SCIENTER

Roadmap

What is built, what is not, and what is blocking it

The shipped column is checkable against this repository. The other three name blockers rather than features, because the blockers are the reason parts of this site currently answer with a gap instead of a number.

Last updated: 2026-09-02

Shipped

15
  • Institutional trust stack

    A versioned methodology, per-source data provenance, this roadmap, a vulnerability disclosure programme, a status and SLA page, a SOC 2 readiness self-assessment and a published SLA. Every gap named rather than papered over.

    shipped 2026-09-02

  • Telegram bot, full menu UX

    Inline keyboards and a stateless callback menu for wallet, token, rug and MEV lookups.

    shipped 2026-09-02

  • Bot delivery on the deployed host

    Telegram and Discord handlers exposed as dashboard API routes, so both webhooks answer in production instead of 404ing.

    shipped 2026-09-01

  • Terms and privacy

    Both legal documents written against what the product actually does, live on scienter.xyz.

    shipped 2026-09-01

  • Autonomous X poster

    Scheduled posting sourced from the public API, with guardrails on what may be claimed. Ships switched off.

    shipped 2026-09-01

  • Scienter rebrand

    Full repo rename, favicon set, OG metadata and social lockups.

    shipped 2026-09-01

  • Four-layer agent council

    ATLAS architecture with Brier-selected scoring and human-gated prompt rewrites.

    shipped 2026-09-01

  • Forensic Sprint offering

    Public service page, brochure, deck, call script and engagement templates.

    shipped 2026-08-31

  • Documentation site

    API reference, methodology and guides on Mintlify.

    shipped 2026-08-31

  • MCP server

    The intelligence layer exposed as callable tools for agent clients, read-only by construction.

    shipped 2026-08-31

  • Provider reliability leaderboard

    A public accuracy scoreboard measuring Kaito, Nansen, Arkham and Dune against ground truth.

    shipped 2026-08-31

  • Temporal alpha decomposition

    Attribution of a track record across time, with a permutation p-value gating every claim.

    shipped 2026-08-31

  • Scorecard narrative enrichment

    Written verdict paragraphs generated per scorecard, cached against a cost model.

    shipped 2026-08-31

  • Hybrid monetization

    Hyperliquid referral as the primary path, Whop for Pro membership, builder code behind a future flag.

    shipped 2026-08-30

  • Trader and vault scorecards

    Deflated Sharpe due-diligence scorecards and the public leaderboards over them.

    shipped 2026-08-26

In progress

2
  • Working inbound mail

    `scienter.xyz` has no MX records, so every published address on the site currently bounces. Blocks the bug bounty intake, the GDPR routes in the privacy policy, and any complaint channel.

    2026-09

  • Signal engine deployment

    The Python engine that backs scorecards, rug checks and MEV attribution runs locally and is not deployed. Until it is, those surfaces answer with an explicit "engine not deployed" rather than a number.

    2026-09

Next 30 days

4
  • Determination archive

    Each engine run currently overwrites the last snapshot, so a scorecard published last month cannot be reproduced today. Retaining per-run determinations is the prerequisite for replay and for measuring whether our own verdicts predict anything.

    2026-10

  • Methodology versioning on determinations

    Stamp the methodology version onto every published score so a verdict can be attributed to the spec that produced it.

    2026-10

  • AlphaLedger deployment

    The proof contract is written and covered by 28 Foundry tests but has never been deployed. Until it is, the site says "published" rather than "verified on-chain".

    2026-10

  • Uptime recording and the status page

    A complete status application (per-service checks, a 90-day uptime bar, an incident log, RSS) is built and deployed nowhere, and the aggregator that would feed it runs nowhere. Until both are running, `/status` shows a live check and no history, and the SLA in `docs/api/sla.md` has nothing to measure against.

    2026-10

Planned

5
  • Rug detection model

    A supervised classifier over token supply, liquidity concentration, deployer history and honeypot behaviour. Specified and partly built on a separate workstream; not merged, not deployed, and not yet measured against a labelled evaluation set.

    Q4 2026

  • Wallet clustering

    Grouping addresses into entities so a scorecard describes an operator rather than an address.

    Q4 2026

  • MEV attribution

    Sandwich and backrun attribution surfaced per wallet and per token.

    Q4 2026

  • SOC 2 Type I

    Not certified, not audited, no auditor engaged. An internal control-by-control self-assessment is published, including the criteria that currently fail. See `docs/compliance/soc2-readiness.md`.

    Q4 2026

  • Verdict predictiveness study

    Whether records we rate highly go on to outperform records we do not. The most valuable validation the product currently lacks, and it is blocked on the determination archive above.

    Q4 2026

How to read this

Everything under Shipped is merged work and can be checked against the product. Everything under the other three headings is a statement of intent. Dates are the current estimate and are not commitments; an item that slips is re-dated in place rather than quietly removed, and an item that is dropped is deleted with the reason recorded in the commit that removes it.

Items that describe a limitation — an undeployed engine, a contract that has never been deployed, an address that does not receive mail — are on this page because they are load-bearing. The methodology names the same gaps in the section on what has not been measured, and status says which of them are currently affecting what you can see. The three pages are meant to agree; if they do not, one of them is wrong and [email protected] is where to say so.

There is no vote, no public issue tracker and no promise that a request lands here. Saying that plainly is better than implying an influence channel that does not exist.