# T760 numerical evidence Quantity: ordinary logarithmic Mahler measure of a+bx+cy, natural logarithm and unit Haar measure on both circles. Indices are primitive sorted positive integer strict triangles; angles are paired with their opposite sides. The shape has positive imaginary part and its positive algebraic square root is enclosed. No principal value, multiplicative measure, per-vertex division or sign-reversed free energy is substituted. Scope: all 84 selected identities, exactly those in selection.json and argument-selection.md. Exact integer arithmetic computes the discriminant; rational divisions use QQ. Every returned native number is sage.rings.real_arb.RealBall. The required dry_run.py/check.py report no exactness, import, prose or suspicious-decimal complaints: 84 entries, longest 156 characters and measured block 18.9 KB. Production uses Sage 10.9, numberdb 0.1.10, Python 3.12.3, at 461 bits, including 128 guard bits. There is no manually truncated analytic series: Arb evaluates the closed Cassaigne–Maillot expression with exact inputs enclosed before transcendental operations. The least measured relative accuracy is 457.0387871084 bits. The largest reported radius is 8.6687683e-138. Diagnostics are rounded measurements, not replacement certificates. All 84 balls are finite and nonvacuous. Exact rational comparisons against the stored decimal endpoints establish containment of the complete production ball. The stored decimals have 100 significant digits, each with the NumberDB interval of one unit in its last place. Independent numerical route: checks.py derives a single-circle smooth finite integral by Jensen integration in y. It never imports or evaluates the production formula for its 84 row comparisons. mpmath 1.4.1 quadrature at 145 and 180 decimal digits is a separate backend and formula. The largest difference between its two precisions is 5.9956859e-146; the largest difference between its 180-digit result and a stored center is 0.498463120867 last-place units. Endpoint location is recomputed at both working precisions, not pasted from a decimal. These numerical differences are heuristic agreement evidence, not a certified quadrature error or a universal identity proof. Production rigor rests on the closed-form theorem, branch/domain check and propagated Arb balls. Controls actually executed: polynomial integral 1/3 and the nonzero Smyth Dirichlet-L anchor; boundary (1,1,2), nontriangle (1,1,3), coefficient swap and common scaling. The checker rejects a 1e-50 perturbation of the anchor, an omitted log(2) scale term, a conjugated-shape sign, NaN, infinity and a vacuously wide ball. enclosure_checks.py also checks permutation and doubling for every production ball; that supplementary test shares the production formula. The independent integer enumeration uses defining equations and confirms 68 isosceles and 16 disjoint primitive right triples, ordered by (c,b,a), against the generator's repeated-side and Euclid parametrizations. The displayed mpmath Program executes and reproduces the same selection and its 5-5-9 example. Stored-data check: check-results.json names the final stored.json hash 9567a2b716a5511e7ad846a2b0da7e758735205aba757230467b72023bb7fcca; checked_entries=84, total_entries=84, failures=[], and both checker controls are true. verify(sample=None) matched 84/84, with no differing, missing or extra entries. verify checks generator reproduction; Jensen is the independent mathematical computation. No independent scientific verdict is issued here. Skill manifest: ../20261118T202056Z-build-skills/manifest.json, archived SHA-256 7629363ce07d2e1340528294fe639e6fc7a5b18da808ec52558687fc1db91534. Local/live complete bundle hashes both equal b5c3c7985b7c9e178b4899d083a9e3bc4284989dff76f05e5b0e52835d47eed2. Each archived core/reference byte hash was verified before use. Sources and precise passages are in research.md and sources-manifest.json. Generator SHA-256: c14d65cc5911c93166b01abf131be22e4f73695bce22ef908032c64e4a4b2217. Computed checkpoint SHA-256: 6bbb11e405ec4c99cc92929bfca80d59bf285ac56ea4d1dc2ca15f4e57a5fbdd. Executable independent checker SHA-256: 3e64aea4345ef2f16ada0ec60464a92439533508c11394f5a8e4eee98582645f. Exact containment/identity checker SHA-256: 928519d013710fe95855a604c19d536bccd50a513cfa8f9937b09f50b65a22b7. Actual complete result hashes are recorded in evidence-manifest.json. Presentation: the authenticated API audit is clean. All seven Similar tables records render, and all 84 identity blocks were inspected in a complete local assembly of the site's exact server-rendered preview sections. The public preview rejected a full-document GET with HTTP 414. Individual exact stored sections and row batches were rendered by the same website, combined without rewriting mathematics, and inspected using shared Playwright 1.64.1 / Chromium 156.0.8078.4. This does not inspect the private draft's authenticated page chrome. The rendering manifest, HTML, screenshots and browser-results.json preserve that limitation. Companion-link checks verify all seven destinations and four entry anchors. History retains exact #345/#263 citations. Reproduction in this campaign uses agents/sage.sh with generate.py, checks.py, enclosure_checks.py and stored.json as mounted inputs. final_checks.py runs the displayed Program, all exact enclosure checks and all Jensen comparisons; final-checks.log retains executed output. The attached generator and checks also run outside the campaign with SageMath + numberdb and Python + mpmath, respectively. Credentials belong only in the contributor's configured NumberDB client, never in attachments. Rendering environment: final Chromium inspection reuses FONTCONFIG_FILE=/home/ubuntu/.cache/numberdb/browser-fonts/fonts.conf. The host has no default system fontconfig; earlier screenshots without that setting are diagnostic failures and do not serve as final presentation evidence. The final browser-results.json requires all 84 number rows to have visible text. Source audit repair: after files were attached, the API audit requested an explicit installation command in generate.py. The final docstring now includes `sage -pip install numberdb`. Executable numerical code is byte-identical after its import boundary. The exact originally computed source is preserved as production-source-original.txt (SHA-256 c14d65cc5911c93166b01abf131be22e4f73695bce22ef908032c64e4a4b2217). The final generator SHA-256 is 2d862b983e95284debf8a210ff9aac5ce157bc65cd833ee8eed2fa2bab27e259; verify-final.log recomputes all 84 stored values with that final source. The preexisting computation version/hash record is retained rather than rewritten. A fresh controller critique will use the final source fingerprint.