Skip to content

experiment: KB identity, gap analysis, absorption, docs, and the prose cases - #1063

Closed
leeovery wants to merge 6 commits into
feat/experiment-phase-integrationfrom
feat/experiment-phase-crosscutting
Closed

experiment: KB identity, gap analysis, absorption, docs, and the prose cases#1063
leeovery wants to merge 6 commits into
feat/experiment-phase-integrationfrom
feat/experiment-phase-crosscutting

Conversation

@leeovery

Copy link
Copy Markdown
Owner

PR5 of the experiment-phase stack (design: #1057, stacks on #1062) — the closing slice.

  • KB: experiment paths derive identity (experiment/{topic}/E{n}-{slug}/(design|report).md → phase experiment, topic-shared); per-file replacement keyed on the canonical source path (indexing report.md no longer would wipe design.md's chunks); bulk discovery off the manifest register; remove --phase experiment scoped; new chunking profile (confidence medium); bundle rebuilt with the source. Engine: INDEXED_PHASES, manifest resolve, reactivate re-index. PR3's conclude indexing verified end-to-end against a throwaway world.
  • Gap analysis: reads completed experiment reports beside research and discussions; SOURCE_PHASES gains experiment so live sessions defer the analyses; the dispatch/read-in/category prose and epic-continuation §B updated.
  • Absorption: the experiment item travels whole (records, verdicts, reasons, the discussion's awaiting_experiments), series directory moved, completed series re-indexed at epic identities; pivot verified inheriting as-is. Simulation permutations for both.
  • Docs: CLAUDE.md phase entry Add /start-discussion command and update skill for directory structure #3 + layout/conventions paragraphs; README + docs/work-types.md pipelines.
  • Prose cases (authored clean, none walked): discussion-hits-the-empirical-wall, research-concludes-into-experiments, experiment-walk-honours-the-freeze, experiment-abandonment-releases-the-wait — deterministic perimeter green (110/110 snapshots).
  • Deliberately still two-way: frozen migrations, the requeue pair. The reported roadmap-flag gap was stale — flagJoined already walks all three phases.

Gates: npm test 2687/2687 · npm run test:cli 5/5 · npm run typecheck clean.

🤖 Generated with Claude Code

leeovery and others added 6 commits August 30, 2026 21:01
…ile replacement, engine scopes

deriveIdentity learns .workflows/{wu}/experiment/{topic}/{Eid}-{slug}/(design|report).md
— both files index under the topic's identity, the record dir riding the chunk's
source path. Replacement is per-file (removeBySourceFile; ids carry the record),
bulk discovery enumerates completed series from the manifest register, and the
per-file isIndexed check keeps bulk and status honest. Engine side: kb.cjs owns
INDEXED_PHASES and experimentArtifactPaths, manifest resolve answers the series,
topic reactivate re-indexes it, and absorb moves the experiment item, records,
directory, and the discussion's evidence wait whole — with simulation
permutations for the mid-experiment absorb and the pivoted series.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… on live experiment sessions

collectAnalysisInputs joins completed experiment reports into the gap-analysis
input set (stamp names them past the basename), SOURCE_PHASES gains experiment
so a live experiment session defers the epic-wide analyses, and the two-way
research/discussion enumerations the landed machinery made three-way are swept:
the analysis prose set (dispatch signpost, precondition, read-in, stamp), the
routing decision's third answer, the bridge's §B wording, the epic reactivate
prose's indexed-phase list, summary-backfill's source read, and the engine
reference's commit/presence lines.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…type pipelines

CLAUDE.md gains the experiment phase entry (series per topic, the
design-before-measurement gate, evidence waits, measures-never-decides), the
directory-layout row, the three-way pipeline lines, and the presence / KB /
soft-gate / staleness-hop / absorption / cancellation paragraphs the landed
machinery made three-way. README and docs/work-types.md pick up the optional
Experiment slot in the user-facing pipelines, the KB and self-healing lines,
and one feature bullet.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…e edges

discussion-hits-the-empirical-wall — the mid-discussion exit: gate, waiting
note, straight-through hand-off, and the first conceive recording the
evidence wait (create→await→commit pinned in order).
research-concludes-into-experiments — the conclude gate's needs-measuring
arm: complete-and-index unchanged, then the experiment entry, never the
bridge. experiment-walk-honours-the-freeze — the conceive→conclude mainline
with approve pinned between the design's advance and measurement's, against
a world whose numbers the walk really counts.
experiment-abandonment-releases-the-wait — the discussion's re-entry after
an abandonment: the advisory surfaces the row and its reason, the flag
clears, the point reverts to open, and the conclusion is no longer refused.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ade three-way

The map's fresh gate, the review triage guard, the absorb reconcile comment,
the analysis-cache grade line, and planning's sole-input note all cover
experiment now; the topic-complete reference notes the series' per-file
indexing. Left deliberately two-way: the frozen migrations, the requeue
offer (a concern moves within the research/discussion pair only), and the
roadmap flag's research/discussion targets.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@leeovery

Copy link
Copy Markdown
Owner Author

Closed: this stack implements a superseded model of the experiment phase. The corrected design is PR #1057 (design/experiment-phase.md) — experiment as an umbrella tool spawned from research/discussion, not a peer pipeline phase. A fresh stack built against that design replaces this one. The branch is kept for salvage until the replacement lands.

@leeovery leeovery closed this Aug 31, 2026
@leeovery
leeovery deleted the feat/experiment-phase-crosscutting branch September 1, 2026 17:56
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant