Implementation notes — 2026-07-22 frontier round 5 (Goodspeed/Wooding extension)

Scope: apply ONLY the approved change-set from the dispatch (Goodspeed/Wooding line extension). Source: 2026-07-22-frontier-round5-partG.md + partH.md, tree doc 2026-07-20-wilson-tree-v0.md.

What changed

index.html (ANC object)

• 353 Elizabeth Goodspead: b:null,eb:1733 → b:1731; ed:1795 kept as-is (not recomputed off the new real birth — see judgment call #1 below); added prob:true; rewrote ebasis, story, conf.
• Added 706 Stephen Goodspeed — b:1706,d:1763,prob:true.
• Added 707 Bethiah Wooding — b:1711,ed:1773,prob:true.
• Added 1412 Nathaniel Goodspeed — eb:1680 (no prob, CONFIRMED-existence tier).
• Added 1413 Sarah King — eb:1684 (no prob, CONFIRMED-existence tier).
• Added 1414 Peter Wooding — eb:1671,d:1717 (no prob, CONFIRMED tier, real death date).
• Added 1415 Elizabeth Mallot — eb:1675,ed:1737 (no prob, CONFIRMED tier).
• New block inserted after the existing 11277 entry, before the closing };, with a dated comment header. Insertion position doesn't affect ordering — the file's own comment convention says integer keys auto-sort.

build-people.mjs

• Added 706, 707, 1412, 1413, 1414, 1415 to LINE_BY_AHN, all mapped to 'Olney & Lambert'. This wasn't explicitly in the change-set text but is a mechanical necessity: without it, build-people.mjs would emit an "unassigned line" console warning for all six and they'd be missing from the grouped people/index.html listing (their individual pages would still be written regardless — the write loop iterates all persons unconditionally — but they'd be orphaned from the family-line index). Judgment call #2, logged below.

people-bios.json

• Rewrote elizabeth-goodspead (existing entry) to reflect the new real birth + newly named parents.
• Added 6 new entries: stephen-goodspeed, bethiah-wooding, nathaniel-goodspeed, sarah-king, peter-wooding, elizabeth-mallot. Matched the existing schema (bio: [paragraphs], facts: [{text, source_label, source_url, confidence}]) and the file's own typographic conventions (curly ’ apostrophes, curly “ ” quotes for quoted primary-source language, — for em dashes) rather than straight ASCII punctuation.
• Entry count: 110 → 116 (105 pre-existing untouched + 1 rewritten + 6 new — net +6 as expected).

Tree doc (2026-07-20-wilson-tree-v0.md)

• Appended ## 2026-07-22 frontier round 5 section: headlines, the 6-ancestor Goodspeed/Wooding extension, the 3 reconciled contradictions (Stephen Goodspeed d.1763 ruling, GW Gibby stays 1909/no site change, Alice Pemberton baptism year unresolved), the Oakford doc-only date pins, the Pemberton-upstream doc-only/quarantined notes, and top open keys carried forward.
• Did NOT touch the existing 2026-07-21 "Estimation methodology" table's 353 Elizabeth Goodspead · ~1733–~1795 row — that table is a dated historical snapshot of what was estimated that day; editing it retroactively would misrepresent the record. The new round-5 section documents the retirement of that estimate instead. Flagging this as a possible follow-up if the founder wants that table kept current rather than historical.

Judgment calls (things a fresh reader should know)

1. Elizabeth Goodspead's death estimate (ed:1795) was kept as-is, not recomputed from the new real birth. Strict E3 arithmetic off the new real birth (1731+62) would give 1793, not 1795. The dispatch explicitly said "keep the estimated death (ed:1795)," and the methodology's E4 rule ("a real documented date always overrides its estimated half") doesn't mandate recomputing the other half just because one side went real — so this isn't a methodology violation, just an instruction I followed literally rather than silently "improving." Flagging in case the founder wants ed recomputed to 1793 for arithmetic consistency; leaving it at 1795 per the explicit instruction.
2. LINE_BY_AHN map extension in build-people.mjs (see above) — a plumbing change, not a content/genealogy judgment call, needed so the six new pages surface correctly in the grouped people index rather than silently orphaning.
3. Node numbering required no judgment at all — 706/707/1412/1413/1414/1415 are the exact ahnentafel positions relative to Elizabeth Goodspead (353): father=2n, mother=2n+1, recursively. Verified programmatically (floor(706/2)==353, floor(1412/2)==706, etc. — all true) before writing.
4. Held to the HOLD list exactly: did not add Oakford, Pemberton-upstream, or any Gibby-node changes to index.html or people-bios.json. Those are recorded doc-only in the tree doc's new section, per instruction.
5. Did not add SOURCES or FRONTIER_HINTS entries in build-people.mjs for the 6 new slugs — out of the approved scope; the generator's existing fallback behavior (a muted "no direct web links captured yet" note under Sources, and a generic "not yet found — research frontier" stub for their own unlisted parents) handles their absence gracefully and correctly.

Verification results

• ANC/GROUPS/REGIONS block parse: extracted all three via the same balanced-delimiter new Function approach build-people.mjs uses, in a standalone Node check. All three parse cleanly. ANC has 124 keys (118 + 6 new). Checked every b/d/eb/ed field across all 124 entries is either null or a real (non-NaN) number — zero bad fields. Checked every non-unknown entry's region resolves in REGIONS — zero mismatches.
• Ahnentafel consistency: confirmed programmatically that 706/707 are children of 353, and 1412/1413/1414/1415 are children of 706/707 respectively (all floor(n/2) checks true).
• node build-people.mjs exit status: 0. Output: "people/: wrote 122 person pages + index.html + slugs.js + style.css" / "persons total: 122" (116 prior + 6 new) / "slug-map keys: 250" / pre-existing slug collisions unchanged (william-olney ×3, john-herman-wilson ×2) — no new collisions introduced. No "WARN: unassigned line" printed, confirming the LINE_BY_AHN fix worked.
• Generated pages spot-checked: all 6 new person pages + the updated elizabeth-goodspead.html render with correct year ranges, correct badge (probable vs plain, matching the prob flag choices above), correct bio text pulled from people-bios.json, and fully-linked Immediate family sections (father/mother/spouse/children all resolve to the right pages — e.g. Stephen Goodspeed's page correctly links up to Nathaniel Goodspeed + Sarah King and down to Elizabeth Goodspead).
• grep for literal null/NaN/undefined across all files in people/*.html: zero matches (excluding the JSON source's own "source_url": null keys, which aren't in the generated HTML — confirmed separately that no such matches appear in HTML output at all).
• people/index.html grouping: confirmed all 6 new people appear under the "Olney & Lambert" <h2> section, correctly year-sorted alongside existing members.
• Did not deploy — left for the founder per instruction.

2026-07-22 ~07:10 ET — Timeline width consolidation (founder ask, separate from round 5)

Founder (iMessage + screenshot): the timeline gave every person their own vertical lane → very wide with dead whitespace. Ask: pack non-overlapping lifespans into shared columns; child-between-parents rule can be relaxed; hard constraint = never stack two people alive at the same time. Example he gave: Danny Wilson + Jeremiah Lambert (non-overlapping) can share a column.

Implementation (index.html renderTimeline):

• Sort FLAT by birth year (was: by _f genealogical fraction), tiebreak _f → same-year births still order paternal→maternal.
• Greedy interval-partitioning: laneBottom[] tracks each column's last death-pixel; each person takes the leftmost column whose last occupant died before this birth (+LANE_BUF=12px), else opens a new column. This is optimal (min columns) for interval graphs when processed in start order.
• x = GUT+6 + lane*(BW+GAP) (was per-person x+=BW+GAP).
• Canvas min-width set dynamically to GUT+6+laneCount*(BW+GAP)+30 (overrides the CSS #canvas{min-width:3010px} floor which was sized for one-col-per-person).
• Hint caption rewritten (was "columns run genealogically, child between parents" — now describes packing; points to Fan view for exact structure).
• Descent brackets + marriage connectors unchanged (still keyed off per-person x via colX/COLK/_x) — they now span packed columns; no visible clutter in render.

Verified via Playwright (localhost:8743, cache-busted): 118 bars → 56 columns (was 118); canvas 3010px → 1538px (49% narrower); 0 same-column time-overlaps (bug check passed); 0 console errors. 56 is the true floor = peak simultaneously-alive count, which the founder's constraint forbids collapsing further. Full-page screenshot eyeballed: clean staircase, chips readable, region colors intact.