We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
Docs sweep: brand ThrottleKit Lens; new Scaling/Policy Plans/Replay pages; 8-tab Lens + Monitor door - Home/_Sidebar: (TUI) -> (Lens), Policy-Plans compare row, links to the new pages - Monitoring-and-the-Lens: rebrand to ThrottleKit Lens, 5 -> 8 tabs (Cost Room/Replay/Plan), Monitor door section - new pages: Scaling-and-the-Fleet (Tier-1 + Tier-2 + leaseWindowed), Policy-Plans, Replay - Tier-1/Tier-2 cross-links + Monitor door across Polyglot/Operations/Federation/Pillar-4/Unified/Distributed/Performance/Middleware
docs: officialize TALE full form (Temporally-Accounted Learned Escrow) Match GALE's heading style and put both full forms on the Home intro.
Monitoring: document the five tabbed views + the fairEscrow/concurrency server sources
docs: monitoring is a built-in terminal dashboard (--tui), not the web Lens Rewrite the Monitoring page for the `throttlekit-server --tui` terminal dashboard (binding-axis hero + the actual TUI panels, keybinds, opt-in TTY, headless -> OTel/Grafana); update the Operations dashboard note, the Home feature/guide rows, and the sidebar link. The web Lens package is deprecated.
Monitoring & the Lens: note the 1.2.0 throttlekit.denies_by_axis escape hatch
Add the Monitoring & the Lens page; link from Home, Operations, and the sidebar
Polyglot: document the server's pluggable store backends The service-door section listed only --redis. Add the full store-backend set (--store memory|redis|postgres|dynamodb, inferred from the URL flags), note the client is store-agnostic (same requests, bit-identical decisions), and the edge-only exclusion (Deno KV / Cloudflare not reachable via the service door).
docs: align Home tagline to the live site (beyond rate limiting; govern rate, concurrency & cost, provably)
docs: lead the Home tagline with GALE & TALE
docs: de-academicize positioning — capability-led, no 'research' labels Home 'Research-backed' bullet -> 'Proven, and shipping'; guides/sidebar/cross- link labels 'Research: GALE & TALE' / 'research track' -> 'GALE & TALE'; Research page retitled 'the proven guarantees' and venue mentions (SIGMETRICS/POMACS/NSDI) removed. GALE/TALE kept as engine names; the proof signals (TLA+, fleet-size- independent) kept; FairRide NSDI'16 prior-art citation kept.
Home: add a competitor comparison table (obvious differentiators)
docs: add Polyglot & Python page (service door + throttlekit-py, every axis); link from Home + sidebar
docs: 1.0.0 — API frozen under SemVer; refresh perf numbers Footer no longer says "0.x, API may still be refined before 1.0" — it now states the 1.0 SemVer freeze and links STABILITY.md. Home + Performance perf figures refreshed to the measured 1.0.0 run (169 ns/op checkSync, ~300 ns async, twoTier(leased) ~35x on Postgres) and point at BENCH.md as the canonical perf doc.
Distributed & Provable: lease.adaptive wires GALE Pillar 2 into the lease loop (0.13.0)
Pillar-4: Federated Weighted Fair Escrow (cross-region) shipped (TK-1404) Add a "Federated Weighted Fair Escrow (cross-region)" section: the two-level WFE composition (region weight = Σ tenant weights, the Parekh-Gallager GPS collapse), the regionFairPool + federatedWeightedFairEscrow API + usage snippet, the guarantees table (T-FED-1 Σ≤L safety / T-FED-2 fluid exactness / T-FED-3 bounded discrete error), the demand-proportional weight split for region-spanning tenants, and the honest scope (T3 reserve gap, all-backlogged exactness, in-process pool now + store-backed next). Flip "Not federated / Not hierarchical at 0.9.1" to shipped (region->tenant GPS special case) and flip the roadmap federated-WFE row.
Unified-Admission: 3-axis concurrency shadow price shipped 0.11.3 Add the "Concurrency shadow price — the 3-axis filter" section: Little's law makes an occupancy cap L over a window T a concurrency-seconds budget K=L·T, per-admit consumption = hold time, so the bid test gains p_K*hold to reject a hold-time hog 2-axis is blind to (regret 53->2%). Honest scope: strictly-dominated hog only (can't ration a marginal one); fail-open on a missing/bad hold. Flip the future-work row to shipped; bump the design-record range to D-JLP-1..16. Accumulates for the 0.11.3 release push.
Unified-Admission: online dual refinement (jointLp.adaptive) shipped 0.11.3 Add the "Online dual refinement" section (guarded sample-then-price): the warm-up is priced with the prior, the re-solved duals are adopted only if they beat the prior on the observed sample, else kept. Honest scope: on-sample, not full-horizon (the rho=+1 foil's cousin). Flip the future-work row to shipped; bump the design-record range to D-JLP-1..14. Accumulates for the 0.11.3 release push.
docs(concurrency): 0.11.2 — PostgresConcurrencyCoordinator (shipped) + demand-proportional allocation (opt-in)
Unified-Admission: joint-LP bid-price policy (shipped 0.11.1, opt-in, ε=25.33%)
Distributed adaptive concurrency: eager handoff (D-DAC-20) + self-fencing (D-DAC-21) Document the 0.11.0 completion of the hard bound: eagerHandoff removes the acknowledged-handoff ramp-latency cost (near-floor ramp, no loosening of the bound; the pitch-perfect {acknowledgedHandoff, eagerHandoff} config), and self-fencing (default ON under fail-closed) closes the partition in-flight overshoot under the standard bounded-clock-skew assumption, with the FLP/CAP frontier stated honestly.
Distributed adaptive concurrency: document acknowledged handoff (opt-in hard async bound, D-DAC-19 / TK-1330)
Distributed-Adaptive-Concurrency: occupancy cap (D-DAC-18) The 0.10.0 coordinator reserves each peer's max(share, inflight), eliminating the synchronous rebalance overshoot in Σ inflight (a joiner no longer ramps into an incumbent's still-occupied capacity). Kept the honest framing: this is NOT a hard instantaneous bound across the async fleet — grant-reply + reporting lag leave a bounded (~1.5–2×), self-draining residual; a hard instantaneous bound would need acknowledged handoff. InflightCap is model-checked in the synchronous spec + twin.
docs(wiki): middleware integration page for 0.9.2 (TK-1326/TK-1328)
Pillar-4-Weighted-Fair-Escrow page + nav (TK-1312, queued for 0.9.1) New page: Pillar-4-Weighted-Fair-Escrow.md covering: - The 3-primitive fairness landscape comparison - The streaming algorithm (dynamic g_i + borrow phase with cost-cap) - T1-T4 guarantees + the T5 (FairRide) conceded vertex - L2 multi-process backing via existing Store + fixedWindow lease - Failure modes - Composition with combineDecisions / unifiedAdmission - Roadmap (federated WFE 0.10.x, hierarchical 0.10.x+) Sidebar + Home: added Pillar 4 entry pointing at the new page. Overload-Fairness-and-DDoS: expanded the weighted fairness section into a 3-row comparison table (weightedFairShare vs weightedMaxMin vs weightedFairEscrow), pointing at the new page for the multi-process work-conserving sibling. To push (release tag time only): cd C:\Users\ameya\tk-wiki && git push origin master
Add Unified-Admission page; update Home + Sidebar (0.9.0 prep — pushed at release tag time)
docs(federation): multi-process regional escrow + regional-only outage mode (0.8.5) Adds the TK-1306 user-facing documentation: Outage failure-mode table: - New row: regional-only outage mode (requires regionalEscrow); engine serves from L2 balance during coord outage; re-probes isHealthy() every coordinatorHealthCheckMs (default 5s); resumes on recovery - New row: multi-process within a region sharing a regionalEscrow; in-flight per-region escrow bounded by perKeyBudget instead of M*batch - New row: regional Redis (L2) outage; engine falls through to direct L3 leasing (matches 0.8.4 behavior); federation bound preserved New "Multi-process regional escrow (0.8.5)" section: - The two scenarios where an L2 helps (tight regional sub-bounds + the regional-only outage mode) - Quick-start code with RedisRegionalEscrow + RedisCoordinator wired together; onCoordinatorOutage: "regional-only" - The RegionalEscrow interface contract (lease/refill/release/isHealthy); TestRegionalEscrow as deterministic in-memory mirror - The regional-only outage mode semantics (clock-driven probe, recovery, bound degradation during outage) - Backward-compat note: federate() without regionalEscrow is exactly 0.8.4 - Pointers to examples/federation-regional-escrow.ts and research/regional-escrow/DESIGN.md Soft-deprecates the prior "regional-only (TK-906+) is opt-in" paragraph above, now that the mode is fully wired. Local-only commit; pushed to origin at the 0.8.5 release tag per the standing wiki workflow.
docs(federation): document PostgresCoordinator (0.8.4) Adds the PostgresCoordinator backend to the Coordinator-backends table (now shipped 0.8.4) and a Quick start subsection demonstrating its use. Notes the latency / throughput / HA trade-offs vs RedisCoordinator; both implement the identical GlobalCoordinator interface so they are drop-in interchangeable. References the design doc at research/postgres-coordinator/DESIGN.md. Per the wiki-coupling rule (PLAN.md S2), this commit accumulates locally on tk-wiki master and is pushed only at the 0.8.4 release tag time (alongside the npm publish).
docs(federation): refresh 0.9.0 → 0.8.3 release references (DR-07) Federation #77 ships as 0.8.3 patch (not 0.9.0 minor) per the new DR-07 in research/bigger-bets/PLAN.md §8. Surface is purely additive so a patch is the right semver signal. Federation.md: - Coordinator backends table: 'Shipped 0.9.0' → 'Shipped 0.8.3' (×2) - PostgresCoordinator: '0.9.x follow-up' → '0.8.x follow-up' - SPOF caveat paragraph: 'For 0.9.0' → 'For 0.8.3' Home.md: - Guides table Federation row: '(0.9.0)' → '(0.8.3)' No content change beyond the version reference.
docs(federation): new Federation page + sidebar/home/Distributed-and-Provable cross-links (TK-911) The federation user-facing wiki page for the 0.9.0 release. Pushes to origin happen at release tag time per the local-accumulate-then-push rule (memory/throttlekit-wiki.md). New page Federation.md covers: - Quick start (federate + RedisCoordinator example) - The contribution narrative vs static partition (eval table at K=3) - All four outage shapes + Δ bound per shape - Coordinator backends matrix (TestCoordinator / RedisCoordinator shipped; PostgresCoordinator / Raft 0.9.x / 1.0.x follow-ups) - SPOF caveat + Sentinel/Cluster/Postgres/Raft mitigation ladder - Pointers to the TLA+ spec, BFS twin, DESIGN.md, RESULTS.md Cross-links added: - Home.md table: new Federation row - _Sidebar.md: new Federation link - Distributed-and-Provable.md: new 'Cross-cluster federation' sub-section that says when to use federate() vs the existing twoTier(leased) windowCoupled (essentially, when processes span multiple regional clusters)
docs(home): refresh 0.8.1 breadth — quota, peek/forecast, KV, CLI, config, decorator; 186 ns checkSync