EIKNA LESBIAN ROMANCE AUTHOR INDEX - v1.0 - SCORING ENGINE (PYTHON) Edition: August 2026. Data checked: 16 August 2026. Published at https://lesbianromancereads.com/best-lesbian-romance-authors/ WHAT THIS IS The runnable scoring engine behind the EIKNA Author Index, with the full methodology written into it as comments: the seven component weights, the log normalisation, the Bayesian shrinkage used for Reader Satisfaction, the eligibility rule, the missing-data rule, the momentum formula, and the expected output for the August 2026 edition. HOW TO RECOVER THE RUNNABLE FILE Every code line below is prefixed with a pipe and a space, because WordPress.com refuses to store recognisable source code. Strip the prefix: sed 's/^| //' eikna-author-index-v1-scoring-engine.txt > eikna_score_v1.py On Windows PowerShell: (Get-Content file.txt) -replace '^\| ','' | Set-Content eikna_score_v1.py Then delete these header lines from the top of the resulting file. INPUT DATA The engine imports AUTHORS from eikna_dataset_v1.py. Every raw input it needs is also present as a column in eikna-author-index-v1-scores.csv (gr_total, gr_over_1k, sg_reviews, amz_bsr, amz_cat_rank, amz_reviews, novels, first_year, rel_18m, outlets, award, ku, _v, _R, _adj, _vel), so the index can be rebuilt from that CSV alone if the dataset module is lost. WHAT THE SCORE IS NOT It measures reader reach, reception, catalogue strength, commercial visibility, editorial coverage and momentum. It does not measure literary quality, and it cannot be purchased. ================================================================ | | | """ | EIKNA Lesbian Romance Author Index - scoring engine v1.0 | Run: python3 eikna_score_v1.py | Outputs eikna_v1_scores.csv and eikna_v1_scores.json | """ | import json, math, statistics as st | from eikna_dataset_v1 import AUTHORS, EMERGING | | # =============================== METHODOLOGY =============================== | # WEIGHTS (sum = 100) | W = dict(reach=25, satisfaction=17, catalogue=15, commercial=15, | storygraph=10, community=8, momentum=10) | # NOTE: the original design allocated 10% to Community Attention including | # Reddit and BookTok. Neither could be measured consistently and reproducibly | # for this cohort, so the component was narrowed to *verified independent | # editorial coverage* and its weight reduced 10 -> 8. The freed 2 points went | # to Reader Satisfaction, the most statistically robust component. | | # ELIGIBILITY THRESHOLD (fixed BEFORE the cohort was scored) | # E1 >= 3 qualifying sapphic titles | # E2 >= 5,000 total Goodreads ratings across the qualifying catalogue | # E3 first qualifying sapphic title published 2023 or earlier | # E4 Data Confidence >= 60% | E_MIN_TITLES, E_MIN_RATINGS, E_MAX_FIRST_YEAR, E_MIN_CONF = 3, 5000, 2023, 60 | | TODAY_YEAR = 2026 | | def lg(x): return math.log10(1.0 + max(0.0, x)) | | def minmax(vals): | lo, hi = min(vals), max(vals) | if hi - lo < 1e-12: return [50.0] * len(vals) | return [100.0 * (v - lo) / (hi - lo) for v in vals] | | # --------------------------- DATA CONFIDENCE ------------------------------- | # Seven evidence slots, each scored 1 / 0.5 / 0: | # 1 Goodreads catalogue-level totals read directly | # 2 Goodreads top-book ratings + averages read directly | # 3 StoryGraph figure obtained | # 4 Amazon Kindle BSR + review count obtained | # 5 Career/publisher facts confirmed at author or publisher source | # 6 >= 2 independent editorial outlets verified | # 7 Release dates for the last 18 months confirmed to the day | for a in AUTHORS: | a['confidence'] = round(100.0 * sum(a['conf_flags']) / 7.0) | | # --------------------------- ELIGIBILITY ----------------------------------- | for a in AUTHORS: | reasons = [] | if a['novels'] < E_MIN_TITLES: reasons.append('catalogue < 3 titles') | if a['gr_total'] < E_MIN_RATINGS: reasons.append('< 5,000 Goodreads ratings') | if a['first_year'] > E_MAX_FIRST_YEAR: reasons.append('first title after 2023') | if a['confidence'] < E_MIN_CONF: reasons.append('data confidence < 60%') | a['eligible'] = len(reasons) == 0 | a['ineligible_because'] = '; '.join(reasons) | | C = [a for a in AUTHORS if a['eligible']] # scored cohort | print(f"Eligible cohort: {len(C)} of {len(AUTHORS)}") | for a in AUTHORS: | if not a['eligible']: | print(f" EXCLUDED {a['name']}: {a['ineligible_because']}") | | # ----------------------- MISSING-DATA RULE --------------------------------- | # A missing component input is imputed at the COHORT MEDIAN of that input, and | # the author's Data Confidence already carries the penalty for the gap. Median | # is by construction the neutral position - imputation never flatters. | def median_of(key, transform=lambda x: x): | vals = [transform(a[key]) for a in C if a[key] is not None] | return st.median(vals) | | # =========================== 1. GOODREADS REACH (25) ======================= | r1 = minmax([lg(a['gr_total']) for a in C]) | r2 = minmax([lg(a['top_books'][0][0]) for a in C]) | r3 = minmax([lg(a['gr_over_1k']) for a in C]) | for i, a in enumerate(C): | a['c_reach'] = 0.50 * r1[i] + 0.30 * r2[i] + 0.20 * r3[i] | | # ======================= 2. READER SATISFACTION (17) ======================= | # Rating-count-weighted mean across the author's best-evidenced titles, then | # Bayesian shrinkage toward the cohort prior: | # adjusted = (v*R + m*Cprior) / (v + m) | # v = ratings underpinning R; m = cohort MEDIAN of v (prior strength); | # Cprior = ratings-weighted mean rating across the whole cohort. | # Effect: a 4.7 from 80 ratings cannot outrank a 4.3 from 40,000. | for a in C: | v = sum(n for n, _ in a['top_books']) | R = sum(n * r for n, r in a['top_books']) / v | a['_v'], a['_R'] = v, R | m_prior = st.median([a['_v'] for a in C]) | num = sum(a['_v'] * a['_R'] for a in C); den = sum(a['_v'] for a in C) | C_prior = num / den | for a in C: | a['_adj'] = (a['_v'] * a['_R'] + m_prior * C_prior) / (a['_v'] + m_prior) | sat = minmax([a['_adj'] for a in C]) | for i, a in enumerate(C): a['c_satisfaction'] = sat[i] | | # ======================= 3. CATALOGUE STRENGTH (15) ======================== | # Deliberately NOT raw book count: the >1,000-ratings measure carries most of | # the weight, so a large catalogue nobody reads does not score. | c1 = minmax([lg(a['novels']) for a in C]) | c2 = minmax([lg(a['gr_over_1k']) for a in C]) | c3 = minmax([min(1.0, (TODAY_YEAR - a['first_year']) / 20.0) for a in C]) | for i, a in enumerate(C): | a['c_catalogue'] = 0.30 * c1[i] + 0.40 * c2[i] + 0.30 * c3[i] | | # ================= 4. AMAZON COMMERCIAL VISIBILITY (15) ==================== | # Observable retail signals only. This is NOT sales: Amazon does not publish | # unit sales. Rank is inverted (lower rank number = stronger position). | med_bsr = median_of('amz_bsr'); med_cat = median_of('amz_cat_rank') | a1 = minmax([lg(a['amz_reviews']) for a in C]) | a2 = minmax([-lg(a['amz_bsr'] if a['amz_bsr'] is not None else med_bsr) for a in C]) | a3 = minmax([-lg(a['amz_cat_rank'] if a['amz_cat_rank'] is not None else med_cat) for a in C]) | for i, a in enumerate(C): | a['c_commercial'] = 0.40 * a1[i] + 0.40 * a2[i] + 0.20 * a3[i] | | # =================== 5. STORYGRAPH READER SIGNAL (10) ====================== | med_sg = median_of('sg_reviews') | sgv = minmax([lg(a['sg_reviews'] if a['sg_reviews'] is not None else med_sg) for a in C]) | for i, a in enumerate(C): a['c_storygraph'] = sgv[i] | | # ============ 6. COMMUNITY & EDITORIAL COVERAGE (8) ======================== | # Distinct independent editorial outlets verified, plus a juried-award flag. | # Lesbian Romance Reads is EXCLUDED from every count (no self-reference). | # Reddit, BookTok and Bookstagram are NOT included: see the weight note above. | co = minmax([lg(a['outlets']) for a in C]) | for i, a in enumerate(C): | a['c_community'] = 0.70 * co[i] + 0.30 * (100.0 if a['award'] else 0.0) | | # ======================= 7. CURRENT MOMENTUM (10) ========================== | # Velocity = Goodreads ratings on the most recent qualifying release at least | # 14 days old, divided by months since publication. Free Amazon Original | # Stories are excluded from velocity as a distribution artefact, not demand. | for a in C: | if a['recent_ratings'] is None or a['recent_months'] in (None, 0): | a['_vel'] = None | else: | a['_vel'] = a['recent_ratings'] / a['recent_months'] | med_vel = st.median([a['_vel'] for a in C if a['_vel'] is not None]) | m1 = minmax([lg(a['rel_18m']) for a in C]) | m2 = minmax([lg(a['_vel'] if a['_vel'] is not None else med_vel) for a in C]) | for i, a in enumerate(C): | a['c_momentum'] = 0.30 * m1[i] + 0.50 * m2[i] + 0.20 * (100.0 if a['upcoming'] else 0.0) | v = a['_vel'] if a['_vel'] is not None else med_vel | a['momentum_dir'] = 'down' if a['rel_18m'] == 0 else ('up' if v >= med_vel else 'flat') | | # ============================== FINAL SCORE ================================ | for a in C: | a['score'] = round( | W['reach'] * a['c_reach'] / 100 + | W['satisfaction']* a['c_satisfaction'] / 100 + | W['catalogue'] * a['c_catalogue'] / 100 + | W['commercial'] * a['c_commercial'] / 100 + | W['storygraph'] * a['c_storygraph'] / 100 + | W['community'] * a['c_community'] / 100 + | W['momentum'] * a['c_momentum'] / 100, 1) | a['momentum_score'] = round(a['c_momentum']) | comps = {'Goodreads reach': a['c_reach'], 'Reader satisfaction': a['c_satisfaction'], | 'Catalogue strength': a['c_catalogue'], 'Commercial visibility': a['c_commercial'], | 'StoryGraph signal': a['c_storygraph'], 'Community coverage': a['c_community'], | 'Momentum': a['c_momentum']} | a['strongest'] = max(comps, key=comps.get) | | C.sort(key=lambda x: -x['score']) | for i, a in enumerate(C): a['rank'] = i + 1 | | # ================================ OUTPUT =================================== | print(f"Weights sum: {sum(W.values())} | prior m = {m_prior:,.0f} | prior C = {C_prior:.4f}") | print(f"median velocity = {med_vel:.1f} ratings/month") | for a in C: | print(f"{a['rank']:>2} {a['name']:<24}{a['score']:>6.1f}{a['momentum_score']:>5}" | f"{a['momentum_dir']:>6}{a['confidence']:>5}% {a['strongest']}") | | keys = ['rank','name','score','momentum_score','momentum_dir','confidence','strongest', | 'c_reach','c_satisfaction','c_catalogue','c_commercial','c_storygraph','c_community','c_momentum', | 'gr_total','gr_over_1k','gr_over_5k','sg_reviews','amz_bsr','amz_cat_rank','amz_reviews', | 'novels','first_year','rel_18m','outlets','award','ku','_v','_R','_adj','_vel'] | import csv | with open('eikna_v1_scores.csv','w',newline='') as f: | w = csv.writer(f); w.writerow(keys) | for a in C: w.writerow([round(a[k],4) if isinstance(a.get(k),float) else a.get(k) for k in keys]) | with open('eikna_v1_scores.json','w') as f: | json.dump({'index':'EIKNA Lesbian Romance Author Index', | 'version':'v1.0','edition':'August 2026','data_checked':'2026-08-16', | 'weights':W,'prior_m':m_prior,'prior_C':round(C_prior,4), | 'median_velocity':round(med_vel,2), | 'eligibility':{'min_titles':E_MIN_TITLES,'min_gr_ratings':E_MIN_RATINGS, | 'max_first_year':E_MAX_FIRST_YEAR,'min_confidence':E_MIN_CONF}, | 'ranked':[{k:a.get(k) for k in keys} for a in C], | 'excluded':[{'name':a['name'],'reason':a['ineligible_because']} for a in AUTHORS if not a['eligible']], | 'emerging':EMERGING}, f, indent=1) | print("Wrote eikna_v1_scores.csv and eikna_v1_scores.json") | | # ===================== EXPECTED OUTPUT (16 August 2026) ==================== | # Eligible cohort: 25 of 25 | # prior m = 20,478 | prior C = 3.9773 | median velocity = 186.0 ratings/month | # 1 Haley Cass 70.4 | 2 Ashley Herring Blake 65.8 | 3 Lee Winter 60.7 | # 4 Jae 60.7 | 5 Radclyffe 60.7 | 6 Melissa Brayden 57.0 | 7 S.T. Gibson 57.0 | # 8 Alexandria Bellefleur 55.3 | 9 Rachel Lacey 53.5 | 10 Georgia Beers 53.1 | # 11 Harper Bliss 52.8 | 12 Monica McCallan 52.4 | 13 Gerri Hill 51.4 | # 14 Milena McKay 51.3 | 15 Ruby Roe 51.0 | 16 Meryl Wilsner 48.2 | # 17 E.J. Noyes 48.0 | 18 Roslyn Sinclair 45.5 | 19 Clare Lydon 43.3 | # 20 Eliza Lentzski 43.3 | 21 Melissa Tereze 42.5 | 22 Nicole Pyland 39.1 | # 23 Anna Burke 33.4 | 24 Chelsea M. Cameron 30.6 | 25 Fiona Riley 17.8 | # | # Jossef S. is NOT in this cohort: fails E1 (1 qualifying title), E2 (30 | # Goodreads ratings) and E3 (first published 2026). Applied identically to | # every author. Rankings cannot be purchased. |