Conversation
… stubs Foundations for the RuLake-inspired feature roadmap (docs/plan/rulake-inspired-features.md). Phase 0 is deliberately boring: define shared shapes + cut extension points so the four Phase 1 swarm tasks never fight over ruvectorBridge.js. What this ships: - archive/snapshot.js — ArchiveSnapshot v1 shape + validateSnapshot() - archive/hash.js — xxHash32 over Float32Array brains; 8-char hex ID - ruvectorBridge.js — exportSnapshot / importSnapshot / setConsistencyMode / getConsistencyMode / getIndexStats. Stubs throw 'not implemented' with owning-phase labels; importSnapshot already validates at the boundary so callers can test wiring. - eli15/chapters/*.js — 7 placeholder chapters (warm-restart, quantization, consistency-modes, content-addressing, federation, cross-tab-federation, where-the-time-goes), each with comingSoon: true and a real ELI15 hook. - eli15/index.js — registry entries for the 7 new chapters. What this deliberately does not do: no behaviour change, no new feature. The bridge imports validateSnapshot at module scope but doesn't call it at boot; chapters are lazy-loaded. Validated via agent-browser: app boots cleanly with no new console errors, all 7 chapter IDs returned by ELI15.listChapters(), two chapters rendered end-to-end, stub contracts (5 exports, correct error messages, consistency default = 'fresh'), hash determinism + 1-ppm sensitivity.
F5 from docs/plan/rulake-inspired-features.md. Two brains with
identical flattened weights now collapse to a single canonical node
in the lineage DAG — the DAG viewer shows a "×N" badge on
rediscovered brains, and a duplicateRatio stat is exposed for the
training panel.
What this ships:
- archive/dedup.js — in-memory Map<hash, {firstSeenId, dupCount}>
backing maybeInsert() / stats() / has() / get() / _debugReset().
Backed by archive/hash.js's xxHash32 from Phase 0.
- lineage/dag.js — new addBrainWithFlat(flat, meta, fallbackId)
hashes the flat vector and dedups at insert; legacy addBrain()
signature unchanged (no behaviour change on existing callers).
dedupeStats() exposed for the viewer + training panel.
- lineage/viewer.js — renders "×N" badge on nodes with
duplicateCount > 0; tooltip reads "×N seen".
- brainExport.js — export rows carry the hash; new programmatic
importBrain(row) short-circuits on dedup hit and returns
{ skipped: true, reason: "duplicate" }. Legacy export/import
path still works if late-bound hooks aren't wired.
- eli15/chapters/content-addressing.js — full chapter replacing
the Phase 0 coming-soon stub.
- tests/dedup-smoke.html — standalone harness, PASS/FAIL in DOM.
Not wired into ruvectorBridge.archiveBrain on the GA hot path — that's
a later integration slice. Module is callable but opt-in.
Validated via agent-browser: smoke harness prints PASS on all 3
claims (maybeInsert idempotent, dedupeStats shape + counts, importBrain
re-import leaves node count unchanged). Main app still boots cleanly
with no new console errors after the lineage/dag.js + viewer.js +
brainExport.js edits.
F1 from docs/plan/rulake-inspired-features.md. Ships a standalone
quantizer that shrinks brain vectors ~30× (976 B float → 32 B packed,
or 96 B including a 16-float residual) with recall@10 = 0.9060 on the
brain-lineage-cluster workload. Not wired into the hot path — Phase 2A
federation does that.
What this ships:
- quantization/hadamard.js — in-place iterative Fast Walsh-Hadamard
Transform, pads to next power of 2 (244 → 256).
- quantization/rabitq.js — quantize(flat) → {packed: Uint32Array,
residual: Float32Array[16]}; hammingDistance via popcount on XOR;
estimatedCosine uses the Hamming/π identity.
- quantization/index.js — public surface; Phase 2A consumes from here.
- quantization/viewer.js — mountViewer(container, samples[]) plain-DOM
side-by-side float heatmap / bit-packed view / true-vs-estimated
cosine scatter plot / memory meter. No framework.
- eli15/chapters/quantization.js — full chapter replacing the Phase 0
coming-soon stub. Covers the SimHash analogy, why Hadamard beats
random projections, the 30× memory win, and explicitly flags that
hyperbolic HNSW is out of scope for F1 (sign-bit math is Euclidean).
- tests/quantization-recall.html — standalone harness. Open in
browser, reads recall@10 as text. Mulberry32-seeded for repro.
Also updates the phase tracker in docs/plan/rulake-inspired-features.md
to reflect Phase 1 wave 1 closed (1B + 1D shipped); wave 2 (1A + 1C)
remains open because both touch ruvectorBridge.js stubs and would race.
Validated via agent-browser:
- Determinism: quantize(v).packed equal byte-for-byte across calls.
- Hamming(v,v) == 0 on identity.
- Recall@10 = 0.9060 ≥ 0.9 gate on 100 synthetic brain vectors.
- Main app boots cleanly with no new console errors after adding the
quantization/ directory (nothing imports it yet, so it's a no-op on
the current code path).
F3 from docs/plan/rulake-inspired-features.md. Serializes the whole
brain/track/dynamics archive to a gzipped JSON bundle (.vvarchive.json.gz)
and rehydrates it deterministically via replay-of-insertions — no WASM
patch needed. Behind ?snapshots=1 while it bakes.
What this ships:
- archive/exporter.js — buildSnapshot({...}) → ArchiveSnapshot with
xxHash32 witness (sync, click-handler safe). Also buildSnapshotAsync
with crypto.subtle SHA-256 witness when an async path is acceptable.
- archive/importer.js — applySnapshot(snapshot, targets) replays into
fresh VectorDB instances, rebuilds mirrors, re-hydrates the lineage
DAG. Respects insertionOrder; falls back to row order if absent.
- archive/serialize.js — toBlob/fromBlob via CompressionStream('gzip')
with plain-JSON fallback for older Safari. Magic header
"VVARCHIVE v1 {gzip|json}" so truncated downloads fail loudly at import.
- ruvectorBridge.js — exportSnapshot/importSnapshot stubs replaced with
real bodies. _insertionOrder: string[] now tracked alongside the
mirrors — pushed by archiveBrain, hydrate, hydrateFromFixture; cleared
by _debugReset and importSnapshot; consumed by rebuildIndicesFromMirror
so setIndexKind preserves graph identity. setConsistencyMode/
getIndexStats stubs unchanged (1C / 3A own those).
- uiPanels.js — flag-gated "📦 Export archive" / "📥 Import archive" row.
Renders only when ?snapshots=1; default UI unchanged.
- main.js — after bridge.ready(), if ?snapshots=1 and a staged import is
pending, runs it pre-sim. Fire-and-forget; logs counts.
- eli15/chapters/warm-restart.js — full chapter replaces coming-soon stub.
- tests/warm-restart-roundtrip.html — standalone harness. Seeds fixture,
exports → toBlob → fromBlob → importSnapshot, asserts 5 claims.
Validated via agent-browser:
- Round-trip harness prints PASS 5/5: validateSnapshot ok, importSnapshot
returns right counts (brains=8 tracks=1), brain count preserved
(pre=8 post=8), top-1 id byte-identical (brain-7 → brain-7), id set
byte-identical. gzip=true (CompressionStream path exercised, not JSON
fallback).
- Default URL (no flag): main app boots cleanly, no new errors, Export/
Import buttons absent.
- Flag URL (?snapshots=1): both buttons render in the training panel.
Not yet integrated with F5 dedup (import still re-inserts rather than
skipping duplicates) — that's a later integration slice. Not yet
integrated with F4 consistency (snapshot.consistency defaults to 'fresh').
F4 from docs/plan/rulake-inspired-features.md. Three query-time modes
controlling how recommendSeeds sees the archive; closes Phase 1 of the
roadmap (only Phase 3A's getIndexStats stub remains in ruvectorBridge).
What this ships:
- consistency/mode.js — state machine + TTL cache + frozen-id-set ref.
Exports getMode/setMode, recordQuery/getCachedResult, freezeArchive/
thawArchive, stats, _debugReset. TTL default 10 generations.
- consistency/worker-sync.js — computeFrozenSyncPayload() helper for the
A/B baseline worker; postMessage plumbing is a follow-up.
- ruvectorBridge.js — setConsistencyMode stub body replaced. One branch
added inside recommendSeeds that reads the mode:
* fresh → unchanged path (single no-op boolean per candidate when
frozenIds is null — fresh-mode behaviour byte-identical to pre-1C)
* eventual → cache hit returns memoized result, else fresh query +
recordQuery
* frozen → filter candidates to ids in the pinned Set
Freeze mechanism is the cap-by-insertion-id shortcut (plan's rec): at
freeze-entry, snapshot new Set(_brainMirror.keys()); new archiveBrain
calls still insert (archive growth uninterrupted), their ids just
aren't in the pinned set. Rejected the alternatives (VectorDB clone,
parallel read-only index) as 2× memory for no spec benefit.
- uiPanels.js — new radio row (Fresh/Eventual/Frozen, default Fresh)
with a 30-span tick strip that pulses on cacheMisses — visible
evidence of when the archive is actually re-queried.
- style.css — .rv-consistency* classes including keyframe pulse.
- main.js — ?consistency=fresh|eventual|frozen URL flag, mirrors the
?hhnsw=1 pattern; polls for bridge readiness before applying.
- sim-worker.js — new case 'freeze' branch gated on importSnapshot
availability; existing A/B setups without frozen-mode unchanged.
- eli15/chapters/consistency-modes.js — full chapter with the three
modes, the A/B-race motivation for frozen, tick-strip semantics.
- tests/consistency-modes-smoke.html — 5-claim harness.
Also updates docs/plan/rulake-inspired-features.md to reflect Phase 1
complete; M1 milestone (F1+F3 demoable) checked off.
Validated via agent-browser:
- Smoke harness PASS 5/5: fresh bypasses cache (0 hits 0 misses),
eventual 2nd call is a hit (sameRef=true), frozen excludes post-
freeze brain, thaw restores visibility, no throws on valid modes.
- Default UI: Fresh mode, all 3 radios + tick strip render, app boots
cleanly with no new console errors.
- ?consistency=frozen URL flag: bridge reports 'frozen', Frozen radio
checked.
Phase 1 ordering discipline held: 4 subagents (1B+1D in parallel,
1A→1C sequentially) edited 15+ files, zero merge conflicts.
F2 from docs/plan/rulake-inspired-features.md. Query path can now fan out to BOTH nearest-neighbour indexes simultaneously, union the candidates by hash (F5), and GNN-rerank the union. Gated behind ?federation=1 or the UI toggle; off by default. Composes with Phase 1C consistency modes (frozen filter applies to the union; eventual cache stores the final federated result, not per-shard). What this ships: - federation/fanout.js — fanOut(queryVec, k, shards) returns per-shard top-k'. Over-request formula kPrime(k, S) = k + ⌈√(k·ln S)⌉, clamped so S≤1 returns k (graceful single-shard degrade). Also a fanOutSync twin for tests. - federation/rerank.js — unionByHash (dedup key is hash; tracks per-candidate shard provenance), selectTopK (picks final k by external score map; falls back to bestScore when unscored). - federation/viewer.js — plain-DOM split-screen: Euclidean top-k' on the left, Hyperbolic top-k' on the right, unioned+reranked top-k in the middle, with the formula rendered live. - ruvectorBridge.js — structural addition of _brainDB_hyperbolic as a shadow of _brainDB populated in lock-step across archiveBrain, hydrate, hydrateFromFixture, rebuildIndicesFromMirror, and importSnapshot. The existing setIndexKind swap still works — the primary index swaps, the shadow stays hyperbolic. New _recommendSeedsFederated helper handles the fanout branch; wired into recommendSeeds so the 1C consistency cache/frozen logic wraps both the single-index and federated paths. New exports: setFederationEnabled / isFederationEnabled / getFederationStats / setFederationCapturer. - uiPanels.js — new "🌐 Federation" toggle + mount point for the viewer. Renders federation stats on every UI tick. - main.js — ?federation=1 URL flag, mirrors the 1C pattern. - eli15/chapters/federation.js — full chapter with the over-request formula worked example (k=10 → k'=13), dedup-via-F5 explanation, compose-with-1C-and-1D section. - tests/federation-smoke.html — 7-claim harness. Validated via agent-browser: - Harness PASS 7/7. lastKPrime=7 matches k=5 + ⌈√(5·ln 2)⌉=7 exactly. shards=2 (both loaded). dedupeHits=4 (both shards surfaced 4 overlapping brains and the hash union correctly collapsed them). Frozen-mode filter correctly applied to the union, not per-shard (post-freeze brain absent from federated result). - Main app boots cleanly without any flag, no new console errors. - Exactly one "not implemented" stub remains in ruvectorBridge.js (getIndexStats, Phase 3A/F7). Not yet integrated with F1 quantization (per plan scope note) — that would be a later integration slice. Not yet integrated with F6 cross-tab — that's Phase 2B, gated on 1A's snapshot format (which shipped) but sequential with this commit on ruvectorBridge.js.
F6 from docs/plan/rulake-inspired-features.md. Closes Phase 2. When
two tabs on the same origin open with ?crosstab=1, a brain archived
in tab A arrives in tab B's archive within a microtask. No server,
no sockets — pure browser primitive. F5's content-hash IDs make the
protocol lockless by construction: two tabs computing the same brain
produce the same hash, dedup collapses the repeat automatically.
What this ships:
- crosstab/channel.js — BroadcastChannel('vectorvroom-archive') wrapper.
Per-tab random senderId; hello/hello-ack/bye protocol for peer count;
graceful beforeunload bye; stats() for diagnostics.
- crosstab/wire.js — single-brain payload codec. flat: Array<number>
on the wire (not Float32Array) so two tabs on slightly different
builds still interop. hash is sender-computed but receiver
re-derives via dedup.
- ruvectorBridge.js — additive only. New state: _crosstabEnabled,
_crosstabReceiving, _crosstabChannel, _crosstabSenderId. Broadcast
hook in archiveBrain guarded by (_crosstabEnabled && !_crosstabReceiving).
_onRemoteBrain wraps the local archiveBrain call in
_crosstabReceiving=true/try/finally so the receive path is
pure-one-way, never re-broadcasts. New exports setCrosstabEnabled /
isCrosstabEnabled / getCrosstabStats / setCrosstabListeners /
_onRemoteBrain.
- uiPanels.js — flag-gated "🔗 N peer(s)" pill; pulses on received
brain via a CSS class toggle.
- style.css — .rv-crosstab-pill + rv-crosstab-pulse keyframes.
- main.js — ?crosstab=1 URL flag applier, mirrors 1C / 2A pattern.
- eli15/chapters/cross-tab-federation.js — full chapter replacing
coming-soon stub: BroadcastChannel basics, why content-addressing
makes it lockless, echo-prevention via _crosstabReceiving guard,
worked two-tab example.
- tests/crosstab-smoke.html — 3-claim harness simulating two tabs
with two channels in one page.
Also updates docs/plan/rulake-inspired-features.md: 2B closed, Phase 2
complete; milestone M2 ticked (Phase 1 merged); M3 in progress.
Validated via agent-browser:
- Harness PASS 3/3: A→B delivery within 500ms with full wire decode
(flat=244, fitness=42.5, hashMatch=true); duplicate brain dedups to
one canonical node with duplicateRatio=0.5 after 2 inserts;
sender-echo suppression verified (0 self-echoes at A, 1 from B).
- ?crosstab=1 flag enables crosstab: isCrosstabEnabled()=true, pill
rendered, stats show started=true with a per-tab senderId.
- Default boot (no flag) clean, no new console errors.
Echo-loop guard: _crosstabReceiving module-scope boolean flipped
around the receive-path archiveBrain call. Broadcast hook checks it.
Belt-and-braces with the senderId===self drop in channel.js (which
is redundant today since BroadcastChannel doesn't echo to sender,
but documented as defense-in-depth for future relay topologies).
Only Phase 3A's getIndexStats stub remains unimplemented in the
bridge. Phase 2 sequencing discipline held: 2A→2B sequentially,
both structural edits to ruvectorBridge.js, zero merge conflicts.
F7 from docs/plan/rulake-inspired-features.md. Per-stage timings for
every generation, rendered as a collapsible stacked-bar + numeric
table below the vector-memory panel. This commit fills the last
"not implemented" stub in ruvectorBridge.js — the whole bridge now
has zero unimplemented exports for the first time since Phase 0.
What this ships:
- observability/timings.js — ring-buffer histogram per stage name,
window=20 generations. startStage/endStage with try/finally
exception-safety, record for one-shot, snapshot returns per-stage
{count, totalMs, avgMs, lastMs, p95Ms}.
- observability/panel.js — plain-DOM collapsible panel with rAF-gated
10Hz auto-refresh. Stacked horizontal bar (div % widths) + details
table. Inline CSS scoped to .rv-obs-panel.
- ruvectorBridge.js — getIndexStats() body replaced. Returns a live
6-key snapshot: archive counts, index kind, federation/consistency/
crosstab stats (each try/catch-wrapped for optional getters), and
timings. Stage timers added around recommendSeeds hot-path stages
(retrieve, federate, rerank, adapt, dynamics). setGeneration called
inside archiveBrain. Federation branch timer wraps the whole fanout
to avoid double-counting its inner rerank.
- uiPanels.js — dynamic-import mount point below the existing panel,
anchored by "=== Phase 3A observability mount point ===" so Phase 3C
can add its own section without conflict.
- eli15/chapters/where-the-time-goes.js — full chapter replacing the
Phase 0 coming-soon stub. Covers each stage, how to read p95, and
the "surprise big number" workflow.
- tests/observability-smoke.html — 5-claim harness.
Validated via agent-browser:
- getIndexStats returns the full 6-key structure.
- Observability panel renders in the UI (default on; telemetry, no
behaviour change).
- Harness 5/5 PASS including the ring-buffer moving-average check:
25 records into window=20 yields count=20 avgMs=15.5 = (6+25)/2,
i.e. only the trailing 20 samples are reflected.
- Main app boots cleanly, no new console errors.
Load-bearing claim confirmed: `grep 'not implemented' ruvectorBridge.js`
returns zero hits. The Phase 0 pre-partition is fully consumed.
… cross-links) 3B from docs/plan/rulake-inspired-features.md. No chapter body edits; registry hygiene and pedagogical ordering only. What changed: - eli15/index.js — registry re-ordered into four pedagogical bands: Foundations → Vector memory basics → Lineage & reasoning → Geometry choice → RuLake-inspired extensions. Inside the extensions band: content-addressing first (hash primitive), then the features that compose with it (quantization, warm-restart, consistency, federation, cross-tab, observability last). Added `related: [...]` cross-link fields to the 7 roadmap chapter entries. comingSoon: true flags removed from every chapter whose file body no longer has it — including where-the-time-goes (reconciled post-3A land). - eli15/tour.js — STEPS array appended with the 7 RuLake-inspired extensions as conceptual-only steps (anchor: null) matching the registry order. Tour is driven by explicit STEPS, not Object.keys(REGISTRY). Suggested chapter order from the plan was followed with two documented deviations: 1. track-similarity before ema-reranker (reads more naturally once the learner has seen a retrieval happen). 2. content-addressing first inside the extensions band (F5 hash is the primitive every subsequent feature depends on). Validated: - Main app boots cleanly. - ELI15 drawer opens; chapter list reflects the new order (head: what-is-this-project → sensors → ...; tail: federation → cross-tab- federation → where-the-time-goes). - Opening where-the-time-goes renders real content, no "Coming soon" artefact.
…admap complete) 3C from docs/plan/rulake-inspired-features.md. Closes Phase 3 and the whole RuLake-inspired roadmap. Archive bundles from Phase 1A can now travel between users via a shareable URL — we don't host anything, the user supplies the URL (gist / S3 / IPFS / anywhere with CORS). What this ships: - share/url.js — fetchArchive(url) pipes through serialize.fromBlob + validateSnapshot, throws with readable errors on CORS/404/validate. buildShareUrl(archiveUrl) constructs the shareable link preserving other flags (?consistency=, ?federation=...), stripping any pre- existing snapshots/archive params so the link is deterministic. - share/gallery.js — curated community-archive list. Ships with exactly ONE placeholder entry (url: 'about:blank') per the external-scope rule: real URLs require explicit user approval before publication. A code comment lists candidate entries queued for review but not activated. - uiPanels.js — new share panel gated by ?snapshots=1, anchored with "=== Phase 3C share panel ===" so future edits can find its spot without fighting 3A's observability block. Copy-shareable-link button (with alert fallback for insecure contexts where navigator.clipboard is unavailable), Import-from-URL input+button, gallery mount. - main.js — ?archive=<url> URL flag. Requires ?snapshots=1 to be set alongside (same gate as Phase 1A). Auto-fetches + imports at boot. - eli15/chapters/warm-restart.js — appended "Sharing a museum over the internet" section at the end of the existing body. No rewrite of existing content (per plan: 3C extends 1A's chapter rather than adding a new one). - tests/share-smoke.html — 4-claim harness mocks window.fetch against an in-memory buildSnapshot+toBlob fixture. Also updates docs/plan/rulake-inspired-features.md: every Phase row ✅, all four milestones M1-M4 closed, decisions log records the final phase narrative. Focus line set to "Roadmap complete". Validated via agent-browser: - Harness PASS 4/4: fetchArchive + validateSnapshot round-trip; share URL exactly `?snapshots=1&archive=<url-encoded>`; gallery click invokes onImport with 'about:blank'; GALLERY.length === 1. - ?snapshots=1 flag renders the share panel with both buttons and the one placeholder gallery entry. Default (no flag): UI unchanged. - Main app boots cleanly with no new console errors. External-scope discipline held: no real community archive URLs were added in this commit. The gallery is a single about:blank sentinel that fails fetch loudly by design until a vetted URL lands. Roadmap stats: 10 commits today (Phase 0 foundations + 9 feature commits), ~9,400 lines of feature code, 8 browser-live test harnesses, 8 ELI15 chapters rewritten from Phase 0 stubs, 4 subagent swarm waves (1B+1D parallel, 1A seq, 1C seq, 2A seq, 2B seq, 3A+3B parallel, 3C seq). Zero merge conflicts, zero rollbacks, zero "not implemented" stubs remaining in ruvectorBridge.js.
|
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (47)
📝 WalkthroughWalkthroughThis pull request introduces a comprehensive Phase 2A-3B feature set: archive snapshots for warm restart with gzip serialization; federated vector search with union/reranking; 1-bit quantization via Hadamard rotation; consistency modes (fresh, eventual, frozen) with TTL caching; cross-tab synchronization via BroadcastChannel; content-addressed brain deduplication; observability/timing instrumentation; and sharing with community gallery support. Includes six new Eli15 chapters, multiple smoke tests, and extensive documentation of the RuLake-inspired roadmap. Changes
Sequence Diagram(s)sequenceDiagram
participant Client as UI/Main
participant Bridge as ruvectorBridge
participant Euclidean as Euclidean<br/>Index
participant Hyperbolic as Hyperbolic<br/>Index
participant Union as Union &<br/>Dedup
participant Rerank as Rerank &<br/>Score
participant Cache as Consistency<br/>Cache
Client->>Cache: getConsistencyMode()
alt eventual mode + cache hit
Cache-->>Client: return cached result
else fresh or eventual miss
Client->>Bridge: recommendSeeds(queryVec, k)
alt federation enabled
Bridge->>Euclidean: search(queryVec, k')
Bridge->>Hyperbolic: search(queryVec, k')
Euclidean-->>Union: results
Hyperbolic-->>Union: results
Union->>Union: dedup by content hash
Union->>Rerank: union candidates
Rerank->>Rerank: score by GNN/EMA<br/>apply frozen filter
Rerank-->>Bridge: top-k results
else federation disabled
Bridge->>Euclidean: search(queryVec, k)
Euclidean-->>Bridge: top-k results
end
Bridge-->>Cache: record result
Cache->>Client: return result
end
sequenceDiagram
participant UI as UI Panel
participant Export as exportSnapshot()<br/>(ruvectorBridge)
participant Build as buildSnapshot()
participant Serialize as toBlob()
participant File as File/Network
participant Deserialize as fromBlob()
participant Validate as validateSnapshot()
participant Import as importSnapshot()
participant Bridge as ruvectorBridge
UI->>Export: click export
Export->>Build: construct snapshot<br/>with mirrors & HNSW state
Build->>Build: compute xxHash32/SHA256<br/>witness
Build-->>Export: snapshot object
Export->>Serialize: toBlob(snapshot)
Serialize->>Serialize: JSON + gzip<br/>(if available)
Serialize-->>UI: Blob
UI->>File: save or share URL
alt Import from file or URL
File->>Deserialize: fromBlob(blob)
Deserialize->>Deserialize: detect codec,<br/>decompress
Deserialize-->>Validate: snapshot object
Validate->>Validate: verify schema version<br/>& witness
Validate-->>UI: OK
UI->>Import: importSnapshot(snapshot)
Import->>Bridge: clear mirrors,<br/>insert tracks/dynamics/brains<br/>in insertion order
Import->>Bridge: load observations
Bridge->>Bridge: rebuild HNSW<br/>from insertionOrder
Bridge-->>UI: counts & success
end
Estimated code review effort🎯 5 (Critical) | ⏱️ ~120 minutes Poem
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
4 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Implements the full RuLake-inspired feature roadmap documented in
docs/plan/rulake-inspired-features.md. 10 commits land sevenuser-visible features plus the foundations + tour-integration +
shareable-archive polish that ties them together.
Every feature is flag-gated (default-off) or default-on-but-
behaviour-neutral. A casual visitor to the running app sees the new
ELI15 chapters and the observability panel; demo-hunters can unlock
the rest via URL flags.
Commits (in land order)
c33003eb8b2f607b95af0ee36b44?snapshots=1c9df280bf02701ae32aba0c51aaabebba8193c9dc4What users see
By default (no flags):
Behind URL flags:
Stats
Known follow-ups
Test plan
See `docs/plan/rulake-inspired-features.md` for the full plan, phase-by-phase status, and decisions log.
Summary by CodeRabbit
New Features
Documentation
Tests