Wiki · Backlog
Refresh Rift Forge research from the candidate branch
Shelf
Backlog
Source
docs/backlog/archive/WI-030-refresh-rift-forge-research.md
Status
done

Proposal (rationale)

The existing Rift Forge extraction was surveyed on 2026-08-14 against a detached working tree and records 3,236 commits, 401 branches, 51 live worktrees, 13 skills, 69 scripts, and 102 live work items. The local candidate/0.1.0 branch has since advanced materially and is the branch the source repo calls its active candidate. Those claims are now stale unless re-derived from that ref.

This item refreshes the extraction against the candidate branch, preserving the original research boundary while making the evidence reproducible at a full commit SHA. It is a research correction, not a module or product change.

Decision

accepted — 2026-08-15. The local candidate/0.1.0 branch is the requested authority; its full SHA and divergence from the stale remote-tracking ref will be recorded before any conclusion is drawn.

Plan

Requirements

  • Pin the local candidate/0.1.0 ref to a full commit SHA, record its read date, branch/ref, commit count, and the candidate-vs-remote-ref relationship.
  • Re-read the candidate’s own instructions, backlog/session/worktree workflow, skills, gates, release posture, and generated-artifact rules before updating the extraction.
  • Recompute every headline inventory claim retained in the refreshed extraction with a named command and date; distinguish current candidate facts from historical observations.
  • Record changed, retired, and newly introduced practices with file paths and commit evidence, while marking interpretation as Opinion in the first person.
  • Check direct dependent research claims for the old Rift Forge counts and update or explicitly mark each one stale; do not silently leave contradicted numbers in the synthesis/index.
  • State the candidate checkout’s licence evidence explicitly; do not infer a licence from metadata.

Impacts

  • Primary artifact: docs/research/repos/rift-forge.md.
  • Possible direct dependent updates under docs/research/ where the old Rift Forge counts or branch posture are repeated; no changes to modules/, CLI source, or the Rift Forge checkout.
  • Site content and link-check surface change because research documents are published as wiki pages.
  • Research claims may change pattern confidence in a later synthesis, but this item does not edit the pattern catalogue or shipped modules.

Approach

Use a temporary detached worktree at the local candidate/0.1.0 ref so the source checkout and its existing worktrees are not switched or mutated. Run the candidate’s own inventory and self-test commands where dependencies permit, supplementing with read-only Git/tree measurements. Compare the new evidence against the 2026-08-14 extraction, then update only claims that the candidate checkout supports. If the local candidate and origin/candidate/0.1.0 diverge, treat the requested local candidate ref as the authority and record the divergence rather than blending the trees.

Acceptance criteria / tests

  1. docs/research/repos/rift-forge.md identifies candidate/0.1.0, a full SHA, read date, source checkout, and licence evidence, and every retained headline count has a reproducible command/date.
  2. The refreshed document covers the current candidate’s setup, working practices, failure/retirement evidence, pain-point table, and extraction verdict; changed or retired practices are not hidden by copying the old narrative forward.
  3. A repository-wide search finds no unqualified stale Rift Forge headline count in direct research dependents; each changed claim is updated or labelled historical/stale with a reason.
  4. The source checkout remains unchanged and no modules/ or CLI implementation files change.
  5. rungs check, site build, and site link/type checks pass with zero broken links; the refresh’s evidence and limitations are recorded in the work item review.

Out of scope

  • Editing or merging the Rift Forge candidate branch, its working tree, or its remote.
  • Re-running the six public-framework extractions or changing pattern-catalog.md/modules/; open a follow-up item if refreshed evidence changes a catalogue decision.
  • Reconstructing every historical branch or every product feature; only claims used by the extraction and its direct dependents are in scope.

Execution

completed — 2026-08-15. Candidate anchor established before editing:

  • Local candidate/0.1.0 is 4a51848cfc9a2acbcdeddcd028418572406e2950, 3,585 commits, dated 2026-08-15, with tip Merge WI-790: route coverage rows to live owners.
  • origin/candidate/0.1.0 is the older 159f9f030c32df6828c16b3637ae572513d34d4e (3,262 commits). The requested local candidate is authoritative; the remote-tracking divergence is recorded rather than blended.
  • The existing source checkout is detached at 472d45ed; research will use a temporary worktree at the candidate ref and will not switch or mutate that checkout.

Evidence gathered on 2026-08-15 from the temporary candidate worktree:

  • Inventory: 3,585 commits (git rev-list --count), first commit 21b71fa2 on 2026-07-29, candidate tip 4a51848c on 2026-08-15; 433 branches; 105 registered worktrees. The candidate’s worktrees.mjs report measured 80 prunable, 16 merged-but-dirty, and 13 dirty for 10+ days (oldest 12 days).
  • Repository surfaces: 14 skills; 85 .github/scripts/*.mjs; 181 package scripts (58 check:, 46 test:, 30 report:, 7 gen:); 16 specs; 25 ADRs; 163 live + 537 archived work items; 3 active + 2 archived sprints; CLAUDE.md 555 lines; PROJECT-STATE.md 1,223 lines; 74 open finding rows and 200 archived finding sections.
  • Candidate-vs-checkout delta: 66 commits after 472d45ed, including WI-829’s history split, F-323’s shared status predicate and land preflight, F-360’s self-declared-finding closure gate, WI-790’s live triage-owner enforcement, and WI-772’s product-language enforcement ledger.
  • Candidate self-tests passed: verify.mjs --self-test (7 path/tier, 122 gate paths, cache contract), land.mjs --self-test (4 lock + 5 verdict + 8 worktree-guard + 10 status-preflight), session-start.mjs --self-test (6 unclaimed-id + 16 claim-boundary), check-working-rules.mjs --self-test (29 cases), and check-ids.mjs (700 items, 25 ADRs, 5 sprints, no duplicates/dangling citations).
  • pnpm verify --fast completed 111/113 gates. The two failures (product-language and product-language-self) were attributed as inherited because the temporary worktree has no installed typescript package; they are recorded as an environment limitation, not introduced by this research change.
  • Licence evidence: candidate root has no LICENSE, LICENCE, COPYING, or NOTICE, and root package.json has no license field. The research therefore says licence not established and makes no all-rights-reserved or open-source inference.
  • Source safety: only docs/research/** and this work item are changed in this repository; no files under the Rift Forge source checkout or modules//CLI implementation files were edited.

Review

approved — 2026-08-15.

Acceptance review

  1. Met. docs/research/repos/rift-forge.md pins local candidate/0.1.0 to the full 4a51848cfc9a2acbcdeddcd028418572406e2950, names the source checkout and read date, records the local-vs-remote divergence, and states that licence evidence is not established. Headline counts have named read-only commands and a 2026-08-15 measurement note.
  2. Met. The extraction now covers current setup, 14 skills, 85 scripts, 181 package scripts, current gates/skills, the candidate delta (WI-829, F-323/F-360, WI-790, WI-772), failure and retirement evidence, current worktree pain, a refreshed pain table, and an updated verdict. The former 1,513-line/401-branch/51-worktree claims remain only as explicitly historical comparisons.
  3. Met. docs/research/README.md, synthesis.md, harness-landscape.md, repos/hexguard.md, and pattern-catalog.md were searched for the old Rift Forge headline counts. Current comparison claims now use candidate measurements; incident counts that explain the historical design are labelled as pre-refresh evidence.
  4. Met. git -C C:\Development\Repositories\rift-forge status --short --branch remained clean; no files under that checkout, modules/, or CLI implementation changed.
  5. Met with one recorded limitation. npm run rungs -- check: 20 pass / 0 fail; site npm run build: 89 pages built (pre-existing duplicate-content-id warnings); site npm run check: 0 Astro diagnostics and 1,134 internal links / 0 broken. npm test: 5 pass. Candidate pnpm verify --fast: 111/113 gates completed, with the two inherited missing-typescript product-language failures recorded in Execution rather than hidden.

The review is limited to research and evidence propagation. No source-repository mutation, module change, or catalogue decision was smuggled into the item; follow-ups remain available if the refreshed candidate evidence changes a pattern decision later.