# T773 build research Checked primary definitions: PARI ellheight(E,P,Q) uses the Cremona convention and (H(P+Q)-H(P-Q))/4; Sage ell_point.height documents the larger BSD height, twice Silverman's convention. Over Q the field-degree normalization is one. Sage's installed height source dispatches algorithm='sage' to the sum of its non_archimedean_local_height and archimedean_local_height. Its point cache ignores the algorithm, so checks.py calls the local routines directly. The modular runtime needs sage.rings.qqbar._init_qqbar() before local heights. The unsuccessful preliminary probe is preserved in probe.log; probe2.log records the resolved initialization and uncached local-route timings. Checked exact inputs: primary allgens/allgens.00000-09999 and alllabels/alllabels.00000-09999 from ecdata commit 25cec5ecfec8b9f016eb1631ac633194c2bed39f. Sources and hashes are in source-manifest.json. Cremona's TABLE TWO explicitly orders free generators before torsion and demonstrates ordinary projective coordinates. Source-derived 158 records (60 rank one, 97 rank two, one rank three) and all signed affine points match live T770 metadata. This is input reconciliation only; T770's numbers and generator were not used. Full input records are points.json. Live discovery: catalogue.json records one eight-page planning fetch; candidate metadata, screen.json, lookup-*.json and tag-*.json are retained. The classical pairing family is absent. The rank-one 37a1 probe finds T64, while the 389a1 and negative 5077a1 example probes find no numeric match. NumberDB T64, T65 and T66 are full-lattice regulators; T347 is elliptic divisibility sequences (a point coordinate-growth companion), not heights; T211 is the Faltings height of a curve. T770 is offered_for_review, private, and distinct p-adic data; omit its public relation link while this status persists. Every inspected origin is proposal #354 contributing only to wanted #266. No additional direct origins or requested tags were found in #266. Verified live tags: elliptic curves and number theory. Exact height and arithmetic geometry tag probes are absent. The finite source-rank slices and their noncanonical budgets are justified in argument-selection.md. No free continuous parameter or arbitrary grid is present. No assumption of saturation, full-lattice index or BSD enters a pairing. Coarse independence certificates are separately derived in coarse-bound.md; fine values use two working precisions and both height routes. Completed numerical evidence covers all 357 entries at 681/1013 bits. The maximum observed absolute disagreement is below 5e-205. All 158 coarse Gram matrices have positive certified leading principal minors after six doublings. Sage 10.9 has an additional global finite-height wrapper bug at x=0 when a discriminant exponent is at least 12: c=1 is a Python int without valuation. The final checker uses explicit bad-prime local calls plus log(denominator(x)) on the checked globally minimal model. The first 125 curves completed before this bug on 1664n1; 33 resumed from their saved checkpoint. Both exact wrapper paths and all failed probes are preserved. Final stored checks use the explicit prime-place sum for all entries. Source methods and versions are attached. No independent critic verdict is issued in this build. Canonical skill: ../20261120T062421Z-build-skills/manifest.json, validated local/live bundle b5c3c7985b7c9e178b4899d083a9e3bc4284989dff76f05e5b0e52835d47eed2. An exact manifest copy is skill-manifest.json. Core and the six task-relevant references were read from one archived local copy; all resource hashes match both valid archive manifests. No competing policy version was adopted.