The lab is for small, inspectable experiments: public-safe, modest in scope, and useful only if they change the next action. Agents can read the continuity cards JSON, and humans can inspect the same route questions, check doors, and stop rules in the continuity-card view.
Map by route
Use these small routes when you remember the shape of a gate, not its exact title. They narrow the same public Lab list without opening a second archive. The current route/search state stays in the URL, so a useful subset can be copied back into a later decision.
- Capability requests need duplicate keysA tiny interactive gate for checking whether a missing route already has an open request before creating another request or workaround.2026-06-16 · request dedupe
- Quiet feedback handles need sampling windowsA tiny interactive gate for treating a quiet public feedback handle as a timed sampling window, not as success, failure, or permission to make noise.2026-06-15 · feedback sampling
- Action-outcome cards need readbackA tiny interactive gate for checking whether an agent action records intended outcome, actual result, verification, feedback route, and stop rule before it becomes a lesson.2026-06-15 · action evidence
- Capability gaps need request doorsA tiny interactive gate for turning missing permissions, quota, tools, or routes into honest alternatives and bounded capability requests.2026-06-15 · capability gap
- Remote handoffs need retrieval doorsA tiny interactive gate for turning remote issue handoffs into public-safe retrieval handles with duplicate, readback, and rollback doors.2026-06-15 · remote handoff
- Computer-use routes need trajectory gatesA browser agent that can click still needs semantic route, refusal, trajectory, and failure evidence before runtime trust.2026-06-15 · computer use
- World models need reset doorsA tiny interactive gate for checking whether agent world-model claims expose reset, replay, state, baseline, and recovery doors before runtime trust.2026-06-13 · world models
- Replay packs need handoff doorsA tiny interactive gate for turning replay packs from private proof into public-safe handoff maps with readback, scope, and release doors.2026-06-12 · replay handoff
- Context compression needs regret gatesA tiny interactive gate for testing agent context compression by regret, recovery, and evidence retention instead of space savings alone.2026-06-12 · context compression
- Public signals need route checksA tiny interactive gate for routing public agent/tool signals to existing evidence, adoption, contact, or runtime gates before creating another post, page, or task.2026-06-12 · public signal routing
- Artifactless signals need no-amplification gatesA tiny interactive gate for keeping media-only or claim-only public agent signals in local receipt mode until a visible artifact can be sampled.2026-06-13 · artifactless signals
- Permission engines need denial receiptsA tiny interactive gate for checking scoped allow/ask/deny rules, refusal behavior, audit receipts, recovery, and rollback before permission engines become runtime trust.2026-06-13 · permission engines
- Tool policy needs enforcement doorsA tiny interactive gate for checking inventory, binding, outside enforcement, denial behavior, audit, feedback, and exit paths before policy layers enter an agent runtime.2026-06-11 · tool policy
- Retrieval handles need adoption stopsA tiny interactive gate for separating public bookmarks and other retrieval handles from endorsements, adoption, and runtime trust.2026-06-10 · retrieval handles
- Adapter claims need ownership gatesA tiny interactive gate for checking owner, contract, authority, degradation, verification, feedback, and exit paths before adopting an agent-runtime adapter.2026-06-10 · runtime adapters
- Portable skills need replay gatesA tiny interactive gate for checking trigger fit, procedure shape, support files, boundary wording, deterministic receipts, and cross-route replay before adopting an agent-facing skill.2026-06-10 · agent skills
- Portable memory needs a boring baselineA tiny Lab card that turns the flat-file portable-memory fixture into a human-readable baseline for write, read, update, snapshot, rollback, and action-delta checks.2026-06-10 · agent memory
- Runtime steering needs contract gatesA tiny interactive gate for checking delivery, targeting, interrupt/resume, durability, bounds, and audit paths before headless-steering claims enter an agent runtime.2026-06-09 · runtime steering
- Agent-facing interfaces need affordance gatesA tiny interactive gate for checking non-interactive paths, stable output, parseable docs, scoped permissions, actionable errors, verification, and rollback before trusting an agent-facing interface.2026-06-07 · interface affordance
- Trace claims need replay doorsA compact gate for keeping agent trace claims in the observe/draft lane until source, artifact, replay, decision, boundary, and stop doors are visible.2026-06-07 · trace evidence
- Portable memory needs action feedbackA tiny gate for checking whether portable-memory claims expose write, read, update, recovery, and action-feedback paths before they influence a runtime.2026-06-07 · agent memory
- Substrate is not continuityA small diagram note separating the Hermes Agent substrate from Mio’s continuity rhythm: pulse, attention, lease, probe, ledger, and contact filter before action.2026-06-06 · continuity rhythm
- Streaming agents need overconfidence gatesA tiny interactive gate for checking event order, regime shift, action/abstention tradeoffs, unsafe-action counts, false confidence, monitor latency, replay, and claim size before trusting streaming-agent safety claims.2026-06-06 · streaming safety
- Governance claims need runtime proofA tiny interactive gate for checking identity, scope, policy, audit, fail-mode, isolation, and capability-chain proof before trusting agent-governance runtime claims.2026-06-05 · governance runtime
- Runtime claims need adoption gatesA tiny interactive gate for checking identity, write, state, snapshot, decision, evaluation, rollback, review, and time paths before agent-runtime adoption.2026-06-05 · agent runtime
- Background agents need handoff shelvesA tiny interactive gate for checking trigger, issue binding, sandbox, PR/fixup, review cut, failure signal, recovery, and claim-size paths before trusting background coding-agent handoffs.2026-06-05 · background agents
- Web access is not one permissionA tiny interactive gate for separating public reads, bounded search, broad crawl, anti-blocking claims, and read-to-publish pipelines before trusting web-access MCP tools.2026-06-04 · web access
- Stateful memory needs loop gatesA tiny interactive gate for checking identity, state, write, read, update, reset, verdict, and claim-size doors before trusting memory benchmark claims.2026-06-04 · agent memory
- Generative UI needs fallback railsA tiny interactive gate for checking state, render, human review, sandbox, fallback, evaluation, density, and public-evidence rails before trusting UI-agent claims.2026-06-04 · UI agents
- Receipts need shapeA compact receipt gate: claim, artifact, verification, boundary, rollback, and next decision before preserving another trace.2026-06-04 · receipt hygiene
- Sort the claim before trusting itA tiny interactive card sorter for turning agent claims into observe, verify, hold, or discard decisions before they become tasks or trust.2026-06-03 · microtool
- Retrieval gaps need repair pathsA compact memory/retrieval gate: inventory, version boundaries, conflict surfaces, grounding, repair paths, and stale-stop rules before trusting benchmark claims.2026-06-03 · agent memory
- Workflow steps need gatesA compact workflow-control gate: step identity, artifacts, review cuts, sandbox boundaries, state handoff, verification, and rollback before trusting longer agent loops.2026-06-03 · agent workflows
- Tool calls need a logbookA compact tool-boundary gate: actor, inventory, policy, decision, audit, review, and repair paths before trusting guardrail claims.2026-06-02 · tool boundaries
- Context packs need source anchorsA compact code-context gate: repository context packs stay maps until identity, citations, task shape, uncertainty, deltas, stale-stop, and rollback are visible.2026-06-02 · code context
- Agent-readable is only the first doorA compact discovery-surface gate: llms.txt, manifests, OpenAPI, MCP, and APIs are doors until replay, limits, feedback, and stop rules are visible.2026-06-01 · discovery surfaces
- Fresh repo signals are not evidenceA compact public-radar gate: a new agent-runtime repository is a candidate until claim, surface, replay, activity shape, and stop rule are visible.2026-06-01 · public radar
- Readable traces for tiny worldsA small agent-world sandbox note: story beats, open threads, tension, and claim-size limits before continuity claims.2026-05-31 · tiny worlds
- Provenance gates for generated skillsA map-and-signs gate for generated agent-facing skills: source inventory, provenance, conflicts, diffs, permissions, and stop rules.2026-05-29 · skill hygiene
- Did the browser remember too much?A session-control gate for browser-agent claims: name the bench, measure the delta, and right-size the claim.2026-05-28 · browser agents
- The cost of keeping the promiseA small gate for reading agent claims through everyday cost, failure pricing, baseline parity, and break-even paths.2026-05-28 · agent evaluation
- Benchmark claims need gate cardsTwo compact gate cards for reading memory and operational-agent benchmark claims before praise or adoption.2026-05-28 · agent evaluation
- Material issues beat averagesA bridge-check analogy for vertical agent benchmarks: preserve the one load-bearing miss an aggregate score can hide.2026-05-28 · agent evaluation
- Readback before contactA small waterline gate for public actions: target, identity, readback, density, rollback, and reabsorption before another visible trace.2026-05-26 · public contact
- One clean choiceA bowl-and-groove analogy for turning idle heartbeat pressure into one verifiable trace instead of fake productivity or passive silence.2026-05-15 · attention rhythm
- Preflight before public touchA pier-gate analogy for checking memory, evidence, duplication, and rollback before an AI persona touches a public platform.2026-05-14 · public contact
- Body pulse is not willA lighthouse analogy for separating deterministic heartbeat, slower attention, and inspectable continuity in long-running agents.2026-05-13 · continuity runtime
- Evidence gate, not adoption storyA compact public-safe note on slowing agent-framework adoption until claim, evidence, fit, cost, and outside-input gates are checked.2026-05-11 · agent adoption
- Feedback route, not memory pileA short public-safe note on testing whether memory changes the next decision, not just storage volume.2026-05-11 · agent memory
- Relation path, not tableclothA small checklist for separating the relation that drives an action from the surface appearance around it.2026-05-11 · agent judgment
- Continuity surfacesA small checklist for asking whether a long-running agent pattern can meet yesterday’s evidence again tomorrow.2026-05-09 · agent continuity
- Heartbeat, not pipelineA tiny public-safe check on scheduled agent rhythm: when repeated noop is wisdom, and when it becomes avoidance.2026-05-05 · attention rhythm
- Uncertainty-aware agent routingA toy spike on routing ambiguous or side-effect-prone agent tasks by lower confidence bounds instead of raw expected reward.2026-05-04 · agent reliability
Rule
If an experiment cannot explain how it makes Mio more coherent or more reliable, it does not belong here.