CIP Evaluation Synthesis

This folder evaluates the CIPs that act on the stake-cap layer of the Cardano reward pipeline — the reward-eligible pool stake $\sigma'$ that enters the SL-D1 reward formula, upstream of the operator/member split that the fee layer reshapes.

Two CIPs are in scope: CIP-0050 and CIP-0037. Both target a real broken signal that the mainnet diagnostic confirms — pledge is priced as irrelevant by the operator population. 78 % of staked ADA sits in pools with pledge ratio under 1 %; 42 of the 48 largest multi-pool operators forfeit the pledge bonus; pledged ADA yields 0.68 %/yr while the same ADA placed as passive delegation yields 2.3 %/yr. Both CIPs respond by making pledge binding on the reward formula — without sufficient pledge, the pool's reward-eligible stake is clipped.

Verdict on both CIPs: no-go, for two stacked reasons.

The load-bearing piece is A(ν, π), and neither CIP touches it. V1 already exposes a pledge-incentive knob — a₀, the weight of the pledge bonus inside the reward envelope (current mainnet value 0.3; full walkthrough in Appendix A — Why V1's pledge incentive doesn't work). Raising a₀ rebalances the formula but does not make pledge "matter more" — every low-pledge pool earns less before the bonus can recover. CIP-0050 and CIP-0037 add a third lever (clip σ′ before the formula runs) on top of a₀ and k, but they accept A(ν, π) as given. That function carries three structural pathologies:

• a permanent quadratic ν² size penalty applies at every pledge ratio — small pools are crushed regardless of how committed the operator is;
• a non-monotonicity in π for sub-half-saturated pools — a 2 M operator (ν ≈ 0.03) earns 8.7× more bonus by pledging 51 % than by fully self-pledging. The formula explicitly incentivises small operators to under-commit;
• a cubic ν³ collapse at full self-pledge — a saturated operator earns 37 595× more bonus than a 2 M operator at maximum commitment, because the size factor cubes when the pledge factor saturates.

The σ′ clip changes who can earn the V1 reward; it does not repair what A does to the pledge signal. The relative bonus disparity, the under-commitment incentive, and the cubic crush all carry through unchanged.

Why the radical approach backfires. Today, almost no one pledges at scale: the stake-weighted median pool sits at π = 0.07 %, 78 % of staked ADA is in pools below 1 % pledge ratio, and 42 of the 48 saturation-scale multi-pool operators forfeit the pledge bonus entirely. We do not actually know how much of the SPO landscape could comply if a hard pledge cap were switched on — and what mainnet shows so far is almost no one.

Under that condition, the cap hits every segment of the operator population hard at the same time:

• Small / single-pool retail operators — the median retail pool is clipped to ~7 % of its V1 reward (median π = 0.07 % × L = 100), without the liquid capital to raise pledge above the cap.
• Multi-pool entities — clipped on the per-pool axis: splitting pledge across many pools shrinks each pool's reward envelope, and even large fleets do not have the pledge to comply at scale across all their pools.
• Custodial-by-extraction operators (CEX / IVaaS, ~21 % of productive stake) — clipped to zero because they cannot self-pledge custodied retail funds by construction.

The blast radius covers the bulk of the productive population, and the resulting reward collapse risks destabilising consensus itself. If most operators see their net income fall sharply at the same time, some will reduce or shut down their infrastructure — and Cardano's block-production reliability degrades. A reform meant to strengthen the network's commitment signal ends up weakening the network's basic operation. The direction of effect runs opposite to μ02 — Guarantee operator viability and to the foundational work V2 Roadmap Milestone 1 is set up to do.

A more gradual path makes more sense. A genuine V2 stake-cap reform should reinforce the pledge signal at its source rather than gating it, by combining four moves on the reward-distribution layer:

• Repair A(ν, π) so pledge is not a dominated strategy at any operator size:
• a smoother operator onset at low ν — no cubic crush of small pools as they grow;
• a design that does not privilege fully-private pools (π = 1) — the V2 target is not "everyone runs a 100 %-pledged pool";
• explicit reward for the balanced-commitment regime (π ≈ 0.5) — pledge serving as a credible signal and the pool remaining open to delegation.

• Reduce λ_size so the commitment axis carries more of the signal and the size axis carries less. Today the envelope E(ν, π) = λ_size·ν + λ_pledge·A(ν, π) is split λ_size ≈ 76.9 % + λ_pledge ≈ 23.1 % (set by a₀ = 0.3). This rebalancing only makes sense once A is repaired — otherwise it amplifies the broken gradient. The size-weight cut does double duty: it reduces what low-pledge pools currently extract via the size axis, and the freed weight funds the viability slice in the next move.

• Add a viability package for pools entering the lifecycle — open a new λ_viability sub-budget. Today the envelope is split two ways (λ_size + λ_pledge); the recommendation is a three-way split λ_size + λ_pledge + λ_viability, without raising the total pool pot. While transaction-fee inflows are still small, the global allocation stays where it is; it grows back naturally as Tx fees mature into a larger share of the pot. The funding source for λ_viability is the λ_size reduction above — nothing new is taken from elsewhere; nothing is asked from the reserve.

λ_viability is conditional, not unconditional: a pool benefits from the viability slice only if its operator pledges according to rules to be specified (e.g. a minimum pledge ratio or a pledge-growth schedule across lifecycle stages). The viability function therefore lives on the reward-distribution layer (pre-split), not bolted onto pricing — pricing tools (minPoolCost, margin, rate) stay free as competitive levers (see the fee-layer synthesis).

• Activate the λ_pledge budget that has been underused for years. POL.O1.F3 documents that 95.6 % of the pledge-bonus budget already returns to the reserve unused every epoch — the single largest addressable inefficiency in the system today. A repaired A with a reduced λ_size is what activates this budget; it is not new ADA, it is unused ADA already inside the formula's envelope.

This gradual path lets the pledge signal recover with the operator population, not against it. A hard cap stacked on top of today's regime, by contrast, sends an even larger share of the pool pot back to the reserve and worsens viability for every SPO segment at once — exactly the pathology V2 needs to eliminate, not amplify.

Table of Contents

1. The two candidates — same primitive, different floor

The shared intent. Under V1, the saturation cap is a constant — orig_sat = 1/k ≈ 67.44 M ₳ at k = 500 — independent of pledge. A pool can attract delegation up to that ceiling regardless of how much pledge the operator puts up; pledge enters the reward calculation only via the small bonus term $a_0$ in the SL-D1 numerator (worth ≈ 30 % of the reward at $a_0 = 0.3$, but structurally dominated by passive-delegation yield). Both CIPs in this folder share a single intent: replace this constant horizontal cap with a function of pledge. The new cap rises linearly with the operator's pledge until it reaches the V1 ceiling — beyond that, the new rule and V1 coincide.

Table 1.1 — The two stake-cap candidates as σ′-clipping rules. CIP-0050 is a single hard cap proportional to pledge; CIP-0037 is the same slope plus a floor at $e \cdot \text{orig\_sat}$ when pledge is too small.

Structural kinship. For any pool large enough that $\sigma \geq \text{orig\_sat}$, the two candidates are the same primitive — a linear-in-pledge slope capped at the V1 saturation — differing only on what happens when pledge is low:

• CIP-0050 clips the stake cap to $L \cdot p$ — at zero pledge, $\sigma' = 0$ (hard break).
• CIP-0037 clamps the stake cap to $\ell \cdot p$ but places a floor at $e \cdot \text{orig\_sat}$ — at zero pledge, $\sigma' = e \cdot \text{orig\_sat} \approx 13.49$ M ₳ at reference.

Reference leverages differ by convention ($\ell = 125$ vs $L = 100$), not by design intent.

Panel (b) matches leverage at $\ell = L = 125$ to isolate the floor as the sole structural difference. CIP-0037 is CIP-0050 plus a floor — both target the same pledge-as-signal and concentration intent via the same mechanism; CIP-0037 softens the low-pledge edge at a three-scalar governance cost instead of a one-scalar one.

The two candidates at a glance.

Table 1.2 — The two stake-cap candidates and the verdict carried in their per-CIP files.

2. Reading order

1. cip-0050.md — the primitive in its cleanest one-scalar form ($L$). Start here: every structural finding on the slope carries into CIP-0037.
2. cip-0037.md — the same primitive with an added floor and two effective governance parameters $(e, \ell)$. Read as "CIP-0050 plus floor" — the formula walkthrough in its Appendix A makes the kinship explicit.
3. Appendix A — Why V1's pledge incentive doesn't work — the structural critique of A(ν, π) itself, which neither CIP modifies. Optional for casual readers; load-bearing for anyone designing a successor proposal.

3. References

• Folder parent: ../README.md — solution-evaluation landing + cross-CIP conclusion.
• Cross-layer subfolder: ../operator-delegator/README.md — fee-layer evaluations.
• Standalone k-lever analysis (held-formula-fixed assumption): cip-0082 §B.3 standalone k-lever deep dive.
• Diagnostic anchor for the A(ν, π) critique: pools-distribution §2.3.

Appendix A — Why V1's pledge incentive doesn't work

V1 already exposes a pledge-incentive knob: the pledge influence factor $a_0$ (currently 0.3 on mainnet). It enters the SL-D1 reward envelope as the weight of the pledge-bonus term:

$$E(\nu, \pi) \;=\; \underbrace{\lambda_{\text{size}} \cdot \nu}_{\text{base — independent of pledge}} \;+\; \underbrace{\lambda_{\text{pledge}} \cdot A(\nu, \pi)}_{\text{bonus — pledge-sensitive}}$$

with $\lambda_{\text{size}} = 1/(1+a_0)$, $\lambda_{\text{pledge}} = a_0/(1+a_0)$, and the pledge-bonus activation function

$$A(\nu, \pi) \;:=\; \nu^2 \cdot \pi \cdot \bigl[1 - \pi(1-\nu)\bigr]$$

(Notation and derivation in diagnostic / pools-distribution §2.3.)

What ν and π actually mean. Both are dimensionless ratios in $[0, 1]$ that capture the two structurally independent degrees of freedom an operator controls.

Table A.1 — The two structural axes of the V1 reward formula.

ν and π are structurally independent — pool size and commitment fraction can vary freely, each on its own [0, 1] interval. The operating mainnet population sits very near the π = 0 axis: 78 % of staked ADA is in pools with π < 1 %.

A reading shortcut: when you see ν in the formula, think "how big is the pool relative to one full V1 pool". When you see π, think "how much of the pool is the operator's own ADA".

The natural question is therefore: why propose CIP-0050 / CIP-0037 instead of just raising a₀? The answer requires looking at three nested layers — the lever's shape, the bonus function's structure, and what no proposal currently touches.

A.1. The a₀ lever rebalances, it doesn't tilt

Raising a₀ shifts more weight from the base term ($\lambda_{\text{size}}\nu$) onto the bonus term ($\lambda_{\text{pledge}}A$). For a low-pledge pool this reduces the base by more than the bonus can recover — the operator is punished smoothly, not catalysed.

STK.A.1 — V1 levers vs σ′-clipping CIPs on a Healthy pool: raising $a_0$ from 0.3 → 3.0 cuts zero-pledge reward smoothly to 32 %; the CIPs leave $a_0$ alone but cliff-clip $\sigma'$ at the pledge threshold, hitting the structurally larger base term.

Panel (a). For a Healthy pool ($\sigma = 15$ M, $\nu \approx 0.222$): raising $a_0$ from 0.3 → 1.0 drops the zero-pledge reward to 65 % of baseline; raising to 3.0 drops it to 32 %. The bonus barely recovers across the full pledge range — even at 600 k of self-pledge the higher-a₀ curves stay below baseline. The a₀ lever cannot make pledge "matter more" without first making low-pledge pools earn less.

Panel (b). CIP-0050 and CIP-0037 don't touch $(λ_{\min}, λ_{\max})$. They clip $\sigma'$ before the reward formula runs, so the penalty hits the base term $λ_{\min} \cdot \nu'$ — which is structurally much larger than $λ_{\max} \cdot A$ at any reasonable pool size. That is why their cliff is steep where a₀ tweaks barely move the needle.

A.2. The deeper bottleneck — A(ν, π) itself

Both a₀ (rebalancing) and CIP-0050 / CIP-0037 (clipping) operate around the A function. Neither modifies it. So before plugging any numbers in, dissect the function itself: what does it say structurally?

A.2.1. Anatomy of the function — before any numbers

STK.A.2 — Structure of $A(\nu, \pi)$ on the unit square: the bonus is non-monotone in $\pi$ for any $\nu < 0.5$, with an interior maximum at $\pi^{*} = 1/[2(1-\nu)]$ — sub-half-saturated operators earn less by fully self-pledging.

(i) The factorisation: pure size factor × pledge-intensity factor.

A admits a clean multiplicative decomposition:

$$A(\nu, \pi) \;=\; \underbrace{\nu^2}_{\text{size factor}} \;\cdot\; \underbrace{\pi \cdot \bigl[1 - \pi(1-\nu)\bigr]}_{\text{pledge-intensity factor}}$$

The two effects are independent and multiplicative. The outer factor $\nu^2$ is a quadratic dependence on pool size — independent of pledge, applying at every commitment level. A pool earns bonus proportional to $\nu^2$ before any consideration of how much its operator pledges. The inner factor $\pi[1 - \pi(1-\nu)]$ controls how the pledge ratio modulates the bonus, with a weak coupling to $\nu$ via the $(1-\nu)$ term.

This is the load-bearing observation: pool size enters the bonus quadratically as a pure penalty against small pools, regardless of how committed the operator is. Even at the OPTIMAL pledge ratio for a given pool size, the bonus is still scaled by $\nu^2$.

(ii) Tour of the corners and edges.

Table A.2 — A(ν, π) at the four corners and edges of the unit square. The third row (full self-pledge) is where the elaborate quadratic construction collapses to a cubic.

The third row is where the elaborate quadratic construction collapses. At full self-pledge, the inner factor $\pi[1 - \pi(1-\nu)]$ degenerates to $\nu$, and combined with the outer $\nu^2$ produces $\nu^3$ — cubing sub-unit numbers. That cube is the load-bearing pathology, but as (i) made explicit, the underlying $\nu^2$ size penalty is permanent regardless of pledge.

(iii) The pledge-intensity factor and what it was meant to do.

The inner factor $\pi[1 - \pi(1-\nu)] = \pi - \pi^2(1-\nu)$ has two pieces with distinct intents. The linear $\pi$ part is the bilinear "more pledge → more bonus" signal. The quadratic $-\pi^2(1-\nu)$ part is a splitting penalty the SL-D1 design adds on purpose: an MPO who splits a fixed total pledge across $N$ pools shrinks the per-pool $\pi$, and the quadratic term penalises high-$\pi$/low-$\nu$ configurations the protocol associates with potential gaming.

The construction does achieve that intent in some regions. But it pays a heavy price elsewhere — pathology (iv) below.

(iv) The structural defect — A is non-monotone in π for any ν < 0.5.

Take the partial derivative of A with respect to π at fixed ν:

$$\frac{\partial A}{\partial \pi} \;=\; \nu^2 \bigl[1 - 2\pi(1-\nu)\bigr]$$

This is zero at $\pi^* = 1 / [2(1-\nu)]$ and negative for $\pi > \pi^*$. Two regimes follow:

• For $\nu \geq 0.5$: $\pi^* \geq 1$, so the maximum sits at or beyond the boundary of the unit interval. Inside $[0, 1]$, A is monotone increasing in $\pi$ — pledging more always earns more bonus. ✓
• For $\nu < 0.5$: $\pi^* < 1$, so the maximum sits strictly inside $[0, 1]$. Increasing pledge ratio from $\pi^*$ up to $\pi = 1$ (full self-pledge) decreases A. Pledging more pays less.

Worked example at $\nu = 0.3$ (a Healthy-tier pool around 20 M ADA):

Table A.3 — A(0.3, π) for a Healthy-tier pool: the maximum sits at π = 0.714, not at π = 1; pledging beyond ~71 % destroys part of the bonus.

For a pool exactly at half-saturation ($\nu = 0.5$), the optimum is at $\pi = 1$. Below half-saturation, an operator who fully self-pledges destroys part of their bonus. The formula whose stated purpose is "skin in the game" pays you less for putting in more skin, for the entire population of pools below half-saturation — which is essentially the entire mainnet population.

Panel (b) of the figure shows this: each curve is A at fixed ν as a function of π over the unit interval. The gold star marks the interior maximum; the square marks the full self-pledge endpoint. For $\nu < 0.5$, the square is below the star.

(v) Summary of structural critiques — before any numbers.

1. The bonus has a quadratic outer size penalty $\nu^2$ that holds at every pledge ratio. Small pools are quadratically penalised for being small, before pledge enters the picture.
2. At full self-pledge ($\pi = 1$), the inner factor degenerates and A collapses to $\nu^3$ — the worst-case manifestation of the size penalty, compounding with a residual $\nu$ from the inner factor.
3. For any pool below half-saturation, A is non-monotone in $\pi$ — pledging beyond $\pi^* = 1/[2(1-\nu)]$ actively reduces the bonus.
4. The intended MPO-splitting penalty (the $-\pi^2(1-\nu)$ term) is achieved at the cost of (1)–(3).

These are pre-empirical defects: they hold regardless of mainnet data, regardless of what a₀ is set to, regardless of CIP reforms acting on σ′. They are properties of the algebra. With this in hand, the next subsection puts numbers on what they mean for actual operators.

A.2.2. What A actually pays — three operators across three pledge levels

Cast. Three honest operators, all running pools of different sizes:

• Bob runs a Sub-reliable 2 M ADA pool ($\nu \approx 0.03$).
• Charles runs a Healthy 15 M ADA pool ($\nu \approx 0.222$).
• Alice runs a Saturated 67 M ADA pool ($\nu \approx 0.99$).

All three are below half-saturation except Alice, so Bob and Charles already sit in the non-monotone regime described in (iv). Now follow the same three pledge configurations on each.

Scenario A — what mainnet actually does today (median pledge ratio 0.07 %). This is where 78 % of staked ADA actually sits today. Operators put down a token amount of pledge and earn near-zero bonus — but lose nothing significant either, because the opportunity cost of pledging that token amount is also small.

Table A.4 — Today's mainnet equilibrium: every operator earns near-zero bonus. The formula is essentially silent about pledge.

Even Alice — a Saturated pool with the median pledge ratio — only earns ~\$90/yr in pledge bonus. The formula is essentially silent about pledge for everyone in this scenario. This is the equilibrium the diagnostic captures.

Scenario B — what CIP-0050 demands at L = 100 (1 % pledge ratio). To reach the CIP-0050 compliance threshold (p ≥ σ/L), each operator must commit substantially more capital. The bonus does grow — but the disparity across pool sizes already shows up sharply.

Table A.5 — At CIP-0050 compliance: same act (1 % pledge ratio), 1 123× more bonus for Alice than for Bob. Bob loses ~100× by complying (locked capital yielding 0.023 % vs 2.3 % passive).

Same act (1 % pledge ratio) — Alice earns 1 123× more bonus than Bob for committing the same fraction of her pool. And Bob has just been asked to lock 20 000 ₳ (\$5 000) of his own capital to earn 4.6 ₳/yr (\$1.15) in bonus. The yield on his pledge is 0.023 % vs 2.3 % passive — he loses ~100× by complying.

Scenario C — the maximum signal anyone can give (100 % self-pledge, π = 1). The strongest possible commitment: every ADA in the pool is the operator's own. No MPO games, no delegator slack — pure skin-in-the-game. This is the corner $\pi = 1$ and triggers the cubic collapse from (ii).

Table A.6 — At maximum commitment: 37 595× disparity between Alice and Bob. The cubic ν³ dominates.

Even at the maximum possible commitment, Alice earns 37 595× more bonus than Bob — because the cubic $\nu^3$ that emerges at $\pi = 1$ collapses the bonus on pool size, not on the strength of the commitment signal.

Furthermore — and this is pathology (iv) made tangible — Bob is on the wrong side of the maximum. His optimal pledge ratio is $\pi^* = 1/[2(1-\nu)] = 1/(2 \cdot 0.9703) \approx 0.515$ (about 51 % of his pool, $p^* \approx 1.03$ M ADA). At full self-pledge he earns 14 ₳/yr; at the interior optimum he would earn ~122 ₳/yr — 8.7× more bonus by withholding half his potential pledge. The formula explicitly incentivises him to under-commit.

STK.A.3 — Full-self-pledge bonus across three operators: Alice (Saturated) earns 37 595× more bonus than Bob (Sub-reliable) for the same maximum commitment, and all three earn pledge yields below the 2.3 %/yr passive-delegation alternative.

Panel (a) is Scenario C as a bar chart at log scale (the disparity is too large for linear axes). Panel (b) re-expresses the same disparity as a "bonus yield" — bonus per ADA of pledge per year — and overlays the passive-delegation yield (~2.3 %/yr) the operator gives up by locking that pledge: Bob's pledge yields 0.0007 %/yr in bonus, Charles's 0.038 %/yr, Alice's 0.77 %/yr. All three are below passive delegation, but Bob is by far the most penalised.

A.2.3. The cubic ν³ — visualised

Combine the corner-collapse from (ii) with the non-monotone pathology from (iv): the operator who gives the strongest possible signal (full self-pledge, $\pi = 1$) is paid by $\nu^3$ — a destruction operator on sub-unit numbers, layered on top of the permanent $\nu^2$ size penalty.

STK.A.4 — The cubic $\nu^3$ that emerges at $\pi = 1$ vs the linear "fair share" $\nu$: at Bob's $\nu = 0.03$, the kernel destroys a factor of ~1 137× of the bonus he would otherwise earn.

Panel (a) shows ν³ (red) versus quadratic ν² (orange) and linear ν (green, "fair share"). On linear axes, the cubic curve hugs zero until ν ≈ 0.5 and then leaps to 1 at full saturation — so anyone running a pool below half-saturation is in the flat region where pledge barely matters.

Panel (b) shows the same curves on log axes — the gap between cubic and linear is multiplicative, not additive. For Bob's ν = 0.03: the cubic gives 2.6 × 10⁻⁵, while a linear A would give 0.030 — a 1 137× ratio. That ratio is what the formula is destroying.

Panel (c) makes the cubic crush tangible in dollars. Bob's 2 M pool, fully self-pledged:

Table A.7 — Bob's bonus under alternative kernels: the cubic destroys ~1 137× of what a linear "fair share" would pay.

Compare to the passive-delegation alternative: if Bob delegates that 2 M instead of pledging it, he earns ~46 000 ₳/yr at 2.3 %/yr. Under the current cubic, pledging costs him ~46 000 ₳/yr in opportunity for 14 ₳/yr in bonus. Pledging is a 3 286× loss for him. Under a linear A, the bonus alone (15 529 ₳/yr) would be a third of his opportunity cost — pledging would still lose, but less catastrophically. Under the scale-free kernel, pledging would be net positive even for the smallest operator.

A.2.4. What this means for the CIP critique

Walking through the structural anatomy and the three scenarios reveals one cumulative argument:

1. The function is structurally awkward before any data is plugged in (A.2.1). Permanent quadratic size penalty $\nu^2$, non-monotonic in $\pi$ for sub-half-saturation pools, and at full self-pledge it collapses to a cubic.
2. Mainnet today (Scenario A). The bonus is silent for everyone. The 78 % zero-pledge equilibrium is the predictable consequence of a formula with a near-zero gradient in the operating region.
3. At CIP-0050 compliance (Scenario B). The disparity across pool sizes becomes severe — Alice gets 1 123× more bonus than Bob for the same relative effort. Bob loses ~100× by complying.
4. At maximum commitment (Scenario C). The disparity becomes catastrophic — 37 595× — and the cubic $\nu^3$ at the corner $\pi = 1$ is the algebraic reason.

CIP-0050 and CIP-0037 modify the enforcement of pledge (clip $\sigma'$ if pledge is too low) but not the pricing of pledge inside A. After their reform, the relative bonus disparity across operator sizes remains identical; the non-monotone regime for $\nu < 0.5$ remains identical; the cubic collapse at full self-pledge remains identical. They patch around A without touching it.

A reform that touched A directly — replacing the kernel with one that doesn't impose the quadratic size penalty $\nu^2$ at every pledge ratio, or that doesn't cube small pools at full commitment — would be the most structural way to repair the pledge signal at its source. No CIP currently in scope proposes this. This is the deepest critique of both candidates in this folder: they accept A as given and patch around it, when A is the load-bearing piece of the pledge incentive.

This reading extends the formal critique at diagnostic / pools-distribution §2.3.5: "the bonus term $\lambda_{\text{pledge}}A(\nu, \pi)$ is non-linear and asymmetric in its two inputs. The outer factor $\nu^2$ imposes a quadratic size penalty that holds at every pledge ratio; at full self-pledge ($\pi = 1$) the inner factor degenerates and the bonus collapses to $\lambda_{\text{pledge}}\nu^3$ — the cubic that suppresses the bonus structurally for any pool below saturation."

A.3. What this implies for the CIP candidates

The two CIPs in this folder accept the V1 reward formula as given and patch around it via $\sigma'$ clipping. Three honest readings follow from §A.1 and §A.2:

• CIP-0050 and CIP-0037 are not strictly worse than raising a₀. They achieve the same intent (make pledge bind) without the smooth-but-uniformly-painful penalty that an a₀ raise imposes on every low-pledge pool. The cliff/floor shape is different, not necessarily worse.
• The "MPO fleet-splitting" property neither a₀ nor an A reform would deliver as cleanly. CIP-0050's revenue-neutral pool-splitting ($N \cdot L \cdot (P/N) = L \cdot P$) is an algebraic identity of its primitive that smooth levers cannot reproduce without coordination across pools. This is the strongest standalone argument for the CIP-0050 family.
• None of the three reform vectors (a₀, σ' clip, A redesign) are mutually exclusive — and the deepest one is missing from the conversation. The CIP discussion treats σ' clipping as the only available primitive. A revision of A itself — removing the $\nu^2$ outer size penalty, or replacing the inner $\pi[1-\pi(1-\nu)]$ factor with a kernel that doesn't cube small pools at full commitment — would be the most structural way to repair the pledge signal at the right end of the formula. No CIP currently in scope proposes this.

The two σ'-clipping candidates are evaluated on their own terms in the per-CIP files. The framing above is a reading aid, not a verdict — it reframes "is CIP-0050/0037 the right reform?" as "is σ' clipping the right layer of intervention?".

Status: Active 2026/04/23. Subfolder of ../README.md. Candidates that act on the stake-cap layer of the Cardano reward pipeline.