fix(stats): surface real MCP client labels (drop k-anonymity floor; allowlist labels are public)#80
Merged
LakshmanTurlapati merged 1 commit intoMay 25, 2026
Conversation
…keeper
- showcase/server/src/telemetry/housekeeper.js: K_ANONYMITY_FLOOR 2 -> 1
- Rewrite top-of-file + inline comment blocks documenting v0.9.69 -> v0.9.70 -> next-version floor transition and the public-allowlist rationale (MCP_CLIENT_ALLOWLIST is a fixed PUBLIC 13-label set; the label set itself carries no identifying information beyond per-label install counts, which are already exposed via total_users)
- tests/server-telemetry-housekeeper.test.js: destructure K_ANONYMITY_FLOOR from housekeeper exports; replace single k=2 "Other" bucket assertion with two assertions pinning K_ANONYMITY_FLOOR === 1 and asserting both real labels (Claude + Codex) surface as separate rows with uniq=1 (no "Other" bucket)
- Fixes live /stats "Popular agents" doughnut showing 100% "Other" (3 distinct installs, 3 distinct mcp_client labels, each uniq=1 collapsed into a single "Other" bucket)
- No extension/** changes; no API contract changes (popular_mcp_clients [{label,uniq}] shape unchanged); no DB schema changes
- All four required tests exit 0: server-telemetry-housekeeper (25 passed), server-public-stats-headline (33 passed), server-telemetry-allowlist (15 passed), fsb-telemetry-service (68 passed)
LakshmanTurlapati
deleted the
fix/stats-popular-agents-no-k-anonymity-on-public-allowlist
branch
LakshmanTurlapati
pushed a commit
that referenced
this pull request
Jul 4, 2026
Brings main's 62 post-branch-point commits into the FSB v0.9.90 / MCP v0.10.0 release branch and resolves the 62 resulting conflicts. Merge direction is main -> automation so PR #91 (automation -> main) can fast-forward afterward. Versioning & rebrand: - Keep automation's versions (FSB 0.9.90, MCP 0.10.0); reverse-DNS io.github.fullselfbrowsing/fsb-mcp-server already matched on both sides. - Adopt main's fullselfbrowsing org repo URLs (root package.json homepage/ repository/bugs, README + mcp/README shields/links); keep the personal author profile and the ClawHub URL (clawhub.ai/lakshmanturlapati/...). - Union mcp/CHANGELOG (0.10.0 -> 0.9.3 -> 0.9.2). Server-side data-correctness fixes carried from main: - Telemetry K_ANONYMITY_FLOOR 2->1 (#80) + popular_mcp heartbeat-row exclusion (RCA 260530-19i), keeping automation's region k-floor apparatus. - Locale-prefixed prerender routing (#76) grafted onto automation's fuller marketing/client-shell route sets (incl. /legal, /lattice, /phantom-stream). Skills / Hermes: - Adopt main's skills/FSB Skill/ -> skills/fsb/ rename and Hermes support (print-hermes-yaml.mjs, hermes-tool-prefix.md, install-host Hermes block); port automation's tool-decision-tree tweak; extend the skill-spec token allowlist for automation's capability/trigger tokens. Extension runtime: - Graft main's owned-tab chat-input lock, storage.onChanged live refresh, and canonical MCP client labels onto automation's rewritten sidepanel/popup; owner-chip exports both label helpers (lookupClientLabel + clientLabelFor). Showcase (automation's redesign wins UI; carry only main's data/logic): - Keep automation's globe /stats, PhantomStream dashboard, redesigned privacy pages, and routes; reconcile the dashboard-runtime-state test to automation's eager CDN-loader API. Regenerate llms-full.txt. Validated locally: validate:extension, full npm test (200+ files), showcase build (6 locales, no missing translations), lint:i18n, mcp smoke, CI-02 i18n extraction, and a simulated-Node-20 run of the catalog gate all pass.
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
/stats "Popular agents" tab shows 100% "Other" instead of real MCP client names. Live evidence:
Three distinct installs today, each with a different
mcp_client, each atuniq=1, each belowK_ANONYMITY_FLOOR=2inshowcase/server/src/telemetry/housekeeper.js:31, so the housekeeper collapses them all to "Other".Root cause
K-anonymity floor was designed for free-form labels where a unique entry could fingerprint a single user. But
MCP_CLIENT_ALLOWLISTinshowcase/server/src/routes/telemetry.js:47is a fixed public 13-label set (Claude, Codex, ChatGPT, Perplexity, Windsurf, Cursor, Antigravity, OpenCode, OpenClaw, OpenClaw 🦀, Grok, Gemini, Hermes, plusunknown). The labels themselves leak no identifying value -- only the install count does, which is already exposed via thetotal_usersheadline anyway.Fix
Lower
K_ANONYMITY_FLOORfrom2to1in housekeeper. The aggregation math is preserved verbatim and becomes a no-op at floor=1:selectPopularMcpForDayRangeGROUP BY mcp_clientalways emitsuniq >= 1rowsabove = popularMcpRaw(everything),belowInstalls = 0, "Other" bucket suppressedRewrote the two explanatory comment blocks in housekeeper.js to cite the new rationale (public allowlist → no identifying value at uniq=1). Updated
tests/server-telemetry-housekeeper.test.jsto pinK_ANONYMITY_FLOOR === 1as a regression guard and assert that two distinct mcp_client labels with uniq=1 surface as separate rows (not a single "Other" bucket).Non-impact
popular_mcp_clientsshape ([{label, uniq}]) unchanged.OtherJSON intelemetry_global_aggregates.popular_mcp_jsonon first tick after deploy (boot-tick fires viasetImmediateinstartHousekeeper, so within ~1 min of deploy).Test plan
node tests/server-telemetry-housekeeper.test.js-> 25 passednode tests/server-public-stats-headline.test.js-> 33 passed (API contract unchanged)node tests/server-telemetry-allowlist.test.js-> 15 passed (orthogonal: tests 'Other' as a rejected ingest value)node tests/fsb-telemetry-service.test.js-> 68 passedgit diff --name-only extension/-> empty (constraint satisfied)curl https://full-selfbrowsing.com/api/public-stats/globalreturns real label names within ~1 min of fly.io rollout🤖 Generated with Claude Code