Updates the public benchmark leaderboard and the homepage to reflect
the 2026-05-03 re-run of clawbio_bench v0.1.5, after the nutrigx
compat-symlink fix in PR ClawBio#215.

Headline numbers
- Audit baseline (2026-04-05, bench v0.1.0): 80 / 140 (57.1%)
- Latest run (2026-05-03, bench v0.1.5): 149 / 162 (92.0%)
- Skills audited: 7 to 10 (3 new harnesses in v0.1.5)

Per-skill changes
- equity-scorer: 20.0% to 100.0% (P0 findings resolved)
- nutrigx-advisor: 80.0% to 100.0% (today's symlink fix)
- pharmgx-reporter: 42.4% to 97.7%
- bio-orchestrator: 75.9% to 98.1%
- claw-metagenomics: 85.7% to 100.0%
- clinical-variant-reporter: 80.0% to 80.0% (unchanged)
- fine-mapping: now reports 21 harness infrastructure errors,
  excluded from rate (under investigation)
- New harnesses: cvr-acmg-correctness 69.2%, gwas-prs 62.5%,
  cvr-variant-identity 50.0%

Site changes
- benchmarks.html
  - Hero adds "We fix them" arc
  - Audit metadata card "Auditor: Sergey Kornilov" replaced with
    "Bench Author: Biostochastics LLC", focusing the credit on the
    open-source clawbio_bench tool rather than an individual
  - Bench version v0.1.0 to v0.1.5
  - Audit commit 1481fb4 to 925b89a
  - Summary bar: 80/140 to 149/162
  - Scorecard table: full rebuild against new run, 10 rows, status
    pills updated (most P0/P1 now Clear)
  - Footer: original audit baseline retained for transparency
  - JSON-LD dateModified bumped
- index.html
  - Top banner: "80/140 across 7 skills" to "149/162 (92.0%) | Up
    from 80/140 (57.1%) at original audit"
  - Hero CTA pill: "Leaderboard 80/140" to "Leaderboard 149/162"
  - Benchmark Validated feature card: numbers refreshed
  - Skills section header: "7 publicly benchmarked" to "149/162
    publicly benchmarked passing"
  - Recent Updates: leaderboard card now references PR ClawBio#215 too,
    cites clawbio_bench, includes the gain numbers

The auditor-name swap follows the principle that the bench is a
public tool with org credit; a single individual's name is not the
right framing for a permanent front-page asset.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Resolves the 21 harness errors observed in clawbio_bench v0.1.5
finemapping driver. The driver does:

    sys.path.insert(0, skill_dir)
    from core.abf import compute_abf
    from core.susie import run_susie
    from core.credible_sets import ...
    from core.susie_inf import run_susie_inf  # optional

The skill restructured its internal package from `core` to
`fine_mapping_core` at some point but the bench harness still
expects `core`. Adds a tracked directory symlink so the legacy
import path resolves until the bench is updated.

Effect on benchmark
- clawbio-finemapping: 0/0 with 21 harness errors -> 19/20 (95.0%)
  with 1 real algorithm failure (susie_inf_est_tausq_ignored)
- Aggregate: 149/162 (92.0%) excluding finemapping -> 168/182
  (92.3%) including finemapping

Same pattern as the nutrigx_advisor symlink in PR ClawBio#215. The proper
fix is in biostochastics/clawbio_bench: either resolve skill-package
names dynamically or read import paths from a per-skill manifest.
Once that lands, both symlinks can be removed.

Verified
- clawbio-bench --smoke --harness finemapping: 19/20 pass

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>