SISO / Action Model / research pack
evidence-led · v1
Action Model / consolidated research pack

From scattered runs to a decision-ready builder map.

This is the working surface for the platform dossiers, OSS discovery, vertical atoms, private builder research, and reusable-block gates. It keeps the original evidence intact while making the through-line easy to inspect.

Browse the eight tracks ↓ Open platform deep dive ↗ Open full dossier ↗
Current posture
research complete / build gated
0
blocks admitted to build
8tracks collated
389merged OSS candidates
115bounded reviews
17 / 12industries / teams
66 / 72use cases / ideas
3vertical pilots scored
01 / The synthesis

What the research actually says

Read this before diving into vendor names.

The central thesis

The opportunity is not another prompt-to-app clone. It is a governed loop that turns internet code and agent actions into a safe, evidence-backed client outcome.

Commercial builders have made generation fast. The missing layer is controlled composition: provenance, license fit, action authority, recovery, review, and a repeatable path from a semi-custom brief to a deployable vertical system.

Decision posture

  • Keep the platform map broad; use it to define capability boundaries.
  • Run OSS discovery per vertical and reusable capability, not as one undifferentiated pile.
  • Treat every block as an admitted artifact with receipts, not a promising GitHub URL.
  • Do not call a build authorized until the synthesis gates close.
02 / Agent work, preserved

Eight research tracks, one operating picture

Showing 8 tracks

AM / P-001 → P-014complete · gated

Platform capability map

Original task
Map platforms doing something like Lovable: their first-party dossiers, capabilities, control surfaces, and failure modes.
Completed
Covered Manus, Airtable Omni, Zapier Forms, Base44, Onlook, NocoBase, CopilotKit, v0/Lovable API questions, and local reconciliations across the builder field.
Learned
Generation is increasingly commoditized; durable differentiation sits in governed actions, recovery, evidence, and vertical context.
Open gate
No build authorization. Validate current APIs and private-builder claims before implementation.
AM / corpusbounded review

OSS GitHub discovery

Original task
Find code that can become a building block, starting from the merged GitHub candidate corpus.
Completed
389 candidates merged; 115 bounded reviews; clusters 1–6 reviewed; cluster 7 has a plan and preflight artifacts.
Learned
The candidate pool is a funnel, not a parts bin: license, provenance, build, static analysis, data boundaries, and UI quality all matter.
Open gate
114 held, 1 rejected, 0 admitted blocks. Repeat the sweep per vertical/capability to expand coverage.
AM / verticalsmodelled

Vertical atomization

Original task
Break industries, departments, use cases, and ideas down to reusable atoms and identify where value can land first.
Completed
17 industries, 12 teams, 66 use cases, 72 ideas, 12 atoms, 62 edges, 3 pilots, and 10 offers checkpointed.
Learned
Finance, CRM, and operations are practical starting pilots because their workflows expose evidence, approval, and handoff requirements.
Open gate
Pilot scores are directional; they are not customer validation or permission to build.
AM / synthesispartial

Action Model synthesis

Original task
Collate the platform, corpus, and vertical work into a repeatable decision system.
Completed
Decision ledger, queues, checkpointed evidence, and a shared architecture direction are present.
Learned
The system needs an Actionist contract: authority boundaries, review checkpoints, recovery, and receipts as first-class outputs.
Open gate
Ledger remains partial: 4 decisions resolved, 1 open; stale metadata must be reconciled before a build call.
BR / ecosystemmapped

Private builder ecosystem

Original task
Deep-dive commercial and private builders that turn prompts, code, or workflows into working software.
Completed
Commercial matrix, coverage ledger, OSS map, and provisional pilot stack assembled across Lovable, v0, Bolt, Replit, Base44, Manus, Cursor, Claude Code, Bubble, Retool, and more.
Learned
“Prompt in, app out” is crowded. The open wedge is trusted composition for a specific client job.
Open gate
The matrix is research, not a current product benchmark; refresh pricing, APIs, and export behavior before selection.
BR / blocksframework ready

Reusable-block admission

Original task
Define how internet code becomes a safe, repeatable building block instead of a copied dependency.
Completed
Manifest-backed framework with provenance, license, SBOM, static/AST, data, UI, build, visual, and human gates.
Learned
The smallest credible first pilot is a narrow read-only permissive UI/CRUD block with a complete evidence trail.
Open gate
No block has passed every gate. Keep the admission state explicit.
BR / catalogueshortlisted

SaaS catalogue + vertical dossiers

Original task
Build a local estate and public map of SaaS substrates, then test their fit for finance, CRM, and revenue operations.
Completed
17 local systems and 34 public candidates catalogued; finance matrix and CRM/RevOps dossiers produced; shortlists include Chatwoot, Payload, Activepieces, Atomic CRM, Frappe, Cal.diy, and Hi.Events.
Learned
Existing systems are useful control references, but the client ledger and Actionist evidence layer remain authoritative.
Open gate
No candidate admitted and no build recommendation made.
BR / synthesisdecision support

Cross-pack synthesis

Original task
Turn the builder reports into a prioritized, falsifiable backlog and identify what still has zero evidence.
Completed
Eight irreducible problems, coverage matrix, and P0/P1/P2 backlog drafted; the pack makes missing scan/build/eval/API receipts visible.
Learned
P0 is not “build a clone”: prove model evaluation, license execution, and the Actionist contract first.
Open gate
The synthesis contains stale catalogue language and no admitted block; reconcile before presenting it as final.
03 / Capability map

What the builder platforms cover

Capability families, not a leaderboard.

Platform familyWhat the research examinedWhere the capability is strongActionist implication
Lovable / v0 / BoltPrompt-to-app generation, frontend composition, export and iteration loops.speed UI composition feedbackDifferentiate above generation.
Replit / Base44 / ManusAgentic build environments, hosted execution, app scaffolding, and task completion.execution hosting agent loopStudy authority + recovery.
Airtable Omni / Zapier FormsStructured data, forms, workflow primitives, and business-facing automation.workflow data ops UXVertical context is the wedge.
Onlook / NocoBase / CopilotKitVisual editing, low-code internal systems, and embeddable agent/product primitives.control surface extensibility internal toolsCompose governed blocks.
Cursor / Claude Code / Bubble / RetoolCode-native agents, visual builders, internal tools, and workflow surfaces.breadth developer leverage systemsOwn the outcome layer.
04 / Open-source funnel

389 is a starting corpus, not the universe

The next sweep should be vertical-aware.

Merged candidate pool
389
Bounded reviews
115
Held for more evidence
114
Rejected
1
Admitted blocks
0
05 / Industry model

A common spine for every industry

Industry → department → job → atom → offer.

taxonomy17

Industries

Source print for the vertical sweep.

teams12

Departments

Where the work actually sits.

jobs66

Use cases

Concrete automation surfaces.

offers72

Ideas

Potential wedges to test.

Directional pilot scores

Where to pressure-test first

These are internal prioritization signals from the vertical run, not market proof.

Finance · 28 / 35CRM · 28 / 35Operations · 27 / 35
Read vertical checkpoint ↗
06 / Composition library

The reusable-block thesis

Use existing code without inheriting invisible risk.

Shortlist from the local estate + public map

The catalogue is a research shelf, not an approved dependency list.

  • ChatwootSupport and inbox substrate.
  • listmonkCampaign and messaging substrate.
  • AFFiNE / SISO DocsKnowledge/workspace substrate.
  • Teable / BaserowStructured data and CRUD substrate.
  • PostizPublishing workflow substrate.
  • Hi.Events / Cal.diyScheduling and event substrate.
  • Payload / FrappeCRM/ERP and content substrate.
  • Activepieces / Atomic CRMAutomation and relationship substrate.

Admission gates

Every candidate needs an evidence trail that survives review and handoff.

01 · provenanceSource, commit, fork and transformation history.
02 · licenseSPDX fit, obligations and redistribution path.
03 · supply chainSBOM, dependency and static/AST scan.
04 · dataBoundaries, secrets, tenancy and migrations.
05 · UIAccessibility, visual fit and interaction contract.
06 · buildReproducible install, test and smoke evidence.
07 · actionAuthority, approvals, rollback and recovery.
08 · humanNamed owner signs the admission decision.
Open decision ledger ↗
07 / Source shelf

Open the evidence behind this page

Original artifacts remain in place.

LOCAL / PRIVATE

Full internal dossier

Existing research pack with private diligence and client-context material. Keep local unless access control is added.

Open full dossier locally ↗
MARKDOWN

Builder-space audit

What the Builder Research space produced, where it stopped, and which reports are preserved on disk.

Open space audit ↗
JSON / MD

Corpus evidence

Candidate reviews, license gate, checkpoint state, and cluster-seven preflight.

Open corpus status ↗
Deployment boundary: the root project also contains private source material. This hub is ready for a separate sanitized Cloudflare Pages upload; do not publish the entire project folder until the private dossier is protected with Cloudflare Access or excluded from the upload.