Memory · Vendor-neutral · Closed beta — May 2026

The memory layer
for experimentation.

Most teams ship more experiments than they can remember. Prior connects what you already test, document, and decide — and turns it into a searchable, durable record the whole org can reuse.

Join beta Contact us No credit card · we read every reply
prior · search EXP-INDEX
Checkout Q1–Q4 · 2024–25 All squads — matching
EXP-00417 Single-step checkout · new users· EPPO shipped +2.10%
EXP-00298 Collapse address into card form· NOTION held −0.42%
EXP-00351 Saved-card autofill on landing· JIRA flat +0.04%
3 prior tests · 1 winner · 1 hold · 1 flat ⌘K to search
01 · The problem

What you ship
and what you keep.

37%

Hypotheses
disappear.

Writeups live in someone’s docs, someone else’s tickets, someone else’s deck. Nobody owns the index.

Fragmented
2.4×

Decisions lose
context.

Why did we ship the v3 onboarding? The answer is in a Slack thread from eight months ago — if at all.

Untraceable

Failed ideas
get retried.

Negative results are the most expensive thing to rediscover, and the easiest to forget you ever produced.

Rediscovered
02 · How it works

Three layers,
no new workflow.

01 · INGEST
Connect what you already have.

Experimentation platforms, warehouses, docs, tickets, repos. Prior reads from the systems your teams already write into — no new editor, no new tags, no new meeting.

02 · CANONICALIZE
Fragments become experiments.

An assignment, a metric run, a Notion writeup, a launch ticket — joined into one canonical experiment with hypothesis, variants, outcome, and decision intact.

03 · SYNTHESIZE
Search before you build.

Type a hypothesis, get the prior tests, decisions, and outcomes that already explored it. Duplicate detection, related work, and reusable evidence — surfaced before launch.

03 · The product

Every experiment,
one record.

01 AI-generated TL;DR

Every experiment gets a one-sentence summary with key metrics, confidence intervals, and the final decision.

02 Prior finds related work

Before you launch, see what was already tested. Hypotheses, outcomes, and decisions — surfaced automatically.

03 Metrics with full context

Primary, secondary, and guardrail metrics — with confidence intervals, significance, and inline sparklines.

04 Linked artifacts

Docs, tickets, PRs, flags — every artifact connected to the experiment, matched manually or by AI.

prior. Checkout / Experiments / EXP-00417 ⌘K

Checkout — simplified address form (v3)

checkout_simplified_address_v3 Copy
Concluded · Shipped SStatsig · 4m ago MCMaya Chen · Product, Growth checkout forms mobile-web
● Shipped AI-summarised

Treatment B increased mobile checkout completion by ↑ +2.10% [+0.90%, +3.30%]; guardrails held; shipped to 100% on 14 March 2026.

Primary+2.10% Secondary · TTC−18s GuardrailHeld n184,302 Duration22 days
A Hypothesis
Reducing the address form from 7 fields to 4 will increase checkout completion rate on mobile by ≥1.5%, without harming order accuracy.
C Metrics & results
Metric Treatment B 95% CI Significance
Checkout completion rate checkout.complete / checkout.start +2.10% [+0.90%, +3.30%] Sig +
Time to checkout p50 seconds, mobile ↓ −18s [−24s, −12s] Sig +
Order accuracy rate guardrail · orders.accurate / orders.total −0.10% [−0.30%, +0.10%] Held
E Linked artifacts
N Checkout · simplified address form — design doc edited 11 Feb manual
G Mobile checkout — autocomplete vendor evaluation edited 6 Jan ● matched
J CHK-2841 — implement compact address form on mobile web closed · 24 Feb manual
PR #4127 — checkout: simplified address layout (mobile) merged · 26 Feb manual
S checkout_address_form · gate 100% · stable_id ● matched
+6 more linked artifacts
04 · Why now

Three forces,
one window.

Volume
10×/ year

AI collapsed the cost of shipping change.

Variants are cheap to produce. Validating them and remembering what we’ve tried is now the binding constraint.

Stack
4–7tools

Experimentation knowledge is scattered.

A typical org runs across multiple assignment, analytics, and documentation systems. None of them owns the long-term record.

Synthesis
2026

Models can finally read what teams write.

Hypotheses, decisions, and outcomes can be extracted across messy artifacts with quality that wasn’t possible two years ago.

05 · Principles

What we
won’t compromise.

P / 01 You own the knowledge.

Experimentation history belongs to your org. Portable, inspectable, exportable — always.

P / 02 Every experiment becomes durable.

Not a transient dashboard. A structured, reusable artifact that outlives the tool that produced it.

P / 03 Search comes before charts.

The first interface to your past is retrieval — what happened, where, and why.

P / 04 Negative results are first-class.

Preventing rediscovery of a failed idea is worth as much as preserving a winner.

P / 05 Decisions are first-class.

An experiment without a decision is an unfinished thought. Prior captures both, with lineage.

P / 06 Local value before global value.

Useful to one team on day one. Org-wide network effects are a byproduct, not a precondition.

P / 07 AI augments, never narrates.

Models surface prior evidence. They don’t fabricate new certainty.

Closed beta · ~12 design partners

Be one of the
first orgs to remember.

We’re working with a small number of teams who run real experimentation programs. If that’s you, leave an email — or just write to us. We read every reply, and the feedback shapes what we ship next.

Contact us No spam · one founder reply within 48h
12 / 30 design-partner slots hello@prior.tools