fix: correct Human Design gate wheel offset (-1.375 -> -1.75) - #14

Root cause

scripts/chart_engine.py:98 — OFFSET=-1.375 in gate_line(). This shifts the entire 64-gate Human Design wheel by 0.375 deg away from its real position relative to the ecliptic. SEQ (the gate sequence) and the line-width math in gate_line() are correct; only the rotational offset constant is wrong.

def gate_line(lon):
    w=(lon-OFFSET)%360; idx=int(w//5.625); g=SEQ[idx]
    line=int((w-idx*5.625)//0.9375)+1
    return g,line

Sources

• Published Human Design mandala (barneyandflow.com/gate-zodiac-degrees; Scribd "Gates and Zodiac Placements"):
  • Gate 25 = 28°15′00″ Pisces (358.25°) – 03°52′30″ Aries
  • Gate 17 = 03°52′30″ (3.875°) – 09°30′ Aries
  • Gate 41 starts 02°00′ Aquarius (302.0°)
• OFFSET=-1.375 (current main) puts all three anchors in the wrong gate. OFFSET=-1.75 puts all three in the documented gate.
• Independent confirmation: npm hd-chart-engine@0.1.1 (used by life-web's question-first path, life-web/node_modules/hd-chart-engine/dist/index.js) has WHEEL_START=223.25, i.e. -1.75 + 40*5.625 — the same offset, derived independently.

Impact

At OFFSET=-1.375: ~6.5% of gate activations land in the wrong gate, ~39.7% land on the wrong gate-or-line, and ~83% of charts have at least one wrong gate. Every consumer of this engine (server.py, webapp.py, mcp_server.py, synastry.py, the WordPress plugin over HTTP) inherits the wrong values, since they all go through gate_line().

Fix

One line: OFFSET=-1.375 → OFFSET=-1.75.

TDD

tests/test_human_design_gate_wheel_offset.py (new) pins the three published mandala anchors plus a preceding-gate/previous-line case and a 0°-wrap-around case:

tests/test_human_design_gate_wheel_offset.py::test_gate_25_starts_at_358_25_degrees            RED -> GREEN
tests/test_human_design_gate_wheel_offset.py::test_last_line_before_gate_25_is_the_preceding_wheel_gate   GREEN (unaffected — relative to SEQ)
tests/test_human_design_gate_wheel_offset.py::test_gate_17_starts_at_3_875_degrees             RED -> GREEN
tests/test_human_design_gate_wheel_offset.py::test_gate_41_starts_at_302_degrees               RED -> GREEN
tests/test_human_design_gate_wheel_offset.py::test_just_below_302_degrees_is_previous_gate_line_6         GREEN (unaffected — relative to SEQ)
tests/test_human_design_gate_wheel_offset.py::test_wheel_wraps_around_0_degrees_consistently   GREEN (unaffected — relative to SEQ)

RED (at OFFSET=-1.375, captured before the fix):

AssertionError: assert (36, 6) == (25, 1)   # gate_line(358.25)
AssertionError: assert (25, 6) == (17, 1)   # gate_line(3.875)
AssertionError: assert (60, 6) == (41, 1)   # gate_line(302.0)
3 failed, 3 passed

GREEN (at OFFSET=-1.75): 6 passed.

Fixtures regenerated from the corrected engine (no hand edits)

tests/fixtures/update_hd_baseline.py (new) regenerates only build_json.human_design and the human_design_boundary annotations in ephemeris_baseline.json, asserting western stays within a sanity tolerance first so any unrelated drift would fail loudly instead of being silently baked in. golden_example.json/.md and examples/sample-output.json are regenerated with their existing GOLDEN_PROVENANCE.md commands. See that file's new 2026-09-15 entry for the full writeup, including why golden_example_pre_1_2.json is regenerated too (it freezes "1.1→1.2 adds one field", not a known-wrong gate wheel — leaving it stale would permanently pin the bug through its byte-identity test).

Per-case before/after (tests/fixtures/ephemeris_baseline.json, via update_hd_baseline.py)

11 of 12 fixtures changed (the 12th, placidus_lat70_invalid, has ok:false — no human_design to regenerate).

golden_example.json/.md (--example sample): channel 3-60 replaced by 27-50; 9 gate/line values change.
examples/sample-output.json (小明 sample): profile 1/3→2/4; 10 gate/line values change (matches taipei_standard above — same birth input).
README.md + all 18 translations: the embedded 小明 sample snippet updated to match (角色 1/3→2/4, 12.1/36.3→12.2/36.4).

tests/test_synastry_hd.py::test_time_unknown_channel_confidence_exhaustive_split_pinned (not in the original sweep) also pinned a gate set derived from the old offset — person B0's noon fast-planet gates shift {...,60} → {...,41}, dropping the hd-chan-3-60-dominance connection. Updated to match the corrected engine's actual output (verified by direct invocation, not guessed).

Test commands

cd life-chart-engine
uv venv --python 3.12 .venv   # or: bash setup.sh
uv pip install --python .venv/bin/python -r requirements.txt -r requirements-dev.txt
.venv/bin/python -m pytest -q
.venv/bin/python -m pytest tests/test_human_design_gate_wheel_offset.py -v
.venv/bin/python -m pytest -m deploy_regression -q -r s
.venv/bin/python -m ruff check .

Verification run (this branch, HEAD)

447 passed, 2 skipped (1 pre-existing mcp extra skip, 1 other pre-existing skip; baseline on main before this change was 441 passed, 2 skipped — the +6 are the new anchor tests). ruff check . — all checks passed.

Not merged, not deployed

Per instructions this PR is opened for review only. pyproject.toml version bumped 1.2.0→1.2.1 (patch); no CHANGELOG file exists in this repo so none was added. schema_version stays "1.2" (JSON contract shape unchanged).

Also fixed (same bug, separate copy, not part of this repo)

$HOME/Documents/CC Cli/agents-skills/life/scripts/chart_engine.py is an older, separate fork (Swiss Ephemeris/py-iztro, not a copy of this engine) with the same OFFSET=-1.375 at its own line 72. Edited to -1.75 in the working tree; not committed here (different repo, out of scope for this PR) — see the session notes for that repo's status.

Out of scope, flagged not fixed

• GATE_CENTER[19] in the CC Cli skill fork (agents-skills/life/scripts/chart_engine.py:78) is still '情緒'; this engine repo already fixed the equivalent bug on 2026-08-30 (docs/2026-08-30-gate19-root-fix-impact.md), but the older fork never got that fix. Separate issue, separate repo.
• placidus_lat70_invalid's frozen ephemeris_baseline.json error string ("swisseph.houses_ex: error") predates the AGPL→MIT/astronomy-engine migration; the current engine raises a different message for the same input ("placidus undefined at high latitude"). Unrelated to the OFFSET bug — the ok:false flag itself is unchanged and no test asserts the error text — left untouched.

https://claude.ai/code/session_016ZLLbCYxw9wxqX6rsSrbMX