Portfolio interoperability guide · alpha.6
One contract · many native tools · human authority preservedOne Lattice, Many Tools
How open-source UVLM modules can speak with the CoherenceLattice cognition engine without becoming one compulsory monolith.
The Product Module Contract standardizes the governed exchange—identity, grounding, candidate quarantine, native findings, human decisions, bounded repair, telemetry, and replay. It does not standardize what every product must believe, calculate, teach, or conclude.
- v0.1.0-alpha.6
- 9 contract objects
- 10 capability roles
- 3 model postures
- Local-first reference
- Authority effect: NONE
Release basis and claim labels
This guide describes the separately released Product Module Contract. It is not a new release of the CoherenceLattice engine or any mapped product.
Release Safety Successor v1.0.2
adc29787223610c1fd1b5f3188aebe8096c7d7816e5d954906c0bab93208fe0ePublic open-source alpha under the exact recorded MPL-2.0 posture.
CoherenceLattice Community Edition v0.1.0-alpha.3
The cognition engine remains its own product, identity, codebase, and release library.
| Status label | What it means |
|---|---|
| Executed reference | The exact alpha.6 package runs and replays this route. |
| Included design mapping | The source contains a bounded module-mapping design; the underlying product was not rebuilt or conformance-certified. |
| Published direct bridge | A separate release already exchanges typed data with CoherenceLattice, but this is not automatically Product Module Contract conformance. |
| Proposed adapter | The native product appears structurally compatible, but no exact adapter pack or passing conformance evidence is claimed. |
| Source corpus | The work can be authenticated grounding; it is not executable module authority. |
Read status before possibility. The guide explains how products can participate. It does not claim that every adapter already exists or that every combination has been tested.
A common language, not a common personality
Forcing every product into the engine core would make CoherenceLattice responsible for every product’s ontology, terminology, release cycle, scientific assumptions, and failure modes.
Architectural rule. CoherenceLattice owns the governed exchange. Each domain product keeps its native expertise.
To “speak CoherenceLattice” means that a product can enter a common lifecycle while retaining its own data model and validator. The contract does not require all products to rewrite outputs in GUFT language, call a model, or accept a universal score.
| Canonical object | Contribution to the run |
|---|---|
module_manifest | Identity, capabilities, checks, model posture, and nonauthority. |
adapter_manifest | The selected model, tool, deterministic route, or human-operated adapter. |
request_envelope | Task, module, grounding references, constraints, and route. |
candidate_artifact | One quarantined output bound to exact bytes and lineage. |
finding_set | Native findings, executed checks, skipped checks, and unresolved work. |
human_decision | APPROVE, HOLD, REJECT, or REQUEST_REPAIR candidate-review posture. |
repair_packet | Only the changes needed for named findings and allowed scopes. |
telemetry_event | A bounded, local, hash-linked lifecycle event. |
run_receipt | The complete replay relationship and its nonclaims. |
human task + authenticated sources
↓
request envelope
↓
domain module or authorized adapter
↓
quarantined candidate artifact
↓
product-native validation
↓
finding set
↓
human decision
↓
optional bounded repair packet
↓
successor candidate + revalidation
↓
final human review posture
↓
telemetry chain + replayable run receiptCapability roles and model posture
A module declares only the work it performs:
TEACH · GENERATE · TRANSFORM · ANALYZE · VALIDATE · REPAIR · GOVERN · COMPARE · SEAL · VERIFY
These are capabilities, not authority. A generator cannot approve its own draft. A validator cannot certify external truth. A repairer cannot exceed the named repair packet. A sealing tool cannot certify scientific or ethical correctness.
No model needed
The trusted function is deterministic or human-operated.
A model may propose
Native checks remain independent of the adapter.
Adapter required
The task cannot proceed without a separately selected and authorized adapter.
Responsibility remains divided
CoherenceLattice
Owns canonical request, candidate, evidence, finding, repair, decision, telemetry, and replay identity.
Domain module
Owns expertise, protected surfaces, evidence policy, required checks, and native validators.
Adapter
Produces an untrusted candidate through the declared route. It does not bypass the module or human.
IntegrityLock
May authenticate and seal exact artifacts. It does not decide semantic correctness.
Human
Retains repair, acceptance, release, publication, deployment, professional judgment, and real-world authority.
What is implemented now
- The common nine-object contract exists and validates against ten Draft 2020-12 schemas.
- The deterministic Web Steward reference lifecycle executes end to end without a live model.
- The final fixture remains
HOLDbecause five external checks remain unexecuted; this is a safety result, not a defect to hide. - The alpha.6 source includes design mappings for IntegrityLock, CECP, ΔSyn Compression Lab, Governed AHA Lab, Echo Writing Primer, TIAIP, Web Steward, and Trauma-Informed Care work in progress.
- The separately published 432 Atlas and Rosetta modules already send typed observations to the exact CoherenceLattice alpha.3 validation boundary.
- Other public products described below still need thin adapter packs, product-specific conformance evidence, and exact-pack review.
No automatic absorption. A mapping file does not modify, relicense, rebuild, release, or certify the underlying product.
How the current public portfolio can participate
Each entry names its present relationship to the contract before describing the possible workflow.
1 · CoherenceLattice Operational Cognition Engine — the governed exchange hub
Core orchestration product · separately released
Contract role posture: The engine owns canonical run identity and evidence relationships; it is not treated as an ordinary module inside its own contract.
The engine authenticates grounding, validates typed candidate observations, preserves uncertainty, and emits deterministic review and replay artifacts. The Product Module Contract extends that pattern into a portfolio-wide lifecycle: modules contribute bounded candidates and native findings while CoherenceLattice keeps the run identities and cross-object bindings coherent.
A user-selected model may propose output through an authorized adapter. The adapter does not own approval, findings, publication, or release.
2 · IntegrityLock Community Edition — exact identity and lineage
Included alpha.6 design mapping
Contract role posture: SEAL · VERIFY · COMPARE · model posture NONE
IntegrityLock can authenticate module packs before a run and seal requests, candidates, findings, decisions, repairs, telemetry, receipts, and exports afterward. It can compare predecessors and successors so reviewers can see exactly which declared bytes changed.
The division of labor is deliberate: CoherenceLattice verifies semantic and governance relationships; IntegrityLock verifies artifact identity and lineage. Neither function proves the other.
3 · Critical Epistemic Continuity Protocol Community Edition — action-and-evidence governance
Included alpha.6 design mapping
Contract role posture: GOVERN · VALIDATE · model posture NONE
A structured proposal can be bound into a request, assessed by CECP’s native fixed contract, and returned as a finding set with a bounded PASS, HOLD, or REJECT posture. A HOLD can open a named repair route; a prohibited declaration can remain REJECT; a PASS closes only the CECP-specific contract gate.
CECP does not turn a model recommendation into permission. Consent, external facts, legality, publication, and consequential action remain separate.
4 · ΔSyn Conditional Compression Lab — deterministic comparison
Included alpha.6 design mapping
Contract role posture: ANALYZE · COMPARE · model posture NONE
A request can bind the exact target, declared donor, synchronization assumptions, coding profile, and provenance. The lab returns complete accounting—reconstruction, synchronization, payload, residual, provenance, and deterministic gzip—rather than only a favorable ratio.
Negative gain and NO_BENEFICIAL_DONOR remain valid outcomes. CoherenceLattice must not reinterpret conditional coding benefit as consciousness, semantic understanding, energy savings, or validation of GUFT.
5 · Governed AHA Lab — falsifiable cross-domain bridges
Included alpha.6 design mapping
Contract role posture: ANALYZE · VALIDATE · model posture OPTIONAL
The native product remains capable of a model-free route. The contract’s OPTIONAL posture permits an authorized adapter to draft a candidate while leaving source binding, mapping, disanalogies, prediction, falsifier, and rejection criteria under the native validator.
This separation keeps fluency from becoming evidence and makes a failed or non-specific bridge a preserved research result.
6 · Echo Writing Primer Community Edition — source and editorial governance
Included alpha.6 design mapping
Contract role posture: TEACH · GENERATE · VALIDATE · REPAIR · model posture OPTIONAL
The Primer can declare project-specific source-fidelity rules, protected voice features, role boundaries, and human editorial authority. A human or authorized adapter produces a draft candidate; native diagnostics identify unsupported invention, source displacement, agency loss, genericized voice, role confusion, and out-of-scope repair.
A human REQUEST_REPAIR decision authorizes a narrow successor. The revised draft is a new candidate with lineage, not a silent overwrite.
7 · 432 Atlas + GUFT Rosetta Bridge — a separately published typed observation path
Published direct bridge · not an alpha.6 conformance claim
Contract role posture: No Product Module Contract role manifest is claimed by this guide.
The Atlas/Rosetta public alpha already emits typed observations to the exact CoherenceLattice alpha.3 validation boundary. Its packet preserves a declared address, cell-local expression, GUFT/UCM axes, geometry, uncertainty, provenance, consent, residual, and the ability to return competing mappings or no adequate mapping.
That is a working example of domain sovereignty plus common exchange. A dedicated Product Module Contract adapter and conformance pack would still be separate work; the direct bridge must not be relabeled as having passed this contract merely because both systems share CoherenceLattice vocabulary.
8 · Global Philosophy Governance Atlas — structured pluralism
Proposed adapter path · not included in alpha.6 mapping set
Contract role posture: A future pack might declare TEACH, ANALYZE, GOVERN, and COMPARE; those roles are not yet an exact alpha.6 manifest.
A proposed adapter could return source-rooted lenses, internal disagreements, countervoices, translation losses, cultural-jurisdiction boundaries, affected-party authority, and limits on operationalization. CoherenceLattice could compare those packets without manufacturing consensus or ranking cultures.
Sacred, restricted, collective, and community-governed knowledge would remain subject to its own permissions. A cross-lens pattern is a candidate relation, not a vote that creates truth or authority.
9 · Union Controls Toolkit and UICAF — controls, evidence, findings, and remediation
Proposed adapter path · not included in alpha.6 mapping set
Contract role posture: A future pack might declare TEACH, GOVERN, VALIDATE, REPAIR, and VERIFY; those roles are not yet an exact alpha.6 manifest.
UICAF’s controls, risks, evidence packs, findings, remediation records, signatory receipts, and assurance objects are naturally compatible with a governed candidate lifecycle. Schema validation would remain separate from whether a control actually operated or whether a conclusion is professionally sufficient.
A bounded flow could move from local control or risk object, through native schema checks and evidence review, into human remediation and signatory decisions, with IntegrityLock sealing the resulting lineage.
10 · The Cost of Keeping One’s Name — governed source corpus, not software authority
Public source work
Contract role posture: No module role is assigned. The publication can be grounding; it does not execute, validate, govern, or decide.
The book can enter a run through a deterministic grounding bundle with exact source identity, segmented passages, section references, hashes, attribution, privacy boundaries, and publication posture. The engine can then support source-tracing, thematic comparison, controlled analogy testing, editorial review, or provenance sealing.
The text remains evidence, testimony, argument, and design material. Any derived interpretation is a candidate tied to passages and open to correction.
Additional alpha.6 mappings. The exact source also includes design mappings for TIAIP and Web Steward, plus a Trauma-Informed Care work-in-progress boundary. Those mappings remain non-release design artifacts unless separately sealed, reviewed, and authorized.
Three composable workflows
Governed local-AI writing
authenticated source bundle
→ user-selected local model draft
→ Echo Writing Primer findings
→ human repair authorization
→ bounded successor draft
→ final human judgment
→ IntegrityLock sealFluency does not become publication. Source support, voice, role boundaries, repair scope, and final authority remain visible.
Cross-domain research
declared target + donors
→ AHA structural bridge test
→ conditional compression comparison
→ optional Atlas/Rosetta observation
→ proposed plural-lens review
→ CoherenceLattice comparison + receipt
→ human research postureSupport, ambiguity, no benefit, high residual, refusal, and requests for smaller tests remain legitimate outcomes.
Union controls and assurance
UICAF control + risk objects
→ evidence pack
→ CECP action-and-evidence assessment
→ CoherenceLattice findings + repair lifecycle
→ accountable human decision
→ member assurance packet
→ IntegrityLock verification + sealThis would improve inspectability. It would not create an audit opinion, legal conclusion, or automatic compliance certificate.
What each adapter pack must contain
- A module manifest with identity, roles, model posture, checks, grounding policy, and nonauthority.
- A schema mapping that preserves native objects and semantics.
- A binding to the product’s real native validator.
- Positive, negative, held, refused, malformed, and adversarial reference fixtures.
- A deterministic finding policy with stable IDs, order, repair instructions, and scopes.
- A model or tool adapter declaration when candidate production is not deterministic.
- Telemetry and replay bindings to actual objects and bytes.
- A product-specific conformance suite that attacks identity, evidence, findings, repair scope, and authority.
- Fresh exact-package review plus any distinct human release decision required for the adapter pack.
Thin sidecars, not a rewrite. Every product should remain useful standalone. Integration must not become a compulsory monolith.
What the contract does not do
- A candidate is not a verdict, publication, deployment, or release.
- A passing contract check is not truth.
- A hash establishes identity or fixity, not correctness.
- A valid signature binds bytes to a key, not a key to legitimate authority.
- A skipped or unexecuted check cannot be rewritten as PASS.
- Grounding is evidence input, not automatic truth, consent, license, or permission.
- Product-native validation remains necessary.
- Model capability does not authorize model use.
- No model-training or persistent-memory authority is created.
- Human approval inside a candidate-review loop does not automatically authorize public release or real-world deployment.
- Module conformance does not certify accessibility, security, law, medicine, science, compliance, or production fitness.
- No module must join the integrated ecosystem to remain useful.
From a collection of tools to a governed cognition ecology
The contract does not make the portfolio valuable by making every product identical. It makes the products composable by preserving difference.
declare the task
bind the sources
identify the module
quarantine the candidate
run native checks
preserve uncertainty
name the findings
require human authority for repair
revalidate the successor
record the lifecycle
seal and replay the resultThat is how unlike tools begin to speak to one another without erasing what makes them useful.
One lattice. Many products. Shared evidence. Separate expertise. Human judgment preserved.