Studio b. LMMI · Open Methodology
Studio b. Lower Middle Market Direct Lending Index
The Studio b. LMMI exists because the benchmarks didn't. Lower middle market direct lending is one of the largest and least-documented corners of private credit. The methodology is open. The construction rules are published here. The quarterly index print and weekly market commentary are free. Detailed data — the full constituent universe, historical CSV, and quarterly DD-grade brief — is subscription-priced for institutional researchers, journalists, and allocators. Verified accredited LPs access the backtester through invest.b.studio. Three tiers, one methodology, one universe.
There is a segment of the U.S. private credit market that no publicly-accessible benchmark has ever measured directly: companies with $5–25M EBITDA — too large for the SBA's small-business programs, too small for Business Development Companies that target the upper lower middle market. This is the segment where Capital, by Studio b. invests, and where Studio b.'s alpha thesis lives. It is also, by any empirical measure, the segment with the fewest credible lenders, the least price discovery, and the most durable spread premium.
Studio b. LMMI exists to model this segment. It is a multi-anchor interpolated index, not a direct-observation index. No single public data source covers the $5–25M EBITDA band. The index synthesizes it from two boundary observations — the SBA 7(a) program (small-end, sub-$5M) and the publicly-registered BDC universe (upper-LMM, $25–100M) — plus a cross-jurisdictional recovery anchor from UK Companies House workout records. The interpolated middle, derived from those anchors, is the headline LMMI series. The model is the product.
Architecture overview — three anchors, one modeled middle:
What this version delivers. v0.4 implements the BDC upper anchor (v0.1), the SBA 7(a) lower anchor (v0.3), and the UK Companies House recovery anchor (v0.4). The complete multi-anchor index, with the $5–25M modeled middle as its headline series, ships in a v0.x revision as the interpolation model is developed.
The model and the fund. The Studio b. LMMI is not a passive observation index. It is the live methodology that underwrites Studio b.'s investment strategy in the same segment. The methodology is open because the decision logic has to be auditable: every version of the model is published, every revision is changelogged, every material change is footnoted. Verified accredited investors access a backtester that runs the same model — they apply their own assumptions to historical outputs and see what the strategy would have returned. The model evolves; the strategy retunes; the backtester reflects the current state. The methodology is the marketing because it is the strategy.
On CDLI-LMM. The Cliffwater Direct Lending Index Lower Middle Market is the institutional benchmark for the BDC-observable segment. CDLI-LMM observes what BDCs disclose, full stop — it is a direct-observation index of the upper LMM band, and a rigorous one. Studio b. LMMI uses CDLI-LMM as a reference input to the upper anchor, not as a competitor. They are different products: CDLI-LMM observes the BDC universe; Studio b. LMMI models the segment below it. Concordance with CDLI-LMM (currently 80% on the BDC-anchor classification, see §06) validates the anchor's accuracy.
On openness. The methodology is open. Every version publishes its construction rules, weighting math, calibration scope, and data lineage on this page. The index evolves transparently in public: each version adds anchors, refines the model, and documents what changed and why. The open methodology is how the index earns the right to be used in LP diligence and fund comparison.
Studio b. LMMI is built from multiple data anchors, each covering a distinct segment of the LMM credit market. v0.1 implements the BDC anchor; v0.3 adds the SBA 7(a) lower anchor; v0.4 adds the UK Companies House recovery anchor. The interpolation model that synthesizes all three into the headline LMMI series is pending v0.x.
§02a · BDC Anchor — Upper Band ($25–100M EBITDA) v0.1 — implemented
The BDC anchor consists of Business Development Companies registered under §54 of the Investment Company Act of 1940 that file Schedules of Investments with the SEC via 10-K and 10-Q filings.
Eligibility criteria for a given quarter:
Eligibility is computed per quarter. A BDC that misses a quarter due to filing delay is excluded for that quarter; no interpolation is applied.
v0.2 change: The 80% senior-secured floor (v0.1 criterion) has been removed. No CDLI-LMM equivalent gate exists; the v0.1 floor excluded OBDC at 79.3% in Q4 2025 (0.7pp boundary effect), producing an approximately 485 bps artifact in the Q4 2025 asset-weighted spread. Removing the floor restores OBDC to the eligible universe for Q4 2025.
Current universe (v0.2): 10 primary adapters — ARCC, GBDC, MAIN, PSEC, FSK, HTGC, PRSP, TCRD, BXSL, OBDC — covering approximately 85% of public BDC AUM by fair value. OBDE (merged into OBDC 2025-01-13) is retained as a deprecated historical adapter. Full universe expansion to 30+ BDCs is planned for v0.x. Adapters, mandate classifications, and source filing accession numbers are documented in the wdli_mandate_prior_seed.csv curation file.
What the BDC anchor does not include: private BDCs, insurance-company-affiliated direct lenders, CLOs, separately managed accounts, or any fund not subject to SEC filing requirements. The BDC anchor represents the observable upper band of LMM direct lending — it is not a claim about total LMM credit activity.
§02b · SBA 7(a) Anchor — Lower Band (sub-$5M EBITDA) v0.3 — implemented
The SBA 7(a) FOIA loan microdata provides the lower boundary observation for the LMMI model. It captures the floor of US small-business credit: origination volume, charge-off curves, sector distribution, and vintage-cohort default behavior for companies well below the BDC universe's borrower profile. The SBA anchor serves two roles in the eventual interpolation model: (1) constraining the lower terminus of the $5–25M modeled-middle series, and (2) providing the pre-2010 synthetic calibration signal used in the horizon extension (§05).
EBITDA proxy. SBA 7(a) loans are sized by loan amount, not EBITDA. The sub-$5M EBITDA classification is an inferred proxy based on loan size — loans at or below $5M loan amount presumptively serve companies with sub-$5M EBITDA. This is a documented inference, not a direct EBITDA measurement.
Two published series. The SBA anchor publishes two series per quarter, mirroring the BDC anchor's structure: asset-weighted (each loan weighted by origination amount) and equal-weighted (each loan weighted equally, more stable across loan-size mix changes).
Observation window: 2010-Q1 through current. The implied_spread_bps column is populated from 2010-Q1 onward. Pre-2010 rows have NULL implied_spread_bps per source-CSV limitation — FY2009 is ~89% populated; FY1992–FY2008 are 0–11% populated (structural SBA FOIA gap). See §02b-i for the rate imputation method that addresses the pre-2008 gap.
Implied spread benchmark. Computed as origination interest rate minus the WSJ US Prime Rate (FRED series PRIME) averaged over the calendar quarter of origination. Prime is used throughout the observation window (rather than switching to SOFR post-2018) to maintain a consistent benchmark series.
Pre-2010 calibration unchanged. The SBA calibration regression in the horizon extension (§05) — OLS of year-over-year LMM AUM growth on SBA charge-off rates — is unchanged by v0.3. The v0.3 anchor is an additive, parallel series. It does not replace or modify the calibration logic.
§02b-i · SBA Pre-2008 Interest Rate Imputation v0.5 — implemented
The SBA 7(a) FOIA historical archive files for FY1991–FY2007 leave the initialinterestrate field entirely empty — a structural gap in the source disclosure. FY2008 is ~10.9% populated; FY2009 is ~95.8% populated; FY2010 onward is 100% populated. Without imputation, the implied_spread_bps column in the SBA anchor is NULL for all pre-2008 rows.
Formula. Imputed rate = FRED PRIME (WSJ US Prime Rate, FRED series PRIME, forward-filled to origination month) + 2.75 pp constant spread.
Spread source. The 2.75 pp spread is the statutory maximum variable-rate spread for SBA 7(a) loans with terms greater than seven years (SBA 13 CFR 120.214). Calibrated as the empirical median spread for FY2008–FY2010 actuals — the earliest vintages with dense observed rates. Validation:
Coverage. Applied to all FY1991–FY2007 SBA 7(a) loans (1.04M rows with NULL initial_interest_rate). The 504 program carries no reported rates at any vintage; all 504 rows receive imputation_method = 'unknown'.
Confidence flag. Two columns added to panel.loan_dim: initial_interest_rate_imputed (observed where available, imputed otherwise) and initial_interest_rate_imputation_method ∈ {observed, fred_prime_spread, forward_fill, unknown}. Downstream consumers must check this flag before treating imputed rates as observed. The constant-spread assumption does not preserve observed spread variance — the imputed series is a centrist estimate under statutory constraints, not a distributional reconstruction. Required disclosure for any analysis citing pre-2008 SBA rates: "Pre-2008 SBA 7(a) interest rates are model-imputed (FRED PRIME + 2.75 pp constant spread), not observed. See methodology §02b-i."
Horizon disclosure alignment. Imputed rates extend rate-based computations within the SBA anchor's existing observation window. They do not extend the index horizon beyond the 2006 live observation floor in §05, which is unchanged.
§02c · UK Companies House Recovery Anchor v0.4 — implemented
UK Companies House administration and liquidation records provide a cross-jurisdictional stress reference for LGD estimation. US-only public sources (BDC filings, SBA data) cannot supply a direct LGD observation for the $5–25M EBITDA segment; the Companies House workout database — covering a structurally comparable small-to-mid-market lending segment in a jurisdiction with robust filing requirements — provides a partial but useful cross-check.
What this anchor measures. Vintage-cohort recovery curves: for each year of insolvency inception, the fraction of registered secured charges against the company that were ultimately satisfied after the insolvency event. Charge satisfaction is used as a binary proxy for recovery — 1 minus the recovery rate approximates LGD. The anchor does not measure direct recovery distributions or administrator payment amounts.
Size band filter. Companies House does not report EBITDA. The proxy for the £5–25M EBITDA-equivalent band is the maximum registered charge amount per company: companies with maximum charges between £2M and £50M GBP are classified as "size_filtered." Both "size_filtered" and "all_sizes" bands are published. 86% of companies have null charge amounts and are retained in "all_sizes" only.
Sample coverage. 4,442 distinct UK companies with a confirmed insolvency event in the Companies House charges register, spanning 1986–2026. Annual vintage cohorts from 2018 onward have adequate sample sizes (n ≥ 10 per year) for calibration use. Pre-2018 cohorts are published but flagged as "small_sample" where n < 10. LGD reference range: blended all_sizes recovery rate for 2018–2022 vintage cohorts falls within the UK Insolvency Service published range of 20–60 pence per pound.
Cross-jurisdictional disclosure. This anchor is UK-only. It is used as a Bayesian prior on LGD for the $5–25M modeled-middle segment. All published surfaces disclosing this anchor include: "UK insolvency data; cross-jurisdictional calibration reference; not a US LMM observation." Per-cohort N disclosed in the published artifact; any published chart citing cohorts with N < 10 must include: "N < 10; small sample."
§02d · The Modeled Middle — $5–25M EBITDA implementation pending v0.x
The headline LMMI series is the interpolated middle segment: $5–25M EBITDA borrowers, between the SBA-visible lower band and the BDC-visible upper band. No single public data source observes this segment directly. The modeled middle is derived from the three anchors above via an interpolation model that synthesizes the boundary observations into an estimated credit-conditions series for the unobservable interior.
The interpolation model is the core intellectual property of Studio b. LMMI. Its architecture — functional form, anchor weighting, calibration methodology, and uncertainty quantification — will be published in full when the model ships in v0.x. Until then, v0.4 publishes the three anchors only, with this section reserved for the interpolation design.
Every input to the Studio b. LMMI is a free public source. No paid data is required for v0 index computation.
| Source | Purpose | Cadence |
|---|---|---|
| SEC EDGAR BDC 10-K filings (XBRL + MD&A) | BDC mandate classification, EBITDA proxy, annual holding inventory | Annual |
| SEC EDGAR XBRL portfolio disclosures (10-Q) | Quarterly per-holding fair values | Quarterly |
| IRS Statistics of Income — Pub. 1053 corporate source book | NAICS sector-level LMM penetration rates by revenue decile | Annual |
| Census Bureau Statistics of U.S. Businesses (SUSB) | Firm-size distribution by sector | Annual |
| SBA FOIA 7(a) loan microdata | Charge-off curves for pre-2010 synthetic calibration; PD/LGD reference | Annual |
| NBER recession indicators | Macro conditioning in calibration model | Quarterly |
| FRED macro series | Credit spread and rate environment inputs | Monthly / Quarterly |
Live observation window: 2006–2025 (19 years). BDC SOI coverage before 2010 is sparse; the pre-2010 segment is a model-implied synthetic extension, not live observation. The 2006 floor is not arbitrary: SBA FOIA loan-level microdata nominally extends to FY1991, but the pre-2006 portion carries documented chargeoff rate anomalies (observed range 55–92% vs. 2–8% in the post-2006 panel). The Studio b. LMMI treats 2006 as the live observation floor and discloses any pre-2010 extension as synthetic. See §05 for the required disclosure language that must accompany any chart or claim referencing a longer horizon.
This section describes the v0.1 BDC-anchor computation. The middle-interpolation model — the logic that synthesizes the $5–25M EBITDA headline series from the three anchors — ships in a v0.x revision and will be documented here when it does.
BDC-anchor classification method: Each holding in each BDC's Schedule of Investments is classified as LMM or UMM (upper middle market) using a three-layer Bayesian ensemble. The output of each layer is a posterior probability p_lmm in [0, 1]. The BDC-anchor LMM/UMM boundary is EBITDA $50M — this boundary applies to the observable BDC universe classification, not to the eventual modeled-middle index scope.
Layer 1 — BDC mandate prior. Each BDC is assigned a mandate class based on its disclosed or stated-target portfolio-average borrower EBITDA, consistent with Cliffwater's CDLI-LMM definition: lmm_pure if portfolio-average EBITDA < $40M, umm_pure if portfolio-average is consistently above $75M, or mixed if the portfolio-average falls between $40–$75M or if the EBITDA test is inapplicable (e.g., venture BDCs). Prior probabilities: lmm_pure → 0.95, umm_pure → 0.05, mixed → 0.50. The 5% residual on lmm_pure accounts for hold-duration drift and opportunistic UMM investments. v0.2 change: The v0.1 rule classified BDCs by mandate-language upper EBITDA bound; v0.2 uses actual portfolio-average EBITDA, mirroring how Cliffwater operates CDLI-LMM. Five adapters reclassified: GBDC (lmm_pure → mixed), HTGC (lmm_pure → mixed, venture special case), PSEC (lmm_pure → umm_pure), TCRD (lmm_pure → umm_pure), OBDE (lmm_pure → umm_pure, deprecated). OBDC added as new adapter (mixed). Curation documented in wdli_mandate_prior_seed.csv with source filing accession numbers and reclassification rationale.
Layer 2 — Per-holding ensemble. For each observable holding, the mandate prior is updated using three signals: loan size (M2, EBITDA proxy — loans under $25M fair value presumptively LMM at 80% weight, over $100M presumptively UMM); interest rate (M3, floating-rate loans with spread above SOFR+500bps presumptively LMM); and NAICS sector (M5, sector-level LMM penetration rate from IRS SOI, normalized by revenue-size decile). Signals are combined via naive-Bayes update on the Layer 1 prior.
Layer 3 — Expected-value aggregation. The expected LMM AUM for each holding is fair_value_usd × p_lmm_posterior. The quarterly index value is the sum of expected LMM AUM across all eligible BDC holdings.
Primary series: Asset-weighted total LMM AUM. The headline series is WDLI(t) = Σᵢ[fair_valueᵢ(t) × p_lmmᵢ(t)] across all eligible constituents in quarter t.
Secondary series: Equal-weighted. Each BDC is given equal influence regardless of size, reducing concentration from large constituents such as ARCC.
Confidence bands: 95% analytic confidence bands treat each holding's LMM classification as an independent Bernoulli trial with parameter p_lmm. Bootstrap confidence bands are scheduled for v0.1.
Rebalance cadence: Quarterly, coinciding with 10-Q filing cycles. Additions and deletions are applied when a BDC first meets or fails the eligibility criteria in a given quarter; no look-ahead is applied.
Non-accrual treatment: Per-holding non-accrual flags are extracted from BDC SOI filings. Non-accruing holdings are retained in the index at their reported fair value; no mark-to-zero override is applied. The non-accrual flag is available as a decomposition field. As of v0.1, non-accrual extraction covers 2022–2025 for 7 BDCs with validated accuracy (median |error| 0.2 percentage points against disclosed rates). Pre-2022 non-accrual extraction is in progress and not yet included in the index.
What this index does not measure: Total returns, dividend reinvestment, NAV-based performance. The v0 headline series is total LMM AUM in the eligible BDC universe — a level series, not a return series. The total-return series requires NAV-based ingest from BDC Statements of Operations and is deferred to v0.x.
The open methodology is the marketing. But precision about what's observed versus what's modeled is what makes it usable. — Studio b. LMMI, methodology preface
The Studio b. LMMI publishes two distinct horizons. They are not interchangeable.
Live observation: 2006–2025 (19 years). Based on dense BDC 10-K/10-Q coverage and SBA 7(a) FOIA microdata with credible chargeoff rates. This is the conservative floor and the only figure that should appear unqualified in any LP-facing tear sheet, fund deck, or public chart axis.
Modeled extension: approximately 1992–2025 (~33 years). The live observation window plus a pre-2010 SBA-calibrated synthetic segment. The synthetic segment uses OLS regression of year-over-year LMM AUM growth on SBA charge-off rates, with the calibration sample covering 2022Q4–2025Q4 (n=9 quarterly observations, R²=0.477, slope −11.19). Calibration coefficients are persisted to wdli.calibration_coefficients with the git SHA at each run. The pre-2010 segment carries an explicit ±2σ proxy band — where σ equals the residual standard error of the overlap regression — and must be rendered visibly on any chart displaying pre-2010 data alongside live data.
The 33-year modeled horizon may be referenced only alongside a footnote disclosing (a) that the pre-2010 segment is model-implied, not observed BDC AUM, and (b) that the calibration sample is n=9 quarterly observations. The required disclosure language for any public surface:
DLI live observation: 2006–2025 (19 years), based on dense BDC 10-K/10-Q coverage and SBA 7(a) FOIA microdata with credible chargeoff rates. Pre-2010 values shown are SBA-calibrated synthetic projections, carry an explicit ±2σ proxy band, and are derived from a calibration sample of n=9 quarterly observations. Pre-2006 SBA FOIA microdata is excluded from live observation due to documented coverage gaps.
The Studio b. LMMI runs two validation checks before each quarterly publication.
MD&A back-test. Model-estimated portfolio EBITDA is compared against EBITDA disclosed in BDC 10-K/10-Q Management Discussion & Analysis sections. Portfolio-weighted RMSE is computed using XBRL-tagged disclosures only (the more reliable of XBRL-tagged and regex-extracted data). Gate thresholds: RMSE below 15% on tagged disclosures → publish; 15–20% → trigger BLS QCEW sector data ingest for cross-check; 20% or above → trigger RMA/BizMiner industry financial data ingest. A publication blocked by RMSE does not proceed until the cascade ingest chip is reviewed.
CDLI-LMM concordance. A manual fair-use spot-check of Studio b. LMMI LMM classifications against publicly-visible CDLI-LMM constituent disclosures on Cliffwater's public fact pages. No automated scrape is performed. Current concordance rate: 80% (4 of 5 sampled BDCs, point-in-time, May 2026). The gap is ARCC: the Studio b. LMMI classifies ARCC as mixed (partial LMM inclusion); CDLI-LMM includes ARCC as a full constituent given ARCC's scale. This is a known and documented divergence, not a defect. The v0.2 reclassifications (PSEC, TCRD, OBDE from lmm_pure → umm_pure; GBDC, HTGC from lmm_pure → mixed) are designed to improve long-term concordance by aligning with Cliffwater's portfolio-average EBITDA test.
Known uncertainty: Approximately 12% of holdings in the current universe fall within the borderline EBITDA range ($40–60M) where LMM/UMM classification is genuinely ambiguous. For these holdings, the ensemble returns a p_lmm in [0.3, 0.7]. If more than 20% of holdings are ambiguous in any quarter, a methodology committee review is triggered.
Issuer: Benchmarks, by Studio b. LLC, a dedicated indexing entity legally and operationally separate from any fund advisory entity. All public artifacts attribute the index to Benchmarks, by Studio b. LLC.
Methodology author: Kevin Bibelhausen (v0 single-author).
Methodology committee: To be constituted at the external audit milestone, targeted Q1 2027. v0 uses single-author governance.
Conflict-of-interest controls: Any Capital, by Studio b.-affiliated fund must attest to a pre-publication trading freeze before any quarterly publication proceeds. Attestation is logged in the publication's lineage.json with operator email and timestamp. The attestation flag cannot be set by the publication pipeline itself — it requires a human operator.
Versioning: Methodology revisions increment a semantic version. The current version is 0.1. v1.0 is reserved for post-IOSCO-external-audit release, targeted Q1 2027 (~$20–40K engagement scope). All revisions are git-tagged wdli-v<x.y> in the wasala repository.
Version-bump policy.
lmmi.b.studio/methodology/v<x.y> (this routing convention starts after the first revision; v0.1 is the live state until v0.2 publishes). When the §09 changelog grows past a small number of entries, it will move to a dedicated page at lmmi.b.studio/methodology/changelog.IOSCO self-attestation: v0.1 is self-attested against IOSCO Principles for Financial Benchmarks P1, P3, P4, P7, P8, P10–P12. P2, P5, P6, and P9 are not applicable because the Studio b. LMMI derives entirely from public regulatory filings — no data submitters, no panel members. The full self-attestation document is available at lmmi.b.studio/methodology/iosco-v0-self-attestation.
Every Studio b. LMMI quarterly publication is independently reproducible. Given the methodology_version, the git_sha of the dbt project at materialization, and the ingest_run_ids for each upstream source, any third party can re-download the source files (all are public regulatory filings with stable URLs), run dbt build --full-refresh against a fresh Postgres instance, and obtain an identical index value within floating-point rounding tolerance.
Each quarterly publication includes a lineage.json artifact with the following fields: git_sha, methodology_version, publication_timestamp, period_end, ingest_run_ids per source family, dbt_run_id, and freeze_attested_by.
The Studio b. LMMI publishes across three access tiers. Methodology and the headline quarterly print are open. Detailed institutional data is subscription-priced. LP-track access is gated by accreditation.
Free — open methodology and headline print:
lmmi.b.studio/methodologySubscription — institutional data and detail (in development, target Q1 2027):
The institutional tier is priced for family offices, fund-of-funds, sell-side analysts, and journalists who need depth beyond the headline print. Pricing and launch timing publish alongside the external IOSCO audit, targeted Q1 2027 (see §07). To be notified, subscribe to the newsletter and indicate institutional interest.
LP track — free for verified accredited investors:
How the model tunes the fund.
Studio b. operates a fund — Studio b. LMM Bridge Fund I — that invests in the same $5–25M EBITDA segment the LMMI models. The fund's flagship product structure is itself a model output: parameters chosen because the model's analysis of segment behavior shows where structural alpha exists. As the model evolves across versions — adding anchors, refining interpolation, tightening confidence intervals — the strategy retunes. The published methodology IS the audit trail of those decisions.
This is what makes the open-methodology promise load-bearing. An LP underwriting Studio b. LMM Bridge Fund I can verify the model that designs the strategy. A non-LP subscriber can read the same methodology and run the same backtester (subject to accreditation gating). The strategy is not opaque; the model that produces it is the open product.
Specific fund parameters — interest rate, term length, conversion structure, valuation floor — are set during fund formation and disclosed in the offering documents. They are downstream of the model's segment analysis. Future fund vintages may set different parameters as the model's confidence intervals tighten and segment characterization improves. Each set of parameters is an instance of the model's output at a point in time.
The wasala dbt project (the Studio b. LMMI's full source code) is privately held pending external IOSCO audit, targeted Q1 2027. Replication scripts for all public ingest sources — SEC EDGAR, SBA 7(a), UK Companies House — are available on request at invest@b.studio. Every quarterly publication is independently reproducible from public source files given the methodology_version, git_sha, and ingest_run_ids recorded in that quarter's lineage.json.
Questions and methodology feedback: invest@b.studio.
v0.5 · 2026-05-06 · SBA pre-2008 interest rate imputation
initial_interest_rate in SBA FOIA source. FY2008+ uses observed rates.initial_interest_rate_imputation_method enum on panel.loan_dim ∈ {observed, fred_prime_spread, forward_fill, unknown}. Downstream consumers must check this flag.0031_sba_rate_imputation.sql + dbt loan_dim post-hook. Applied to prod before PR merge.v0.4 · 2026-05-05 · UK Companies House recovery anchor — implemented
v0.3 · 2026-05-05 · SBA 7(a) lower anchor — implemented
wdli.sba_anchor_quarterly dbt model publishes quarterly SBA credit conditions for the sub-$5M EBITDA lower band.implied_spread_bps populated from 2010-Q1 (FY2010 SBA FOIA data carries 100% interest-rate coverage).sba_anchor key added to quarterly publication artifact. sba_ingest_run_id added to lineage.json per §08 reproducibility contract.v0.2 · 2026-05-05 · BDC-anchor methodology refinements
v0.1.2 · 2026-05-05 · Documentation patch — model→product→backtester loop framing
v0.1.1 · 2026-05-05 · Documentation patch — multi-anchor architecture clarified
wasala/wasala/wdli/ is unchangedv0.1 · 2026-05-04 · Initial publication
lmmi.b.studio/methodologyv1.0 reserved for post-IOSCO-external-audit release, targeted Q1 2027