v26 - Lending Regimes & the Freedom Tradeoff

In plain language

July 10, 2026. Companion to RESULTS - v26.0 Lending Regimes.

The question

You want EDEN to stand on Freedom — including the freedom to let banks and people make whatever loan deals they want, and to take on compound interest if they choose the risk. The worry was whether that freedom bites anyone. We modeled four rulebooks — total freedom, freedom-minus-debt-spirals, a rate cap, and profit-sharing instead of loans — and measured what each does.

What we found — and your instinct mostly holds

The floor makes "let banks do whatever" surprisingly fair. In today's world, no-rules lending means payday loans at 400% because a desperate person has no choice. In EDEN, everyone has the essentials floor, so nobody has to borrow to survive — which means every borrower can say "no thanks" and walk. That single fact drags the free-market lending rate down from about 36% (what a desperate market charges) to about 12.5% (what a market of people-who-can-walk-away charges). So freedom plus a floor isn't predatory the way freedom without a floor is. Your core intuition is right.

And by an even-handed scoring, total freedom actually wins. When we score "how much freedom" as choice-of-deals times freedom-from-getting-trapped, the no-rules option comes out slightly ahead — because the floor keeps the downside small. So the free-market lean is genuinely defensible here, not just ideologically.

The one honest catch

Full freedom still leaves a small group trapped — about 4% of borrowers, almost all of them the financially fragile who took a loan, then hit a bad stretch, and watched compound interest snowball faster than they could pay. They chose the loan freely; what they lack is a way out. And here's the subtle part: a person trapped in a debt spiral is one of the least free people in the system — their future earnings are pledged before they earn them. So "freedom" cuts both ways: the freedom to sign a compounding loan can cost someone the freedom to control their own future.

The guardrails shrink that trapped group step by step (4% → 3% → 1.6% → 0%), but only the profit-share option removes it entirely — because with profit-sharing there's no debt to spiral: if the venture fails, the funder loses, not the borrower.

The idea that keeps your freedom and fixes the catch

You don't have to choose between "total freedom" and "protect people." The finding points to a third path that's more freedom, not less:

• Keep the free market — banks and people can offer and take any loan, including compound interest. Default is laissez-faire.
• Add a guaranteed right to exit — anyone caught in a spiral can freeze it (stop the compounding) or discharge it against their floor, whenever they choose. This bans nothing — you're still free to take the compounding risk — it just guarantees you can always get out.
• Offer profit-sharing alongside loans — so someone who'd rather share risk than owe fixed interest can pick that instead. More choices, not fewer.

That's freedom that includes the freedom to leave — which the trapped-borrower finding says is the freedom that actually matters. A choice you can never escape isn't really a free choice; it's a cage you walked into.

The bottom line

Let the market be free — the floor already keeps it honest. Just make sure "free to choose a risk" always comes with "free to walk away from it," because the people who get hurt aren't the ones who chose badly, they're the ones who chose freely and then couldn't get out. Whether to add that exit-right, or to stay purely hands-off, is your values call — the numbers say it costs you almost no freedom to add it, and buys the fragile a lot.

Figures

Technical results

Run: July 10, 2026. Spec: v26 SPEC - Lending Regimes (registered).md — bars Q0–Q5 fixed before code. Engine: lending_regimes_sim.py (reduced-form credit-market ABM; deterministic given seeds 7+11); committed: results_v26.json, fig_v26_lending.png. Owner-directed to price the Freedom pillar's laissez-faire lean against guardrailed alternatives, modeled even-handedly. Every number traces to results_v26.json.

Verdict in one line: your instinct is largely right — inside EDEN's floor, letting banks do whatever they want is far tamer than laissez-faire anywhere else (the floor cuts lending rates from ~36% to ~12.5% because every borrower can walk away), and on an equal-weight "total freedom" measure full freedom even comes out ahead; the honest cost of pure freedom is a small debt-trap tail (~3.8% of borrowers, concentrated in the financially fragile) that no interest-debt rule fully removes — only equity-share funding does — so the real decision is how much you weight one person's freedom-to-choose-compound-risk against another person's freedom-from-entrapment.

The four regimes (seed-averaged)

*EQUITY's "rate" is the equivalent funder take; it structurally can't fund pure consumption borrowing (no return to share) — a real limitation.

*Two lender columns, two different things: the top-1% share is scale-invariant and therefore identical across the three interest regimes by construction (a disclosed tie, not a finding); lender growth (total lender capital, final÷initial) is the accumulation magnitude that actually separates them and is what Q4 is scored on (see F3).

Bars: Q0 sanity — conservation residual ~2e-7 of volume ✓ and structural aggregates agree <2% across seeds ✓, but scored over ALL reported metrics as registered the unqualified 3% seeds clause is breached by the two tail statistics — FAIL (finding; v5 re-score July 12 — tail-sampling noise, headlines seed-averaged and unaffected) · Q1 credit flows (all regimes fund 100% of positive-EV projects) PASS (structural, cannot fail — now flagged in-JSON, see F4) · Q2 walk-away discipline PASS · Q3 NOCAP-is-trap-free FAIL (finding) · Q4 FREE accumulates fastest on the r>g magnitude (lender_growth 3.46×) PASS (computed; top-1% share is a scale-invariant tie, disclosed) · Q5 two-freedoms descriptive — freedom-from-entrapment computed, contract-freedom stipulated (disclosed).

Findings

F1 — The floor tames laissez-faire lending (Q2 — your hypothesis, validated). The single most important result: with EDEN's guaranteed floor, the equilibrium lending rate under full freedom settles at ~12.5%; strip the floor out (the desperation counterfactual) and the same free market charges ~36% — a 24-percentage-point difference. The mechanism is exactly the one you reasoned to: because nobody must borrow to survive, every borrower has a credible walk-away, and that outside option caps how much any lender can extract. So "let banks do whatever they want" produces something close to fair lending in EDEN, where the same rule produces predation in a world without a floor. Freedom and a floor are complements, not opposites. This is the strongest argument for your lean, and it holds.

F2 — Full freedom still leaves a debt-trap tail, and only equity removes it entirely (Q3 — honest FAIL). I registered the expectation that banning capitalized arrears (NOCAP) would make debt traps ~impossible. It doesn't — it halves them (3.8% → 2.8%) but a chronically-fragile borrower can still accumulate frozen arrears past the trap threshold. The trap rate falls monotonically as guardrails increase — FREE 3.8% → NOCAP 2.8% → RATECAP 1.6% → EQUITY 0.0% — and only equity funding, which carries no fixed debt claim at all (if the project fails the funder loses, the borrower owes nothing), is structurally trap-proof. So the honest picture is a gradient, not a switch: every guardrail helps, none of the interest-debt ones fully protect the fragile, and the only complete protection is the non-debt instrument. The traps are concentrated in the ~25% financially-fragile population — the people the Freedom pillar is least able to see, because they entered the contract freely and it's the exit they lack.

F3 — The accumulation worry has two levers: interest-vs-equity AND capitalizing-vs-frozen (Q4). Two distinct measures matter, and separating them is the point of the Q4 recode. (a) Concentration — the top-1% lender wealth share — is ~15.0% for all three interest regimes and ~11.9% for equity; but this share is scale-invariant, so uniform interest compounding leaves it identical to the 13th decimal across FREE/NOCAP/RATECAP (0.1501242016 — a tie by construction; the earlier PASS rested on a meaningless 1e-9 epsilon over it, now disclosed). (b) Accumulation magnitude — total lender capital, final ÷ initial (lender_growth) — is the r>g signal that actually distinguishes the regimes: FREE 3.46× > NOCAP 2.88× > RATECAP 2.15× > EQUITY 1.29×. FREE, compounding on the full arrears-capitalizing base, grows the passive-lender-wealth pool fastest — by a real ~20% margin over NOCAP — and that computed magnitude (not the share tie) is the honest basis for the Q4 PASS. Equity grows only with realized project returns, tracking real value created. So both levers are real: fixed-claim vs shared-risk (interest vs equity, the ~15% vs ~12% concentration gap) and capitalizing vs frozen arrears (FREE vs NOCAP, the 3.46× vs 2.88× accumulation gap). The magnitudes are stylized (the frozen-arrears regimes carry a fixed 0.85 realization factor in the reduced form), so read the ordering as robust and the exact multiples as illustrative.

F4 — Credit isn't strangled by any regime (Q1). All four fund 100% of positive-expected-value investment projects at these dials — so none of the guardrails "starve lending" for productive projects. The differences are in who bears the risk and what happens to the fragile, not in whether good projects get funded. (EQUITY's one real gap: it can't serve pure consumption-smoothing borrowing, which needs a debt instrument — so an all-equity system would leave a consumer-credit hole the floor only partly fills.)

Honest disclosure (Q1 is structural, not a finding): the "100% of positive-EV projects funded" result cannot fail by construction. Investor willingness-to-pay is defined as the project return itself (wtp = R), and the accept rule funds any deal with offer ≤ wtp and positive EV — so every positive-EV project is funded by definition of how demand and acceptance are modeled. Q1's PASS is therefore a property of the reduced-form setup, not empirical evidence that credit is un-strangled; read it as "the model imposes no funding bottleneck," not as a discovered result. (Q1 was outside the recode scope, so the bar itself is unchanged; this line is the honest disclosure the register asked for, not a faked failure.)

F5 — The two faces of freedom (Q5 — the values call, quantified not decided). Freedom-from-entrapment (1 − trap rate), computed from the sim, ranks EQUITY 1.00 > RATECAP 0.984 > NOCAP 0.972 > FREE 0.962. Contract-freedom (FREE 1.00 > NOCAP 0.98 > RATECAP 0.80 > EQUITY 0.60) is owner-stipulated, not computed — the engine models no origination refusal for the interest regimes (RATECAP re-prices to the cap, NOCAP only alters servicing), so it yields no "share of willing contracts permitted" that could distinguish them; only EQUITY refuses origination. These scores are disclosed as priors (stipulated_not_computed), and everything downstream of them is stipulation-dependent, not a sim verdict. On an equal-weight product of the two, FREE edges it (0.962) because the floor keeps its entrapment cost small — defensible on the numbers in EDEN specifically, given the stipulated contract-freedom. The weighting itself is a genuine values choice the sim cannot make: a trapped debtor is arguably maximally unfree (their future labor is pledged).

Correction to the earlier draft (the "~1.5×" was wrong). Using the committed values with a weighted-geometric score (contract-freedom × freedom-from-entrapment^w), NOCAP overtakes FREE only at w ≈ 2.0×, RATECAP at w ≈ 9.8× (and EQUITY at ~13×). At w = 1.5× FREE still wins (FREE 0.944 vs NOCAP 0.939). So you must value freedom-from-entrapment at roughly twice contract-freedom before even the mildest guardrail (NOCAP) overtakes full freedom — a materially higher bar than the ~1.5× first reported. (These crossover weights inherit the stipulated contract-freedom, so treat them as "given the owner's contract-freedom priors," not as sim output.) The sim's real job is to show that (a) the entrapment gap between regimes is small because the floor does most of the work, and (b) the choice reduces to how you weigh one person's freedom to choose a compounding risk against another person's freedom to not be trapped by one.

What this implies for the Banking & Savings Layer spec (DR-17)

A defensible reading that honors the Freedom pillar: adopt FREE as the default — let banks and individuals contract however they wish, including compound interest — because the floor already disciplines it (F1), it funds everything (F4), and it maxes contract-freedom. Consider two freedom-preserving additions that cost almost no contract-freedom (both leave people free to choose compound-interest loans): 1. A universal right to exit — any borrower may convert a spiraling debt to frozen (non-capitalizing) arrears, or discharge against the floor, at will. This doesn't ban compound interest; it guarantees the exit that makes the choice free (turns FREE into "FREE with a rip-cord," recovering most of NOCAP's trap protection without removing anyone's choice). 2. Equity funding available alongside debt — offer revenue-share as a competing instrument (it already exists as the builder-royalty model), so borrowers who prefer shared-risk to fixed-debt can choose it. Pure market freedom: more instruments, not fewer.

This keeps the system laissez-faire by default while ensuring freedom includes the freedom to get out — which the fragile-borrower finding (F2) says is the freedom that actually protects people. But this is a recommendation for the spec, not a decision; the owner's call on the FREE-vs-guardrail weighting stands.

Honest limits

Reduced-form ABM with stated dials (fragile-share 25%, market-power 0.6, shock/project rates) — the directions are robust (floor disciplines rates; traps concentrate in the fragile; interest concentrates more than equity; credit isn't starved), the exact magnitudes (36% vs 12.5%, 3.8% traps, 15% vs 12%) are illustrative, not forecasts. The no-floor counterfactual's 80% consumer WTP is a stand-in for desperation, not a measured value. Lender-concentration is measured on a scale-invariant top-share, so within-interest differences are understated. And the deepest freedom question — whether a freely-chosen contract that ends your autonomy is "free" — is philosophy the sim can only illuminate with magnitudes, not resolve.

Run and written July 10, 2026 (Fable), owner-directed. Q3 reported as a FAIL against its registered expectation — the gradient finding is more useful than the switch I predicted. Modeled even-handedly; FREE was given its fairest showing and, on equal weighting, it earns it. Numbering checked against max before registering v26.

Gate-tightening EXECUTED July 11, 2026 (Backlog #4b, Opus 4.8) — addressing Verification v4 D13c. Physics preserved: every pre-existing leaf in results_v26.json is byte-identical to the July-10 run and the sim is deterministic run-to-run; only gate scoring and added diagnostics changed (verified by flatten-diff: zero changed leaves; only intended removals/additions). What each recoded gate now computes:

• Q0 — was hardcoded {"pass": true} ("conservation is structural"). Now COMPUTES both registered clauses from real state: (1) the debt-ledger conservation residual — principal disbursed + interest accrued = payments + still-outstanding + written-off — equals 1.96e-7 of volume (~0, tolerance 1e-4); (2) seed-agreement on the structural market aggregates (fund_share, mean rate, disbursed volume, lender_growth) = max 1.63% < 3%. Two small-count TAIL statistics — trap_rate (~11-14%) and the scale-invariant lender_top1_share (~17.9%) — exceed 3% as single-seed estimates and were disclosed, not bound. Re-scored July 12 (Verification v5 D4, action 4): the registered clause is unqualified ("seeds agree within 3% rel"), so binding only the aggregate subset was a narrowed reading — scored over ALL reported metrics as registered, Q0 is an honest FAIL (finding): conservation holds and the structural aggregates agree, but the two tail statistics breach the 3% band (committed per-metric under Q0.seed_spread_all_metrics; overall under Q0.seeds_agree_all_reported_metrics). Tolerance unmoved. Downstream verdicts unaffected — Q3 keys off sign (both seeds show traps in FREE/NOCAP and zero in EQUITY) and every headline is seed-averaged with both seeds committed. Whether the clause should be re-registered onto the structural aggregates (the defensible reading) is queued as an owner decision, not taken by the verifier.**
• Q4 — was PASS on a 1e-9 epsilon over regime values identical to the 13th decimal. The registered metric lender_top1_share is scale-invariant, so it is identical (0.1501242016) across FREE/NOCAP/RATECAP by construction — now disclosed as a tie. The r>g claim is re-scored on accumulation magnitude lender_growth (total lender capital, final÷initial): FREE 3.46× > NOCAP 2.88× > RATECAP 2.15× > EQUITY 1.29×, FREE fastest by a real ~20% margin. Genuine PASS, no flip (metric fixed, not the verdict).
• Q5 — contract-freedom scores {1.00, 0.98, 0.80, 0.60} were stipulated constants where the SPEC said "compute." freedom-from-entrapment (1 − trap rate) is genuinely computed and kept; contract-freedom is DOWNGRADED to a disclosed non-result (stipulated_not_computed: true + note) because the engine models no origination refusal for the interest regimes, so there is no sim quantity behind those numbers. total_freedom_proxy / best_total_freedom / the crossover weights are flagged depends_on_stipulated_contract_freedom and fully_computed: false. Descriptive gate (SPEC: "not a verdict"); pass reflects the computed clause with the stipulation disclosed.
• Q1 — outside recode scope; disclosed in F4 as structural (100% positive-EV funding is the willingness-to-pay construction, cannot fail by definition). No fake failure filed. v5 addendum (July 12, D5): the disclosure is now machine-readable — Q1.structural_cannot_fail: true + note in the JSON, so a harness reading the JSON alone sees the caveat (previously prose-only).
• F5 prose corrected — the "~1.5×" weighting crossover was wrong; on the committed values it recomputes to w ≈ 2.0× (NOCAP) / ~9.8× (RATECAP), and at 1.5× FREE still wins. Now traceable via Q5.crossover_weight_ff_over_cf in the JSON.
• Q3 — left untouched: honest FAIL finding, computed and correctly reported (F2).

No headline verdict changed — no PASS→FAIL flips, and FREE still wins the equal-weight two-freedoms proxy (given the stipulated contract-freedom). Executed under the gate-tightening house rules: never loosen a registered bar; failures are findings; do not fabricate; preserve physics.

Raw data

⬇ results_v26.json