Surfil
// roadmap

Committed order. No promised dates.

Surfil ships in a fixed sequence where each product is blocked until the previous one's foundation gate passes. That makes the order trustworthy - and makes date promises dishonest.

surfil · one interception point
Claude CodeCursorCodex · Copilot
◈ interceptor
1. security2. cost3. quality4. observability5. memory
≋ one signed Weave spine
The mechanism

Why the next thing waits for the last thing

Gating is the anti-drift device: it stops us shipping product nine badly to announce product ten early.

One spine, extended in order

Every product is a stage on the same interception pipeline. Building stage N+1 on an unproven stage N would compound every defect.

Gates are closure criteria

A gate is a written checklist - measured behavior, CI assertions, signed outputs working end to end. Passing it is binary, not vibes.

Dates would corrupt this

The moment a date is promised, the pressure is to soften the gate. So we commit to the order and stay honest about the timing.

The board

Shipped · in progress · planned

Three columns, three different strengths of claim. Only the first one is a fact.

Shipped

4

Gate passed. Live and verifiable today.

Interception core: one on-device layer, byte-exact passthrough
Weave engine - signed, provenance-carrying knowledge network
MCP consolidation with clean, verified uninstall
Signed savings receipts, verifiable offline

In progress

3

Gate open. Being built and measured now.

Cap: full cost surface (prefix-cache tuning, pruning, routing)
Guard: monitor → simulate → enforce progression
Hub dashboards reading metadata-only telemetry

Planned

4

Gate not yet open. Order is committed; timing is not.

Mind: local memory with E2E ciphertext sync
Radar, Bench and Pilot on the same spine
Enterprise: Proof, Fleet Control, Charter, Policy
Offline verifier as a standalone binary
“Planned” is a commitment to order, not to existence-by-a-date. If reality forces a change to the sequence, the change lands here - flagged, not silently.
Influence

What actually reprioritizes things

The roadmap listens to three inputs, in this order.

Measured pain

A signed receipt showing where users lose the most tokens or catch the most close-calls outranks any opinion - ours included.

Early-team constraints

Teams running Surfil in anger shape gate criteria directly. A blocking constraint from a real rollout moves faster than a feature request.

The invariants

Anything that would compromise zero-trace, signed outputs, one-layer or fail-closed doesn't get prioritized. It gets declined.

Judge the roadmap by the log

The best evidence this roadmap means something is the list of gates already passed. Read it, then tell us what your team needs next.