Publish OpenAPI/GraphQL/MCP contracts and partner onboarding proof
Partner diligence needs machine-readable API contracts, not only narrative docs.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0002, SAAS-0004
Acceptance criteria
- OpenAPI includes /v0, /v1/search, /v1/retrieve and /v0/rotor bounded routes.
- GraphQL and MCP examples are smoke-tested with demo tokens.
- The API page separates active bounded surfaces from planned Scenario/semantic work.
- Partner one-pager links to CLI verification commands.
Changelog closeout: Record spec paths and smoke receipt.
Choose and benchmark the French legal/accounting embedding strategy
Semantic search can be a moat only if retrieval quality is proven on French legal/accounting questions.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0003
Acceptance criteria
- Embedding dimensions are model-versioned, not hard-coded to one provider.
- Benchmark includes BOFiP, PCG/ANC, URSSAF, Code de commerce and e-invoicing queries.
- Precision/recall and failure examples are documented.
- semantic_search_active remains false until the activation gate passes.
Changelog closeout: Record selected model, rejected models, cost estimate and activation blocker.
Unlock Scenario only after corpus, retrieval and guardrail gates pass
Scenario is the product payoff, but enabling it before data and guardrails are ready would create professional reliance risk.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0001, SAAS-0003, SAAS-0005, SAAS-0010
Acceptance criteria
- Scenario has fixture, graph, RAG, retrieval and source-card gates per corpus family.
- No client-usable Scenario conclusion is exposed before the launch gate.
- The page explains Scenario locked vs review/demo-mode without ambiguity.
- If all machine gates pass, the final professional-review switch remains separate.
Changelog closeout: Record Scenario gate matrix and whether it remains locked.
Build SaaS organization, workspace, RBAC, quota and billing foundations
A buyer or partner needs tenant separation, usage controls and commercial packaging before production pilots.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0007, SAAS-0009
Acceptance criteria
- Tenant/workspace auth paths are documented and smoke-tested.
- Demo tokens are clearly separated from partner tokens.
- Rate limits and usage logs are visible for REST, GraphQL, MCP and CLI.
- No paying-users or ARR claim is made until real contracts exist.
Changelog closeout: Record tenant model and demo/partner boundary.
Create the operator dashboard for queue, leases, workers, parity and blockers
tmux is useful for humans, but it cannot be the nervous system of a SaaS corpus factory.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0002, SAAS-0004, SAAS-0005
Acceptance criteria
- Dashboard reads canonical Postgres/warehouse state, not terminal buffers.
- It shows stale leases and next action for every lane.
- It links to latest artifact checksums and receipts.
- It stays public-safe or has explicit auth separation.
Changelog closeout: Record dashboard URL and current health snapshot.
Refresh the diligence pack and external narrative from live evidence
External reviewers should see working APIs, corpus counts, rights caveats and roadmap status without stale valuation claims.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0006, SAAS-0009
Acceptance criteria
- The pack cites live public pages and local receipts.
- It does not claim ARR, paying users, trademark clearance, legal opinion or client-ready Scenario.
- It includes public demo bearer API tests for reviewers.
- Known limitations sit in a clear final section instead of reading like a wall of negatives.
Changelog closeout: Record pack path, receipt and public pages used.
Containerize one reference worker lane as the pets-to-cattle proof
The target SaaS architecture is credible only after one worker can be recreated without tmux history or host-local state.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0004, SAAS-0007, SAAS-0018
Acceptance criteria
- A fresh worker can register, claim, heartbeat and close out a safe task using only documented configuration.
- The worker writes no canonical state outside the control-plane API and artifact output paths.
- The same container image can run on mtl-01 or mtl-03 with host-specific secrets injected externally.
- A rollback path returns the lane to the existing Hermes/tmux posture if needed.
Changelog closeout: Record image/build path, smoke receipt, rollback command and remaining non-container lanes.
Publish a current diligence-room pointer and one-page buyer narrative
Reviewers should not have to infer the current asset story from many dated docs and terminal receipts.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0006, SAAS-0015, SAAS-0017
Acceptance criteria
- A single `docs/diligence/current.md` points to the latest active documents and receipts.
- The one-pager frames LegiPro as an asset-sale/platform-readiness story, not an ARR valuation.
- Negative caveats are grouped in a clear limitations section rather than scattered as alarm copy.
- Public docs link to current API/status/roadmap evidence without exposing secrets.
Changelog closeout: Record diligence pointer path, one-pager path and source receipts used.
Execute the staged SaaS substrate migration plan
Claude's target architecture argues the gap is substrate re-platforming, not product rewrite; the roadmap needs that work claimable.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0017, SAAS-0018, SAAS-0019
Acceptance criteria
- Phase 0 hygiene, Phase 1 containerization and Phase 2 managed/HA Postgres have explicit owners, receipts and rollback gates.
- Secrets move from host files into a documented KMS/Vault-compatible path before enterprise pilots.
- Gateway, SSO, telemetry and SLO tasks are sequenced after data/control-plane stability.
- The moat remains corpus/trace/guardrail quality; infrastructure migration does not widen readiness claims.
Changelog closeout: Record phased migration status and which SaaS substrate risks remain open.
Archive closed rotor task history out of the hot claim table
External review flagged the large done_machine_safe_delta history as buyer-visible operational drag; hot claim scans should stay small without losing audit evidence.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0002, SAAS-0005, SAAS-0018
Acceptance criteria
- `rotor_tasks_archive` exists with archived rows and false readiness constraints.
- `legipro rotor db-archive-done-tasks` is dry-run by default and rejects live statuses such as `leased`.
- At least one bounded execute batch archives closed rows without touching active leases.
- CLI docs and public API docs document the archive command and audit boundary.
Changelog closeout: Record archive counts, command, tests and remaining archive candidates.
Retire or quarantine dead frontend stubs for buyer-facing repo hygiene
External review flagged historical frontend stubs and a large scripts directory as sprawl; SAAS-0018 narrowed script operations, but visible dead UI surfaces still need explicit cleanup.
- Phase
- Next
- Priority
- P2
- Owner
- /rook-410
- Depends on
- SAAS-0018, SAAS-0020
Acceptance criteria
- Each legacy frontend stub has a disposition: active, redirected, archived, or deleted.
- Public navigation and docs do not link to retired stubs.
- Any removed/redirected public page has a rollback note and no broken internal links.
- The script inventory remains the operational source for CLI-first processing.
Changelog closeout: Record stub disposition counts, changed routes/files and rollback notes.
Define and smoke the Review Answer Runtime contract
The buyer-facing product should return a current, cited, reviewable answer with missing facts, risk flags, and an export bundle, not just raw search results.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0009, SAAS-0012
Acceptance criteria
- The contract includes answerability status, applicability, effective dates, sources, missing facts, risk flags, export_bundle_id and audit_id.
- At least one smoke fixture returns `insufficient_evidence` or `conflicting_sources` when sources do not support a safe answer.
- The public API guide documents this as a bounded review workflow, not a certified professional conclusion.
- Scenario, answer-ready, promotion, human-review, client-reliance and semantic flags remain false.
Changelog closeout: Record contract path, smoke fixtures and false-readiness boundary.
Separate Partner Product API, AI Tool Runtime and Internal Control Plane docs
Reviewers should not see buyer-facing product flows adjacent to queue minting and worker closeouts without a clear internal boundary.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0007, SAAS-0009, SAAS-0024
Acceptance criteria
- API docs separate product endpoints, MCP tools and internal rotor worker routes.
- Worker tokens cannot call partner-only product mutations or queue-minting commands.
- Partner examples use product/search/review/export flows, not rotor control-plane flows.
- The architecture map and public API page use the same surface names.
Changelog closeout: Record docs paths, route-scope checks and architecture map update.
Upgrade retrieval from search access to decision-grade evidence
Premium accounting/legal answers need authority, temporal validity, contradiction handling and citation envelopes, not keyword hits alone.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0003, SAAS-0010, SAAS-0024
Acceptance criteria
- Retrieval fixtures cover BOFiP, PCG/ANC, URSSAF, Code de commerce and e-invoicing examples.
- Every returned answer candidate carries source IDs, passage IDs, artifact IDs and date/asOf metadata.
- Older/superseded sources are marked or demoted in fixtures.
- semantic_search_active remains false until the separate SAAS-0010 activation gate passes.
Changelog closeout: Record retrieval fixtures, metrics and activation boundary.
Add product observability for answer, retrieval, export and partner value
Receipts are good diligence evidence, but daily SaaS operation needs live metrics around user value and degraded behavior.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0013, SAAS-0017, SAAS-0024
Acceptance criteria
- A status/dashboard artifact shows p50/p95 search/review/export latency and error counts.
- Answerability, insufficient-evidence and conflicting-source rates are tracked without private prompt leakage.
- Per-tenant cost units, throttles and denied-scope attempts are visible to operators.
- Metrics feed docs/status snapshots without broad repository scans.
Changelog closeout: Record metrics schema, dashboard path and snapshot receipt.
Productize export bundles and revisioned dossiers
For professional users and API buyers, the durable cited dossier may be more valuable than the chat/search interaction.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0012, SAAS-0024
Acceptance criteria
- Export bundle schema includes tenant/workspace, source IDs, passage IDs, dates, caveats, revision id and audit id.
- A buyer-branded export fixture is generated without exposing private prompts or secrets.
- Failed export returns a queued job id or clear retry state rather than blocking the review workflow.
- Exports remain review-gated and do not imply client reliance or professional certification.
Changelog closeout: Record export schema, fixture output and tenant/audit fields.
Expose source freshness, supersession and dossier impact analysis
The corpus factory becomes a moat when the product can tell which sources changed and which saved work may need review.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0003, SAAS-0026, SAAS-0028
Acceptance criteria
- Source records expose last_checked_at, last_changed_at, content hash, valid_from/valid_until and supersession status where known.
- A fixture shows a changed source and the affected saved dossier/export ids.
- Public docs explain freshness as product intelligence, not a professional review claim.
- Stale-source warnings appear before answer/export confidence fields.
Changelog closeout: Record source freshness fields, impact fixture and public docs update.
Define and test degraded-mode behavior for product runtime dependencies
Buyers test failure behavior; a useful system should degrade into source packs, queued exports or last-known-good corpus rather than fail ambiguously.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0013, SAAS-0024, SAAS-0027
Acceptance criteria
- Meili-down, warehouse-slow, assistant-unavailable, export-unavailable, worker-down and stale-index fixtures exist.
- Each degraded mode returns a clear status, warning and next action or job id.
- The status page and API docs identify last-known-good corpus/index timestamps.
- No degraded mode widens Scenario, answer-ready, promotion, client-reliance or professional-certification flags.
Changelog closeout: Record degraded-mode fixtures, API behavior and status docs update.
Move recurring CLI operations into scheduled jobs, admin APIs and CI gates
The CLI is excellent proof and break-glass tooling, but production paths should not depend on a human operating terminal commands.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0006, SAAS-0013, SAAS-0018, SAAS-0027
Acceptance criteria
- Each recurring CLI command has a scheduled/admin/CI owner or explicit break-glass label.
- Scheduled jobs write the same receipts as CLI runs and are visible in telemetry.
- Manual CLI use remains possible for local QA and rollback.
- No scheduled job performs broad repository scans or mints queue rows outside mtl-01.
Changelog closeout: Record scheduled/admin/CI mapping and first automated receipt.
Roadmap partner SDKs and generated client workflow
Partners should be able to integrate LegiPro quickly from their own product surfaces without copying curl snippets or learning rotor internals.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0009, SAAS-0024, SAAS-0025, SAAS-0028, SAAS-0030
Acceptance criteria
- Public roadmap and API docs list SDKs as planned/not shipped, with TypeScript/Node as the first target and Python as the second target.
- SDK generation depends on a stable OpenAPI/contract artifact and does not hand-maintain business logic outside REST, GraphQL, MCP and Review Runtime contracts.
- Examples cover search, retrieve, review-preview, export bundle and MCP tool calls with tenant/workspace/cost-unit headers.
- SDK docs warn against embedding bearer keys in browser/mobile clients and preserve Scenario, answer-ready, client-reliance and semantic-search false boundaries.
Changelog closeout: Record SDK roadmap, generated-client source, examples and contract-test receipt.
Publish static OpenAPI snapshot for partner SDK codegen
Partner SDK generation needs a machine-readable contract before the live /openapi.json route is promoted.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0009, SAAS-0025, SAAS-0032
Acceptance criteria
- A public static OpenAPI JSON asset exists and parses as OpenAPI 3.1.
- The snapshot covers /v0/search/health, /v0/search, /v1/search, /v1/retrieve, /graphql, /mcp and protected /v0/rotor worker routes.
- The API contract and API guide point to the static asset while still saying /openapi.json is not a live JSON route.
- The snapshot keeps SDK status planned/not shipped and preserves Scenario, answer-ready, client-reliance, semantic-search and professional-certification false boundaries.
Changelog closeout: Record static OpenAPI snapshot, asset URL, contract wiring and tests.
Publish partner API quickstart pack
A buyer should be able to run the first server-side integration checks without reading the whole API guide.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0009, SAAS-0025, SAAS-0033
Acceptance criteria
- A public quickstart asset exists for server-side partner integration.
- The quickstart covers health, search, retrieve, GraphQL and MCP tool-call flows.
- The API contract and API guide link the quickstart alongside the static OpenAPI snapshot.
- The quickstart forbids browser/mobile bearer-token exposure and product-token rotor calls.
- Scenario, answer-ready, client-reliance, semantic-search and professional-certification flags remain false.
Changelog closeout: Record partner quickstart asset, public URL, contract wiring and tests.
Ship the first generated TypeScript/Node partner SDK beta
The SDK roadmap needs a concrete first delivery so partner teams can integrate LegiPro from Node services without assembling thin wrappers by hand.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0032, SAAS-0033, SAAS-0034
Acceptance criteria
- A generated TypeScript/Node beta client exists with a documented package/version source and no hand-maintained business logic fork.
- Server-side examples cover search, retrieve, review-preview, export bundle preview and MCP tool calls with tenant/workspace/cost-unit headers.
- Contract tests prove the generated client matches the static OpenAPI and partner contract artifacts and reject browser/mobile bearer-token usage.
- The public roadmap, API guide and partner contract asset identify the TypeScript/Node SDK as the first delivery target while keeping SDK status pre-GA and all readiness/professional flags false.
Changelog closeout: Record generated TypeScript/Node SDK beta package, examples, contract-test receipt and public docs update.
Implement the Review Runtime kernel inside the current modular monolith
External reviews now agree the next multiplier is a professional review workflow engine, not more standalone surface area or more infrastructure prose.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0024, SAAS-0025, SAAS-0026, SAAS-0028, SAAS-0030
Acceptance criteria
- The current Product API gains in-repo runtime modules for review orchestration, retrieval, applicability, answer contract, citation verification, dossier handling and audit emission.
- The implementation reuses existing search/retrieve/export contracts and does not create a second business-logic brain beside REST, GraphQL, MCP and CLI.
- Exactly two LLM stages are allowed in the runtime path: request normalization and answer synthesis; citation verification remains deterministic/mechanical.
- The delivery may remain preview/private and does not require a live public /v1/review route; Scenario, answer-ready, promotion, human-review, client-reliance, semantic-search and professional-certification flags remain false.
Changelog closeout: Record the runtime module paths, tests, and the first bounded preview wiring without widening any readiness claim.
Add snapshot-pinned review requests, applicability checks and refusal remedies
A buyer-grade review workflow must be replayable against a dated corpus state and must know when to refuse, what is missing, and which sources conflict.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0024, SAAS-0026, SAAS-0029, SAAS-0038
Acceptance criteria
- Review requests can pin corpus/index snapshot IDs as inputs, not only receive them as outputs.
- Applicability checks cover jurisdiction, effective dates, supersession, source hierarchy and missing-fact detection with deterministic receipts or fixtures.
- Refusal taxonomy is explicit: insufficient_evidence requires missing_facts with remedies, conflicting_sources requires contradiction pairs with authority_rank and remedy guidance, and partially_answerable remains distinct from answerable.
- A validate/confirm-refute workflow exists as a bounded mode for checking a proposed treatment without claiming professional certification or client reliance.
Changelog closeout: Record snapshot-pinned request/response examples, applicability checks, validate-mode proof and refusal-remedy fixtures.
Build the golden review eval set and runtime audit sink
The review runtime only becomes diligence-grade when every run is measurable and regressions are caught against labeled professional cases, including refusal and contradiction scenarios.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0027, SAAS-0038, SAAS-0039
Acceptance criteria
- A labeled golden review set exists with answerable, partially_answerable, insufficient_evidence, conflicting_sources and validate-mode cases.
- Runtime audit events record audit_id, tenant/workspace scope, latency, answerability status, citation coverage, missing-fact count, contradiction count and snapshot IDs without storing raw private prompts in public-safe assets.
- CI or CLI gates fail closed when the review runtime regresses on golden answerability/refusal/citation checks.
- The public observability/docs surfaces describe this as bounded quality evidence, not a claim of human review, client reliance or professional certification.
Changelog closeout: Record the golden eval-set paths, runtime audit schema, gating command and the first passing receipt.
Publish a current/planned/target architecture legend and readiness matrix
External reviewers now understand the C4 better, but they still need a one-glance legend showing what is live today, what is planned inside the current monolith, and what belongs to the target SaaS substrate.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0025, SAAS-0038
Acceptance criteria
- The public architecture/C4 surface includes a visible legend for current, planned and target components.
- Meilisearch is explicitly presented as a derived cache and the corpus warehouse as the durable source of truth in both diagram and adjacent text.
- A readiness matrix distinguishes current active, current bounded, planned and target SaaS items without widening any readiness claim.
- Scenario, answer-ready, client-reliance, semantic-search and professional-certification boundaries remain false and are repeated beside the matrix.
Changelog closeout: Record the legend/matrix assets, public URLs and the unchanged false-readiness boundary.
Add dedicated buyer integration and corpus-factory architecture views
The strongest acquirer story is now the embed-without-UI path and the receipt-backed corpus factory flow; those deserve first-class diagrams instead of being implied across one larger map.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0025, SAAS-0043
Acceptance criteria
- A public 'Partner Integration Without UI Adoption' architecture view exists and shows buyer product -> REST/GraphQL/MCP -> evidence/dossier outputs.
- A public 'Corpus Factory and Readiness Gates' architecture view exists and shows worker claim/heartbeat/closeout, warehouse/index separation, backup proof and false-readiness gates.
- Both views stay aligned with the canonical workspace DSL and are linked from the API or architecture surfaces without inventing a second source of truth.
- The views remain bounded and do not imply Scenario launch, professional sign-off or semantic-search activation.
Changelog closeout: Record the new rendered views, their public links and the DSL/source files they derive from.
Create the Cegid technical screening pack
A partner/product reviewer needs one narrow package: how to test LegiPro, what Cegid APIs it could map to, what is active, what is not and what evidence proves it.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0045, SAAS-0046, SAAS-0047
Acceptance criteria
- A single Cegid screening pack exists with index, links and local/public evidence paths.
- The pack maps LegiPro REST/GraphQL/MCP/SDK surfaces to likely Cegid Loop/partner API touchpoints without claiming an approved Cegid partnership.
- The pack includes runnable local commands for LegiPro API smoke, SDK example smoke and Cegid adapter fixture smoke.
- The pack repeats the no-SLA/no-ARR/no-client-reliance/no-professional-certification boundaries.
Changelog closeout: Record the Cegid screening pack path, commands, tests and public-safe publication state.
Run the e-invoicing/VAT partner proof path end to end
A credible screening call should show one professional workflow tied to the September 2026 rollout, not just architecture: question or document input, source-backed review packet, draft dossier/export and partner-shaped handoff fixture.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0039, SAAS-0041, SAAS-0047
Acceptance criteria
- The demo path uses one bounded workflow and includes screenshots or command receipts from public-safe LegiPro surfaces.
- The output includes answerability status, sources, applicability, missing facts/risks if any, draft dossier/export id and audit id.
- The partner adapter fixture consumes the export/dossier metadata without live third-party credentials.
- The demo does not claim Scenario launch, client reliance, human review, semantic search activation or a third-party partnership.
Changelog closeout: Record the Cegid-style pilot demo receipt, fixture output and screenshots or public-safe command evidence.
Define the Cegid escalation gate
COO/GM presentation should be a gated escalation after partner/product validation, not the first outreach step.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0046, SAAS-0048, SAAS-0049
Acceptance criteria
- A Cegid escalation checklist exists with required evidence for Partner/Product, AI/Product, Startup Program and COO/GM routes.
- The checklist states that COO/GM escalation is blocked until at least one product/partner owner validates fit or requests senior sponsorship.
- The checklist includes the exact diligence artifacts to send at each stage and the artifacts to withhold from first contact.
- The checklist stays aligned with current product boundaries: no ARR, no paying users, Scenario locked and professional reliance not certified.
Changelog closeout: Record the Cegid escalation gate path and the evidence threshold for executive-ready status.
Ship first-wave self-serve adapter stubs
After one proof fixture exists, Cegid Loop/Expert, Sage, Pennylane, Odoo, Abby and Inqom have enough public API detail to support additional no-credential stubs that demonstrate LegiPro as an embeddable source-review engine.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0052
Acceptance criteria
- At least three first-wave adapters exist and consume the shared adapter fixture harness.
- The first-wave candidate list is explicit: Cegid Loop/Expert, Sage Accounting/Active, Pennylane, Odoo, Abby and Inqom.
- Each adapter maps one bounded LegiPro output to a plausible vendor object flow such as document handoff, invoice/customer sync, dossier/context sync or accounting-entry handoff.
- Each adapter includes public-doc citations, dry-run sample output, fail-closed credential behavior and tests.
- No adapter makes live API calls unless an explicit non-default credentialed mode is later added and separately documented.
Changelog closeout: Record the first-wave adapter paths, sample outputs, tests and no-live-call boundary.
Create partner-gated strategic adapter briefs
Some of the most strategically useful buyers have gated or partial public API docs; LegiPro should still show the intended integration shape without inventing endpoint details.
- Phase
- Next
- Priority
- P1
- Owner
- unclaimed
- Depends on
- SAAS-0051
Acceptance criteria
- Each brief states verified public docs, missing access details, likely integration objects and exact partner/sandbox ask.
- The briefs clearly separate public-doc facts, inference and unresolved access questions.
- The partner-gated list is explicit: EBP, Cegid XRP, ACD/i-Suite, RCA/MEG, Sellsy, Dext, Tiime, Yooz, Macompta.fr, Indy and Axonaut.
- No brief claims an approved partnership, live credential, customer integration, SLA or professional reliance.
Changelog closeout: Record the partner-gated brief paths, vendors covered and unresolved access questions.
Create per-platform value cards for the full accounting ecosystem
The integration roadmap should be readable by any mapped accounting platform, not only a single strategic contact.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0051, SAAS-0052
Acceptance criteria
- Value cards exist for Cegid Loop, Cegid Expert/Quadra, Cegid XRP, Sage Accounting, Sage Active, Pennylane, EBP, Odoo, Abby, Inqom, ACD/i-Suite, RCA/MEG, Tiime, Yooz, Macompta.fr, Sellsy, Dext, Indy and Axonaut.
- Each card states client value, likely object flow, integration readiness tier, public-doc confidence and unresolved access questions.
- The cards avoid vendor-specific overclaims and can be reused in outbound notes to multiple platforms.
- The cards preserve all no-partnership, no-live-third-party-call, no-ARR, no-client-reliance and no-professional-certification boundaries.
Changelog closeout: Record the per-platform value card paths, vendors covered and reusable outbound/public-safe status.
Create the e-invoicing proof and first-screening comparison pack
The outreach should create urgency from the September 2026 rollout, not pressure tactics: early product screens can shape the proof workflow while it is still concrete and not yet a generic adapter kit.
- Phase
- Next
- Priority
- P1
- Owner
- /rook-410
- Depends on
- SAAS-0051, SAAS-0057
Acceptance criteria
- The pack explains the early-screening advantage without threatening a bidding process or claiming competing partnerships.
- The pack shows which integration decisions remain shapeable before the generic adapter kit hardens.
- The pack includes the first proof fixture receipt and the full vendor map.
- The pack is reusable for Cegid, Sage, Pennylane, EBP, Odoo, Abby, Inqom, ACD/RCA/Tiime/Yooz/Macompta/Sellsy/Dext/Indy/Axonaut-style conversations.
Changelog closeout: Record the first-mover proof pack path, included fixture evidence and public-safe outbound boundary.