MESSAI · Roadmap · v2026-09-06

Twelve months, six pillars, one brewery.

8 Sept 2026 to 27 Aug 2027. Eighteen workstreams, each with dated deliverables. Bold-ringed dots sit on the critical path; dashed dots are scheduled to a phase, not a date. Click any dot for the deliverable.

Prove & prepare · 8 Sept 2026 – 19 Dec 2026

Gate: none — start now · Sam + Claude routines

Fund & partner · 5 Jan 2027 – 30 Apr 2027

Gate: funding event OR 2 paid feasibility studies signed · solo → first hire

Build & validate · 3 May 2027 – 27 Aug 2027

Gate: wet-lab partner signed · 1–3 hires

Villano (ISMET president) meeting · ≈ 15 Oct 2026 P1 gate: funding event or 2 paid studies · 19 Dec 2026

TimelineWhere every milestone lands

A Platform & observability B Science & data C Brewery FOAK D Community & partnerships E Commercial F Company critical path phase-scheduled

The crunch is the five weeks to 10 Oct 2026: A1-M1, A1-M2, A1-M3, A3-M1, B1-M1, B2-M1, B3-M1, C1-M1, C1-M2, C2-M1, D1-M1, D2-M1, D2-M2, E1-M1, F1-M1, F2-M1. A1-M3 and B1-M2/M3 slip first if something has to.

Pillar APlatform & observability

Product analytics + observability

Know how people use the platform. Today: no PostHog/Vercel Analytics, Sentry DSN-gated (dark in practice), SP-7 OTel spec written but zero lines built, feedback widget on apps/web only, survey write-only, onboarding persona never persisted.

Milestone	When	Effort	Deliverable
A1-M1 ●	19 Sept 2026	~2 days	PostHog EU via Vercel Marketplace in web + lab + site; page views + 12 named events (prediction_run, archetype_selected, paper_opened, effect_card_viewed, feedback_sent, survey_completed, onboard_persona_chosen, lab_model_loaded, export_clicked, chat_tool_called, kg_node_clicked, demo_slider_changed); GDPR consent banner.
A1-M2	26 Sept 2026	~1 day	Sentry DSN set on all four Vercel projects; the feedback↔Sentry correlationId join implemented (column already exists on Feedback).
A1-M3	10 Oct 2026	~1 week	SP-7 OTel spans on /api/ml/predict, chat tools, extraction; LangSmith opt-in; TraceBadge. Follow SP-7 plan tasks 4–8; defer Modal/Fly tasks 9–10 to P1.
A1-M4 ●	31 Oct 2026	~2 days	/admin/usage page: funnels (land → predict → export), top surfaces, survey read-out (SurveyResponse has no admin surface today); feedback widget mounted in apps/lab.
A1-M5	P1	~3 days	FeedbackAnalytics daily rollup writer (table exists, no writer); UsageRecord moved from the 24h in-memory LRU to a table.

Depends on: F2-M1
Assets: apps/web/instrumentation.ts apps/web/instrumentation-client.ts apps/web/src/app/_components/FeedbackWidgetClient.tsx docs/superpowers/specs/2026-05-19-sp7-observability-design.md docs/superpowers/plans/2026-05-19-sp7-observability.md
First action: Load the vercel:marketplace skill and run discover for PostHog.

Agentic ops loop

Routines that read telemetry, feedback, and effects and propose ranked changes as PRs. One routine exists today (weekly ML audit).

Milestone	When	Effort	Deliverable
A2-M1 ●	7 Nov 2026	~2 days	Weekly usage-digest routine: reads PostHog API + Feedback table → docs/routines/reports/YYYY-WW-usage.md and opens ≤3 issues labelled from:usage-digest.
A2-M2	28 Nov 2026	~3 days	Dark-surface pruner: pages with 0 views in 4 weeks flagged for _prefix archive (per the sitemap value audit).
A2-M3	P1	~1 week	Improvement-PR routine: for a bounded from:usage-digest issue, drafts the PR on a branch for Sam's review. Never merges.
A2-M4	P1	~2 days	Close the PLATFORM_STATE §7/§8 loop: feedback on a prediction surface joined to the Prediction row; the weekly refit routine consumes it.

Depends on: A1-M1, A1-M4
Assets: scripts/routines/weekly-ml-audit.sh docs/routines/weekly-ml-audit.md docs/PLATFORM_STATE.md
First action: Copy docs/routines/weekly-ml-audit.md as the template for the usage-digest routine.

Reliability + tech-debt burndown

The other streams don't die on a wedged build. Known: type-debt baseline, five dark admin sidebar links, stale computed/calibration.json, no CI.

Milestone	When	Effort	Deliverable
A3-M1	30 Sept 2026	~1 day	Dark admin links (/admin/stats, ml-pipeline, database, users, security) removed or stubbed; /admin/analytics renamed corpus-analytics and linked.
A3-M2	31 Oct 2026	~2 h/week	Type-debt burndown routine live; baseline −25%.
A3-M3	19 Dec 2026	~2 days	pnpm pre-deploy runs in a weekly routine on main; verify:science green.
A3-M4	P1	~1 week	Auth hardening for beta users: rate limits on chat, BotID per TECH_STACK.

Depends on: nothing
Assets: apps/web/src/app/admin/components/AdminSidebar.tsx docs/routines/type-debt-burndown.md scripts/pre-deployment-check.sh
First action: grep -n href apps/web/src/app/admin/components/AdminSidebar.tsx

Pillar BScience & data

Data acquisition + harmonization + analysis

Raise modelable rows and close the corpus gaps the brewery design needs (anode-material coverage at MFC-AD is 0%).

Milestone	When	Effort	Deliverable
B1-M1	26 Sept 2026	~1 day, ~$70–150	BioC-PMC XML run: 1,444 files through the v2 extractor (parser exists at scripts/extraction/lib/pmc-xml-to-text.ts).
B1-M2	17 Oct 2026	~3 days	curl_cffi publisher-403 fix + no-DOI Playwright path; +500 PDFs.
B1-M3	31 Oct 2026	~1 week	NULL canonical_slug 80.7% → ≤70% via alias-map extension (mes-data-harmonization skill).
B1-M4 ●	14 Nov 2026	~2 days	Brewery gap-fill: weekly routine step 5b gap-search targeted at the C1 levers (anode material, HRT, sulfate, temperature); refit effects.json; target ≥1 medium→high effect.
B1-M5	P1	ongoing	Monthly modelable-count report; prod parameterDefinitionId FK under-population (~33k rows) fixed.

Depends on: nothing
Assets: .claude/skills/mes-paper-retrieval/SKILL.md .claude/skills/mes-data-harmonization/SKILL.md scripts/derived docs/keeping-corpus-state-fresh.md
First action: Count the unread BioC-PMC XML files to confirm the 1,444 figure.

Model quality + validation

A predictor that beats its own mean out-of-fold. GP-SCM currently has no predictive power (1 of 12 fitted nodes).

Milestone	When	Effort	Deliverable
B2-M1	10 Oct 2026	~1 week	Held-out benchmark locked (bes-benchmark-v1, paper-disjoint) with a scoreboard at /decks/bes-benchmark.
B2-M2	21 Nov 2026	~2 weeks	Per-class MFC/MEC predictor v3 beats the prior mean on held-out by a stated margin; conformal coverage ≥ nominal.
B2-M3	P1	~1 week	Blind prediction on one partner's measured stream (validation-plan item 4).
B2-M4	P2	—	Brewery bench data (C2) as the first out-of-corpus test.

Depends on: B1-M3
Assets: libs/shared/ml services/ml-engine/training/artifacts/benchmark-v1
First action: Open the held-out benchmark plan and lock the paper-disjoint split.

Publications + reporting standard

Credibility currency for Villano, Horizon, and EXIST reviewers.

Milestone	When	Effort	Deliverable
B3-M1 ●	10 Oct 2026	~3 days	Minimum-reporting checklist v1 (proposal 4C + Korth ÇA GRIMPE framing): one-page markdown + JSON schema. The concrete artefact to hand Villano.
B3-M2	5 Dec 2026	~2 weeks	Preprint: corpus-derived effect sizes for MFC-AD — what 684 papers can and cannot tell you (effects.json 2.0 as the dataset).
B3-M3	P1	~1 week	Submit to a journal or ISMET-adjacent venue; dataset on Zenodo with a DOI.
B3-M4	P2	—	Second paper from bench data, with the wet-lab partner as co-author.

Depends on: nothing
Assets: docs/abstracts docs/ismet-2026-whitepaper
First action: Open proposal 4C and the Korth ÇA GRIMPE abstract note; draft the checklist header.

Pillar CBrewery FOAK

Brewery FOAK design dossier

A component-by-component schematic where every decision cites literature and every unknown maps to an experiment. Design point: 50–500 m³/d mid-size brewery. Today only uncited UI seed defaults and a 10 L pilot spec exist; the TEA at brewery scale is NPV −$92M on stub inputs.

Milestone	When	Effort	Deliverable
C1-M1 ●	26 Sept 2026	~3 days	Design basis: brewery effluent envelope from literature (COD 2–6 g/L, sulfate, alkalinity, temperature, CIP caustic events, batch variability) with n + sources; design flow 50–500 m³/d; incumbent = UASB/EGSB + CHP.
C1-M2 ●	10 Oct 2026	~1 week	Dossier v0.1 (docs/brewery-foak/design-dossier.md): process flow screen → EQ → heat exchange → MFC-AD train → biogas cleanup → polishing; ~20 numbered components each with choice, alternatives, literature anchors, corpus evidence tier, confidence; the scale bridge (kg COD/d → anode m² → modules → trains). Ships with the brewery deck at /decks/brewery-foak.
C1-M3	24 Oct 2026	~3 days	Unknowns register + experiment plan E1–E6 with pass/fail numbers and the effects.json gap each closes.
C1-M4	14 Nov 2026	~1 week	Annotated schematic: numbered-callout SVG/P&ID in-app where each callout links to its dossier row; 3D scene reused, not rebuilt.
C1-M5 ●	5 Dec 2026	~3 days	TEA v1 with uncertainty: cited ranges replace the stub constants (BoP $25M, CE 25%, 600 mW/m², 60% AD removal); FOAK/NOAK split mirroring libs/shared/dac/src/tea/lcoc.ts; side-by-side vs UASB + CHP.
C1-M6	P1	—	Dossier v1.0 reviewed by the wet-lab partner (C2) and one brewery engineer (E1).

Experiments to run first (small scale, ≤ 3 months, ≤ €5k each)

E	Question	Closes	Pass
E1	Real brewery effluent characterisation: COD fractions, sulfate, alkalinity, CIP slugs, weekly variability	design-basis assumptions	envelope within dossier ranges, or dossier revised
E2	Bench MFC-AD on real effluent: CE, power density, COD removal vs the corpus prior	validation item 4; mfc-ad power_density gaps	measured value inside the conformal interval
E3	Anode material contrast (cloth vs brush vs felt) on brewery effluent	anode-material coverage 0% at MFC-AD	≥1 medium/high effect after refit
E4	Sulfate competition + CIP caustic shock recovery	not in the corpus at all	recovery time and CE penalty quantified
E5	HRT sweep 6–48 h	HRT coverage 9.8%	HRT → COD-removal effect at high confidence
E6	90-day fouling / cathode drift	durability gap in research-gaps.json	drift slope with CI

Depends on: nothing
Assets: docs/lab-cad/wastewater-preset-engineering-specs.md docs/dac-measurement-plan.md apps/web/src/app/demo/resource-recovery apps/web/public/data/computed/wastewater apps/web/public/data/computed/research/effects.json libs/shared/component-catalog
First action: mkdir docs/brewery-foak and copy the engineering-specs table header.

Wet-lab partner + de-risking experiments

A signed bench partner and E1 running. Candidates: TU Berlin EVUR (Herkendell/Feilner, doubles as EXIST host), UFZ Leipzig (Korth/Harnisch, live collab page), Sapienza (Villano, via D1).

Milestone	When	Effort	Deliverable
C2-M1 ●	3 Oct 2026	~2 days	One-page bench-partnership proposal: what MESSAI brings (dossier, corpus priors, measurement plan, co-authorship) and what we need (bench, effluent, ~3 months).
C2-M2 ●	31 Oct 2026	—	Proposal sent to 3 candidates; ≥1 meeting held.
C2-M3 ●	19 Dec 2026	—	Partner + effluent source agreed (LOI or email); E1 sample plan.
C2-M4	P1	—	E1 + E2 complete; results in the DB via C3.
C2-M5	P2	—	E3–E6 complete; dossier v1.0; second paper (B3-M4).

Depends on: C1-M2
Assets: apps/web/src/app/collab/korth docs/ismet-2026-whitepaper/networking-plan.md
First action: Draft the one-page bench-partnership proposal from the C1 design basis.

Pilot instrumentation / DAQ

Measurements from a bench or pilot land in the platform with provenance, closing predict → measure → refine.

Milestone	When	Effort	Deliverable
C3-M1	21 Nov 2026	~1 week	Measurement ingest: CSV/JSON upload route in apps/api linked to Run/Experiment; derivationMethod = 'measured' provenance.
C3-M2 ●	19 Dec 2026	~3 days	Potentiostat/logger export adapters (Gamry, BioLogic, Arduino-style CSV); /experiments/[id] shows measured vs predicted with CI.
C3-M3	P1	~1 week	Sensor spec for E1–E6 (pH, ORP, T, conductivity, gas flow, V/I at 1 min) + BOM ≤ €2k.
C3-M4	P2	—	Live pilot dashboard; measured rows feed B2-M4.

Depends on: C2-M3
Assets: apps/web/src/app/experiments apps/web/src/app/runs prisma/schema.prisma
First action: Read the Experiment and Run models; sketch the Measurement ingest contract.

Pillar DCommunity & partnerships

ISMET partnership

ISMET adopts MESSAI as a community tool (education, research, reporting standard) and member research feeds the opportunity areas, starting with the brewery case study. Hard date: Villano meeting mid-October.

Milestone	When	Effort	Deliverable
D1-M1 ●	9 Oct 2026	~1 week	Partnership proposal (docs/ismet-partnership/proposal.md) + 6-slide deck: member access tier; reporting-checklist co-ownership; student/course module; the verify-your-paper and claim-an-ontology-category asks; brewery case study as first community validation target; what ISMET gets (usage stats, citation of members' work, a standards paper).
D1-M2 ●	15 Oct 2026	—	Villano meeting held. Ask: a named ISMET liaison + a slot in the next newsletter or webinar.
D1-M3	28 Nov 2026	~3 days	Education module v1: one guided tutorial on /demo/resource-recovery + /lab with a worked brewery exercise; feedback captured via the widget.
D1-M4	19 Dec 2026	~2 days	ISMET member onboarding path: /onboard persona persisted; source='ismet' on signup.
D1-M5	P1	—	Webinar or student challenge run; ≥20 member accounts; ≥5 papers verified by their authors.
D1-M6	P2	—	Checklist v1 endorsed or co-published with ISMET.

Depends on: B3-M1, A1-M1, C1-M2
Assets: docs/ismet-2026-whitepaper docs/ismet-2026-attendee-survey.md apps/web/src/app/ismet-presentations
First action: Create docs/ismet-partnership/proposal.md with the six-point outline.

Researcher outreach (tester + collaborator pipeline)

Turn the 64-abstract analysis into a tracked pipeline and extend it to testers and collaborators beyond EU-ISMET.

Milestone	When	Effort	Deliverable
D2-M1	19 Sept 2026	~½ day	OutreachContact Prisma table (name, affiliation, country, email + confidence, goal tags validation/EXIST/Horizon/bench/tester, tier, status, lastContactAt, notes; optional FK to SurveyResponse/BetaSignup) + seed script from networking-plan.md (~70 rows). Migration local → staging → prod.
D2-M2	26 Sept 2026	~2 h	Send the 10 Tier-1 emails after verifying the guessed addresses; log sends.
D2-M3	24 Oct 2026	~2 days	Second wave (issue #458): 10 more, plus a tester variant offering beta access + a 20-minute call.
D2-M4	28 Nov 2026	~1 day	Per-partner briefing pages for warm replies, cloned from /collab/korth.
D2-M5	P1	—	20 structured interviews (validation-plan item 1); pipeline reviewed monthly.

Depends on: nothing
Assets: docs/ismet-2026-whitepaper/networking-plan.md docs/ismet-2026-whitepaper/outreach-emails.md apps/web/src/app/collab/korth
First action: Open outreach-emails.md and verify Tier-1 address #1.

Open source + community

The mess-* packages are the thing ISMET members touch first.

Milestone	When	Effort	Deliverable
D3-M1	17 Oct 2026	~2 days	Four packages missing barrels fixed; sync-public-mirrors run; CONTRIBUTING + good-first-issues on Messai-io.
D3-M2	14 Nov 2026	~1 day	Release of mess-parameters carrying the reporting-checklist schema (B3-M1).
D3-M3	P1	—	≥3 external contributors or issues; effects.json published as a versioned dataset.

Depends on: nothing
Assets: open-source docs/releasing.md tools/lint-package-shape.ts
First action: Run pnpm tsx tools/lint-package-shape.ts and list the packages missing a barrel.

Pillar ECommercial

Go-to-market + first customer

A named brewery (Berlin region first), site effluent data, and a paid feasibility study.

Milestone	When	Effort	Deliverable
E1-M1 ●	3 Oct 2026	~2 days	Target list: 15 breweries (Berlin/Brandenburg craft + one regional) with discharge context; feasibility-study one-pager ($25–50K).
E1-M2	14 Nov 2026	—	5 conversations; 1 effluent sample commitment (feeds C1 experiment E1).
E1-M3 ●	19 Dec 2026	—	1 LOI or signed study.
E1-M4	P1	—	2 studies signed, 1 delivered (validation-plan item 2).

Depends on: C1-M2
Assets: docs/pitch/2026-09-02-dac-w2v-validation-plan.md apps/web/src/app/solutions/companies/page.tsx
First action: List 15 Berlin/Brandenburg breweries in the 50–500 m³/d band.

DAC track

Keep the DAC model honest and partner-calibrated in parallel; the second FOAK opportunity area.

Milestone	When	Effort	Deliverable
E2-M1	31 Oct 2026	—	Membrane volume quote + one bench V–I sweep partner (validation item 3).
E2-M2	19 Dec 2026	—	LCOC interval ±129% → under ±50% if item 3 lands; otherwise documented why not.
E2-M3	P1	—	Feasibility study with one DAC developer.

Depends on: nothing
Assets: docs/dac-measurement-plan.md docs/dac-evidence-gaps.md libs/shared/dac apps/lab/src/app/lab/dac
First action: Shortlist three membrane suppliers for a volume quote.

Opportunity portfolio

A ranked list of verticals and the rule for opening a second case study.

Milestone	When	Effort	Deliverable
E3-M1	7 Nov 2026	~2 days	docs/roadmap/opportunity-portfolio.md: brewery, dairy, municipal, mine-methane biofilter, DAC, ectoine/PHB high-value — scored on corpus depth, incumbent cost, customer access, partner fit.
E3-M2	P1	—	Rule adopted: open vertical #2 only when the brewery E2 result is in and ≥1 paid study is delivered.

Depends on: nothing
Assets: docs/vertical-prioritization-2026-05-25.md docs/superpowers/specs/2026-08-01-mine-methane-biofilter-design.md
First action: Create docs/roadmap/opportunity-portfolio.md with the scoring columns.

Pillar FCompany

Funding + runway

The P1 gate. Four instruments in play: EXIST, Horizon Europe, pre-seed/angel, paid feasibility studies.

Milestone	When	Effort	Deliverable
F1-M1	19 Sept 2026	~1 day	Funding calendar: EXIST cut-off (to confirm), Horizon calls relevant to MES, angel list, feasibility-study pipeline; runway in months.
F1-M2 ●	31 Oct 2026	~1 week	EXIST application draft with TU Berlin as host (needs the C2-M2 conversation); mentor = Herkendell or Rosenbaum.
F1-M3	28 Nov 2026	—	Pre-seed: investor overview updated with P0 status (validation items 1–2 numbers); 10 conversations.
F1-M4 ●	19 Dec 2026	—	Gate check: funding event OR 2 paid studies → P1 runs funded; otherwise P1 runs the solo variant (A/B/D streams only, C2 slips).
F1-M5	P1	—	Horizon consortium: coordinator identified (Wageningen, Toulouse, or Ghent); MESSAI as SME partner in one proposal.

Depends on: C2-M2
Assets: docs/ismet-2026-whitepaper/networking-plan.md apps/web/src/app/investors docs/grants/foresight-milestones.md
First action: Confirm the EXIST instrument and cut-off date.

Legal / IP / data licensing

Unblock analytics and partnerships: privacy, corpus posture, contracts.

Milestone	When	Effort	Deliverable
F2-M1 ●	19 Sept 2026	~1 day	Privacy policy + cookie consent (PostHog EU, Sentry). Unblocks A1.
F2-M2	17 Oct 2026	~2 days	Corpus posture doc: PDFs private cache; extracted values + effects published CC-BY; OSS licenses audited.
F2-M3	28 Nov 2026	~1 day	Beta terms of use; feasibility-study contract template; data-sharing agreement template for bench partners.
F2-M4	P1	—	IP posture: what is defensible (calibrated priors + measurement plans), what is open by design.

Depends on: nothing
Assets: apps/site/src/pages
First action: Draft the privacy policy page and consent copy for PostHog EU + Sentry.

Team + hiring

Post-funding roles, in order, with owners for the C and B pillars.

Milestone	When	Effort	Deliverable
F3-M1	5 Dec 2026	~1 day	Post-funding hiring plan: role 1 electrochemist/process engineer (owns C1–C3); role 2 data/ML engineer (owns B1–B2); role 3 optional community/BD.
F3-M2	P1	—	Role 1 hired within 8 weeks of the gate; onboarding = AGENT_HANDOFF + this roadmap.

Depends on: F1-M4
Assets: docs/AGENT_HANDOFF.md
First action: Write the role-1 (electrochemist / process engineer) one-page brief.

DecisionsSettled and still open

Decided

2026-09-06 — Brewery FOAK design point = 50–500 m³/d mid-size brewery; the 5,000 m³/d demo scale stays a labelled scale-up scenario.
2026-09-06 — Outreach pipeline = OutreachContact Prisma table, not CSV.
2026-09-06 — Analytics stack = PostHog EU + Sentry live + OTel per SP-7.
2026-09-06 — DAC is its own workstream (E2), not subordinate to brewery.

Open

EXIST instrument + cut-off date (Gründerstipendium vs Forschungstransfer).
Villano meeting ask: named liaison + webinar slot (recommended) vs a formal MoU.
Which Sentry org / PostHog project owns the data (personal vs company account).

RhythmThe P0 week

Mon — usage digest (A2-M1 once live) + weekly ML-audit routine report.
Tue–Thu — one Brewery-pillar block, one Platform/Science block.
Fri — outreach sends + pipeline update (D2); one status line per workstream.

Generated from docs/roadmap/roadmap.json v2026-09-06 by scripts/roadmap/build-roadmap-html.ts. Markdown twin: docs/roadmap/2026-09-roadmap.md. GitHub milestones P0/P1/P2 and ws:* labels mirror this file. If a date here disagrees with the JSON, the JSON wins.