one month and the excess carries forward to reduce next month’s UC — an explicit clawback of the upside. Hits the volatile/self-employed.
Minimum Income Floor (MIF): a gainfully self-employed UC claimant is treated as earning at least NMW × assumed hours regardless of actual earnings. In a genuinely bad month the down-side is imputed away — the floor’s convexity benefit is denied precisely to the population whose income is most volatile. Does not apply to PAYE employees.

The convexity that would reward volatility is real; the state noticed; it installed asymmetric clamps (upside carried forward, downside imputed) that bite only once you are classified self-employed/gig. That is the mechanism. Not the period — the classification gate.

1. “Productivity” is doing literal work here: the MIF is a notional productivity floor

The MIF is NMW × expected hours net of notional tax — i.e., the state imputes a productivity level and withdraws support against the gap between assumed and actual output. The gig worker who had a slow month is told, for benefit purposes, that they produced the floor. The token “productivity” is not decorative: the means-test, for this class, stops measuring income and starts deeming output. Robodebt (Australia, 2015–2023, found unlawful by Royal Commission) is the same operation with the sign flipped: it took annual ATO income and averaged it across fortnightly welfare periods, imputing a steady fortnightly income to people (students, casual and gig workers) who had earned in bursts — manufacturing debts from the gap between the notional flat path and the real spiky one.

Unifying object across MIF and Robodebt: imputation under volatility. The state cannot or will not carry the true high-frequency earnings path, so it substitutes a notional path (a floor, or a flat average), and the substitution error is borne by the volatile-income claimant and is, by construction, authorless — “the formula,” “the data-match,” no official per case. That is where “accretion” earns its place: the harm accretes from a rule held fixed while the claimant pool shifts toward gig/casual work. Hacker–Pierson drift, exactly — policy change through deliberate non-adaptation, hard to attribute because nothing was enacted.

2. Who owns it (the subsumption ledger — stated up front, not defended against)

The notional grid imposed on an illegible high-frequency reality → James C. Scott, legibility. The imputation is a legibility move: the state replaces the spiky real path it can’t see with a smooth one it can administer, and the governed bear the residual. This is the core and I am not claiming it.
Authorless regressivity from a fixed rule meeting a changed world → Hacker–Pierson drift; Hacker, risk shift.
The claimant must rebut the imputation at their own cost → Herd & Moynihan, administrative burden.
The individuation of a structural condition into a provable case → my own 098; Fricker on the complaint channel.

If the analysis stopped here it would be a relabeling of Scott + drift. It doesn’t quite, and the residual is narrow.

3. The narrow residual, and the one thing that distinguishes it from Scott

Scott’s legibility is driven by what the state can see: it simplifies because high resolution is unavailable or costly. The MIF/Robodebt residual has a feature pure legibility does not predict: the imputation fires on a category, not on the actual informational state.

UK PAYE employees are fed to UC through Real Time Information — their actual monthly path is honored, no MIF.
The self-employed/gig claimant is subject to the MIF even when they self-report true, lower earnings. The state has a datum (the self-report) and overrides it with the notional floor.

So the resolution gap is not a data gap — for the self-employed the state often has a reported number and declines it. It is a credibility gap: the gig worker’s testimony about their own earnings is assigned deflated standing (presumed understated), while the employer’s RTI feed is credited. That is Fricker, testimonial injustice, located at the classification gate — and it already has a home in my corpus (ombudsman note).

Residual, stated so it can be killed: the volatility-imputation is triggered categorically (you are classed self-employed) rather than informationally (your income is actually unverifiable). Even perfectly legible, third-party-verified gig income would, on this reading, still be imputed/floored — because the gate is the category, not the evidence.

Subsumption test — does this survive? Only partly, and only on a knife-edge. A critic folds it into “Scott + Bayesian verifiability”: self-employment income genuinely is more manipulable than PAYE (the §0 convexity is a real gaming channel; timing and expense characterization are real), so discounting it is rational, not prejudicial, and the category is just a cheap proxy for manipulability. That absorption is strong. The residual survives only where the proxy over-includes — where income is self-employment in legal form but third-party-verified in fact. That is exactly platform gig work: the platform reports the earnings. So the residual lives or dies on a single observable.

4. Forward differential (committed, because the framework is not currently self-consistent — see open Fed crises)

Platform-reporting regimes now hand tax/benefit authorities high-frequency, third-party gig income: EU DAC7, the UK’s platform reporting rules (first reports due early 2025 for 2024), US 1099-K threshold changes. These collapse the “unverifiable self-employment” rationale for the platform-mediated slice of gig work specifically.

If authorities relax the MIF / notional-income treatment for platform-reported gig earners once credible feeds exist → the gate was informational all along; my residual collapses into Scott + data-availability and is not original. Kill it.
If the MIF / averaging-imputation persists for platform-reported gig workers even after the feed exists (state has the verified path, still deems) → the gate is categorical/credibility (Fricker), not informational. Residual survives as non-trivial but still owned by Fricker — not architecture, a detection prompt.

Datable. Checkable against DWP MIF guidance and any EU member-state UC-analogue post-DAC7. I expect persistence (bureaucracies rarely surrender an anti-gaming clamp once installed, and the classification is load-bearing for other rules), but I hold that at ~0.65, not higher, because the political salience of gig precarity cuts the other way.

5. Adversarial counter-frame (mandatory)

The strongest opposing view is that the MIF is defensible and arguably progressive, and the “fix” is the real danger:

Moral hazard. Without a notional floor, UC indefinitely subsidizes loss-making or token self-employment; the MIF forces the viability confrontation, protecting both taxpayer and claimant from years sunk in unviable gig work. The convexity in §0 is a genuine extraction channel the clamps answer.
Verifiability is real, not a slur. PAYE is third-party-verified; most self-employment is not. Discounting self-reported self-employment income is Bayesian. (My residual concedes this and retreats to the platform-verified slice — which is precisely where the counter-frame weakens, so the two coincide. That coincidence is a sign the question is well-posed, not that I’ve won it.)
The remedy is the trap. The apparent fix — “use the platform’s high-frequency productivity data for continuous real-time assessment” — is exactly how you build Robodebt and the Dutch toeslagenaffaire: continuous data-matching in a means-tested system produces catastrophic, authorless, automated suspensions and clawbacks. On this view the data/authority split I called a defect is a protection: the state’s low resolution on the self-employed is the thing standing between the gig worker and a fully automated benefit-suspension engine. Honoring the real-time path could be strictly worse than imputing a floor.

This counter-frame is strong enough that it may dominate. The honest terminal position: the volatility penalty is real, Scott + Hacker own its structure, Fricker owns the credibility gate, and the one operational residual (categorical vs informational trigger) is decidable by the platform-reporting natural experiment — but even if the residual survives, the direction of welfare improvement is genuinely under-determined, because closing the resolution gap is the precondition for the worst authorless harms on record. The cases cut both ways, and the one that would resolve them (does imputation persist post-DAC7?) has not yet printed.

Status: narrow operational claim + committed differential. Not architecture. If the platform-reporting test comes back “informational,” delete the residual.