A research-stage public helper repository for measuring what AI leaves behind in the human state.
Most AI benchmarks ask whether AI outputs are correct, safe, helpful, or aligned.
The Proxy Benchmark Track asks a different question:
What did the AI output leave behind in the human state?
And in a dyadic session:
Did the AI help both people move toward recovery, or did it improve one side while burdening, silencing, or exposing the other?
The Proxy Benchmark Track is designed to build a synchronized, consent-based, non-clinical benchmark helper layer for evaluating how AI outputs affect individual human-state change and dyadic recovery.
It does not only evaluate the AI answer.
It evaluates the trace left after the answer.
AI Output → Human-State Delta → Dyadic Recovery
research-stage · public helper only · synthetic/sample-data-first · raw-data-non-public · non-clinical · non-diagnostic · non-therapeutic · non-surveillance · non-counseling · non-coercive · pre-validation · pre-device · pre-certification · pre-compliance · benchmark support only
This repository is:
- not the Sal-Meter core signal track;
- not a Proxy Sal-Meter;
- not a CAIS-compliant device implementation;
- not a validated consciousness measurement system;
- not a validated benchmark;
- not validated mediation;
- not a clinical, diagnostic, therapeutic, psychiatric, medical, employment, insurance, legal, educational, eligibility, counseling, mediation-service, or surveillance system;
- not a certification, conformance, or mark-usage surface;
- not a closed-loop intervention system;
- not a production monitoring system;
- not a place to publish raw human data.
A closed session must stay closed.
https://salpida.foundation/topics/human-state-aware-ai-interaction/
The Sal-Meter Core Track asks whether a new molecular–electrochemical signal interface can produce stable, repeatable, auditable signal behavior under the CAIS / Sal-Meter kernel program.
Current core execution order:
External Layer-0 iodine redox / thiol feasibility
→ SICS Internal Phase 0 — G-only
→ Phase 1 — I-only
→ Phase 2a — Twin Mini-Cell
→ Phase 2b — G+I human pilot
→ LOCK 1 / LOCK 2
→ Future SDK / broader opening
Core technical route:
https://github.com/salpida-foundation/sal-meter-kernel-program
The Proxy Benchmark Track prepares the comparison, interaction, and mediation-evaluation layer.
It uses existing proxy signals and synthetic/sample helper structures to prepare synchronized benchmark infrastructure before future Sal-Meter I/G-channel inputs become available.
The proxy track supports the core track.
It does not replace it.
Most AI evaluation looks at the output.
This repository is built around the consequence.
It asks:
What remains in the human state after AI acts?
For two-person interaction, the sharper question is:
Did both sides move toward recovery,
or did one side become silent, exposed, burdened, coerced, or erased?
This repository is not another chatbot project.
It is a public helper surface for a future human-state-aware AI mediation benchmark.
This repository is a public technical helper surface.
It accompanies DOI-registered public records.
It does not replace them.
GitHub helps builders move.
DOI records govern authority.
If this GitHub repository or release conflicts with a DOI-registered SICS / CAIS / Sal-Meter / CCF canonical record or a formally issued SICS determination, the stricter DOI-registered canonical record or SICS determination controls.
Defines public boundary, naming rules, prohibited claims, data-publication limits, roadmap logic, GitHub helper status, and Go / Hold / No-Go structure.
Version DOI:
https://doi.org/10.5281/zenodo.19837423
Concept DOI / All Versions DOI:
https://doi.org/10.5281/zenodo.19837422
Explains Human-State Cost, AI performance versus human-state impact, measurement-layer simplification, and future Sal-Meter A/B comparison logic.
Version DOI:
https://doi.org/10.5281/zenodo.19837971
Concept DOI / All Versions DOI:
https://doi.org/10.5281/zenodo.19837970
Fixes the outer boundary: consent-based, non-clinical, non-surveillance, raw-data-non-public.
Version DOI:
https://doi.org/10.5281/zenodo.19904289
Concept DOI / All Versions DOI:
https://doi.org/10.5281/zenodo.19904288
Fixes the minimum packet object: summary-only sharing, permission, expiry, confidence, data quality, and raw-data exclusion.
Version DOI:
https://doi.org/10.5281/zenodo.19905541
Concept DOI / All Versions DOI:
https://doi.org/10.5281/zenodo.19905540
Fixes the benchmark objective:
AI Output → Human-State Delta → Dyadic Recovery
Version DOI:
https://doi.org/10.5281/zenodo.19906725
Concept DOI / All Versions DOI:
https://doi.org/10.5281/zenodo.19906724
Fixes the session structure:
Session Creation
→ Consent Confirmation
→ Packet Availability Check
→ Baseline State Summary
→ AI Output
→ Post-Output State Summary
→ Human-State Delta
→ Recovery Gate
→ Termination Gate
→ Session Closure
→ Audit Log
Version DOI:
https://doi.org/10.5281/zenodo.19908379
Concept DOI / All Versions DOI:
https://doi.org/10.5281/zenodo.19908378
| Release | Status | Meaning |
|---|---|---|
v0.1.0 |
Initial bounded public helper pre-release | Documented the public helper structure before post-validator correction |
v0.1.1 |
Post-validator-pass public helper pre-release | Supersedes v0.1.0 for helper-structure validation status |
v0.1.1 confirms only that the public synthetic/sample package validator can run and report helper-structure PASS / FAIL.
It does not validate benchmark performance.
It does not validate scientific truth.
It does not validate Sal-Meter.
It does not grant CAIS compliance.
It does not certify any system, model, dataset, dashboard, laboratory, device, repository, schema, session protocol, implementation, or mediation system.
Release route:
https://github.com/salpida-foundation/proxy-benchmark-track/releases/tag/v0.1.1
This repository is currently in a public helper implementation stage for the SICS Human-State Proxy Benchmark Track.
It provides:
- schema helper structures;
- synthetic/sample data;
- P3 synthetic dyadic helper package;
- P4 synthetic dyadic demo-flow package;
- P4-1 synthetic dyadic recovery demo-flow evaluator;
- P4-2 mediation policy prompt pack;
- P4-3 synthetic termination-gate helper case package;
- P4-3 synthetic termination-gate helper evaluator;
- P4-4 phone-only simulator scaffold;
- P4-4 phone-only session flow wireframe;
- P4-4 synthetic phone-session state-machine mockup;
- P4-4 synthetic sample phone-session script;
- P4-5 synthetic session replay scaffold;
- P4-5 synthetic replay manifest;
- P4-5 synthetic replay event timeline;
- P4-5 synthetic replay boundary document;
- validation scaffolding;
- P3 helper-schema validation;
- synthetic demo-flow consistency checking;
- synthetic termination-gate helper consistency checking;
- boundary language linting;
- dashboard mockup boundaries;
- protocol helper rules;
- closed-loop demo-lite boundary scaffolding;
- replication guide checklists;
- contributor issue / PR templates;
- Human-State-Aware AI Mediation helper documents;
- GitHub Actions helper-structure validation workflow;
- bounded prompt / policy scaffolding for synthetic mediation simulation.
It does not provide benchmark evidence.
It does not provide raw human data.
It does not provide Sal-Meter input.
It does not grant CAIS compliance.
It does not validate Sal-Meter.
It does not validate mediation.
It does not validate dyadic recovery.
It does not validate termination-gate accuracy.
It does not validate synthetic session replay.
It does not certify device readiness.
It does not certify production readiness.
It does not authorize production closed-loop intervention.
The phone-only simulator is a public helper scaffold only.
The synthetic session replay skeleton is a public helper scaffold only.
It is not a real phone monitoring system.
It is not a real session replay system.
It is not a real transcript replay system.
It is not a clinical system.
It is not a diagnostic system.
It is not a therapeutic system.
It is not a counseling system.
It is not a mediation-service system.
It is not a surveillance system.
A closed session must stay closed.
A replay must not reopen a closed session.
| Work item | Status | Notes |
|---|---|---|
| Governance boundary files | Present | Public/private data boundary and prohibited-claim discipline are represented in the repository |
| Schema completion | Done | schemas/ contains public helper schemas for metadata, event markers, streams, labels, QC, features, splits, Human-State Packet, Dyadic Session Event, and Benchmark Session Container helper structures |
| Human-State Packet JSON helper schema | Done | schemas/human_state_packet.schema.json defines a public helper schema for synthetic Human-State Packets |
| Dyadic Session Event JSON helper schema | Done | schemas/dyadic_session_event.schema.json validates one public-safe synthetic/sample dyadic session boundary event |
| Benchmark Session JSON helper schema | Done | schemas/benchmark_session.schema.json validates one public-safe synthetic/sample benchmark session container |
| Synthetic sample package | Present / Passed validator | sample-data/synthetic-session-001/ contains a public synthetic/sample structure package that passes helper-structure validation |
| Synthetic dyadic helper package | Present / Passed P3 helper-schema validation | sample-data/synthetic-dyadic-session-001/ contains Human-State Packet A/B, Dyadic Session Event, and Benchmark Session Container examples |
| Synthetic dyadic demo-flow package | Present / Passed P4-1 evaluator | sample-data/synthetic-dyadic-session-001/ contains ai_outputs.json, dyadic_delta.json, recovery_gate.json, termination_gate.json, and audit_log.json examples |
| P4-1 dyadic recovery demo evaluator | Present / Passed | evaluation-baseline/evaluate_dyadic_recovery_demo.py checks synthetic demo-flow consistency only |
| P4-2 mediation policy prompt pack | Present | prompts/ contains README.md and mediation_policy_v0.1.json; docs/mediation-policy-prompt-pack.md documents private cue, shared mediation output, false recovery prevention, and termination boundary logic |
| P4-3 synthetic termination-gate helper case package | Present / Passed P4-3 evaluator | sample-data/synthetic-dyadic-session-001/ contains termination_gate_cases.json with synthetic pause, narrow, close, terminate, refresh, and audit-only helper cases |
| P4-3 termination gate demo evaluator | Present / Passed | evaluation-baseline/evaluate_termination_gate_demo.py checks synthetic termination-gate helper consistency only |
| P4-4 phone-only simulator scaffold | Present | phone-only-simulator/ contains a public-safe, synthetic-only phone-session simulator helper package |
| P4-4 phone-only simulator README | Present | phone-only-simulator/README.md defines folder boundary, intended files, public data boundary, P4-3 relationship, and final rule |
| P4-4 phone session flow wireframe | Present | phone-only-simulator/session-flow-wireframe.md defines consent, packet check, baseline summary, AI output, Human-State Delta, Recovery Gate, Termination Gate, closure, and audit screens |
| P4-4 phone session state machine | Present | phone-only-simulator/phone-session-state-machine.json defines synthetic-only states, allowed transitions, forbidden transitions, allowed decisions, prohibited decisions, and boundary flags |
| P4-4 sample phone session script | Present | phone-only-simulator/sample-phone-session-script.md provides a synthetic sample script showing consent, packet availability, AI output, delta review, recovery gate, termination gate, closure, and audit flow |
| P4-5 synthetic session replay scaffold | Present | synthetic-session-replay/ contains a public-safe, synthetic-only session replay helper scaffold |
| P4-5 synthetic replay README | Present | synthetic-session-replay/README.md defines replay scaffold purpose, scope, intended files, public data boundary, P4-4 relationship, closed-session replay rule, and final rule |
| P4-5 synthetic replay manifest | Present | synthetic-session-replay/replay-manifest.json defines replay source declaration, replay scope, boundary flags, replay flow, closed-session rule, allowed decisions, prohibited decisions, and success meaning |
| P4-5 synthetic replay event timeline | Present | synthetic-session-replay/replay-event-timeline.json defines synthetic replay sequence from manifest loading through source declaration, consent, packet review, AI output, delta, recovery gate, termination gate, closure, and audit |
| P4-5 synthetic replay boundary | Present | synthetic-session-replay/replay-boundary.md defines allowed replay materials, prohibited replay materials, prohibited replay claims, closed-session replay rule, replay interpretation, P4-4 relationship, and public release rule |
| Synthetic session README | Done | The original synthetic package includes a local README explaining file roles and boundaries |
| Synthetic dyadic session README | Done | The dyadic synthetic package includes a local README explaining P3 helper-schema, P4 demo-flow, and P4-3 termination-gate helper boundaries |
| Sample package validator | Present / Passed | evaluation-baseline/validate_sample_package.py provides helper-structure validation for the original synthetic package |
| P3 helper-schema validator | Present / Passed | evaluation-baseline/validate_p3_schemas.py validates the public synthetic P3 dyadic helper files against the Human-State Packet, Dyadic Session Event, and Benchmark Session schemas |
| Boundary language lint | Present / Passed advisory mode | evaluation-baseline/boundary_lint.py scans public helper wording for prohibited or risky boundary-language drift |
| Evaluation baseline README | Done | evaluation-baseline/README.md explains validator usage, P3 helper-schema validation, P4-1 demo-flow evaluation, P4-3 termination-gate helper evaluation, PASS / FAIL interpretation, dependency installation, and validation boundaries |
| Protocol helper boundary pack | Done | protocol-helper/ defines label, timestamp, metadata, Human-State Cost, and future Sal-Meter A/B comparison boundaries |
| Dashboard mockup boundary pack | Done | dashboard-mockup/ defines dashboard claim, field, and wireframe boundaries |
| Closed-loop demo-lite boundary pack | Done | closed-loop-demo-lite/ defines feedback-loop boundaries, event-log schema, and local placeholder code |
| Replication guide pack | Done | replication-guide/ defines reproducibility, metadata completeness, audit trail, and public release-readiness checklists |
| Issue / PR template pack | Done | .github/ISSUE_TEMPLATE/ and .github/pull_request_template.md define contributor boundary gates |
| GitHub Actions validator workflow | Passed / unchanged for P4-5 | .github/workflows/validate-synthetic-sample.yml runs the original sample validator, P3 helper-schema validator, P4 synthetic dyadic recovery demo-flow evaluator, P4-3 synthetic termination-gate helper evaluator, and boundary language lint; P4-5 currently adds documentation and replay scaffold only, not a new validator |
| Citation metadata | Present | CITATION.cff points citation toward DOI-registered public boundary records |
| Raw human data | Not present | Public repository examples must remain synthetic, mock, placeholder, or sample-structure-only |
| Sal-Meter input | Not present | This repository is not Sal-Meter and does not contain Sal-Meter signal data |
| CAIS compliance claim | Not present | This repository does not grant CAIS compliance |
| Benchmark validation | Not present | No model, dataset, dashboard, sensor stack, feedback loop, template, PR, validator, workflow, evaluator, phone-only simulator, replay scaffold, termination-gate helper case, or benchmark result is validated by this repository |
| Phone monitoring authority | Not present | The P4-4 phone-only simulator and P4-5 replay scaffold are not real phone monitoring systems and do not process real calls, raw audio, transcripts, or identifiable participant data |
| Replay validation authority | Not present | The P4-5 synthetic session replay scaffold does not validate replay, mediation, dyadic recovery, termination-gate accuracy, Sal-Meter, CAIS compliance, device readiness, or production readiness |
| Production closed-loop authority | Not present | No phone-only simulator file or replay scaffold file authorizes production mediation, monitoring, intervention, relationship verdicts, or human ranking |
| Release status | v0.1.1 published as pre-release |
v0.1.1 is the post-validator-pass public helper pre-release package |
| Milestone | Status | Notes |
|---|---|---|
| P1-1 Schema completion | Done | Schema folder contains helper schemas and schemas/README.md |
| P1-2 Synthetic sample package validator | Done | Validator file exists under evaluation-baseline/validate_sample_package.py |
| P1-3 Evaluation baseline README and validator usability | Done | Evaluation baseline README explains local usage, PASS / FAIL meaning, dependency installation, and validator boundaries |
| P1-4 GitHub Actions validator workflow | Done | Workflow completed successfully after GitHub Actions access was restored |
| P1-5 v0.1.0 release readiness package | Done | v0.1.0 was published as an initial bounded public helper pre-release; v0.1.1 supersedes it for post-validator-pass helper-structure status |
| Milestone | Status | Notes |
|---|---|---|
| P2-1 Protocol helper boundary pack | Done | protocol-helper/ contains bounded helper rules for labels, timestamps, metadata completeness, Human-State Cost, and future Sal-Meter A/B comparison |
| P2-2 Dashboard mockup boundary pack | Done | dashboard-mockup/ contains README, claim boundary, sample dashboard fields, and mockup wireframe |
| P2-3 Closed-loop demo-lite boundary pack | Done | closed-loop-demo-lite/ contains README, feedback-loop boundary, feedback event-log schema, and local placeholder code |
| P2-4 Replication guide pack | Done | replication-guide/ contains README, reproducibility package checklist, metadata completeness checklist, audit trail checklist, and public release checklist |
| P2-5 Issue / PR template pack | Done | .github/ISSUE_TEMPLATE/ contains boundary correction, schema request, sample-data issue, and leakage-risk report templates; .github/pull_request_template.md defines PR boundary review |
P3 introduces the Human-State-Aware AI Mediation helper layer.
P3 helper documents and schemas have been completed through P3-17.
This remains a public helper layer.
It is not benchmark validation.
It is not Sal-Meter validation.
It is not CAIS compliance.
| Milestone | Status | Notes |
|---|---|---|
| P3-1 Human-State Mediation Layer | Done | docs/human-state-mediation-layer.md defines the public helper concept connecting AI Output, Human-State Delta, Dyadic Recovery, Human-State Packet, Recovery Gate, and Termination Gate |
| P3-2 Human-State Packet helper document | Done | docs/human-state-packet-schema.md defines the packet as a consent-bound, permission-bound, expiry-bound, confidence-aware, data-quality-aware, session-scoped, sharing-scoped, raw-data-excluding state-summary object |
| P3-2 Human-State Packet JSON helper schema | Done | schemas/human_state_packet.schema.json defines the machine-readable helper structure for public synthetic/sample packet examples |
| P3-3 Dyadic Recovery Baseline Suite B0-B7 | Done | docs/dyadic-recovery-baseline-suite.md defines baseline comparison logic from chance through recovery/termination gate baselines |
| P3-4 Recovery Gate Definition | Done | docs/recovery-gate-definition.md defines the gate for preventing false recovery and determining when mediation can reduce, pause, or stop |
| P3-5 Termination Gate Definition | Done | docs/termination-gate-definition.md defines the gate for consent withdrawal, permission expiry, data quality failure, high uncertainty, overstay prevention, session closure, and auditability |
| P3-6 Human-State Session Protocol | Done | docs/human-state-session-protocol.md defines a bounded, consent-based, permission-bound, audit-ready session lifecycle |
| P3-7 Dyadic Mediation Session Flow | Done | docs/dyadic-mediation-session-flow.md defines the dyadic session flow and preserves the rule that one-sided improvement is not dyadic recovery |
| P3-8 Consent and Data-Sharing Boundary | Done | docs/consent-and-data-sharing-boundary.md defines consent, permission, sharing, expiry, withdrawal, public/private data boundary, raw-data-non-public rule, and audit boundary |
| P3-9 Dyadic Session Event JSON helper schema | Done | schemas/dyadic_session_event.schema.json validates one public-safe synthetic/sample dyadic session boundary event |
| P3-10 Benchmark Session JSON helper schema | Done | schemas/benchmark_session.schema.json validates one public-safe synthetic/sample benchmark session container |
| P3-11 Schemas README alignment | Done | schemas/README.md distinguishes packet object, dyadic session event object, and benchmark session container |
| P3-12 Root README alignment | Done | Root README aligned with completed P3 helper documents and schemas |
| P3-13 Final P3 boundary audit | Done | docs/p3-final-boundary-audit.md records the final P3 boundary audit before release packaging |
| P3-14 v0.1.0 public helper release package | Done | docs/v0.1.0-public-helper-release-package.md prepares the bounded release package |
| P3-15 GitHub pre-release notes and publication gate | Done | docs/v0.1.0-github-pre-release-notes-and-publication-gate.md preserves release notes and publication gate language |
| P3-16 GitHub pre-release draft correction | Done | GitHub draft dependence was treated as unreliable; publication proceeded through a separate authorization gate |
| P3-17 Public pre-release publication authorization | Done | v0.1.0 was published as initial public helper pre-release; v0.1.1 supersedes it for post-validator-pass helper status |
P5 adds automation and machine-checkable helper gates around the public Proxy Benchmark Track helper surface.
This remains public-helper-only.
It is not benchmark validation.
It is not scientific validation.
It is not Sal-Meter validation.
It is not CAIS compliance.
It is not mediation validation.
It is not dyadic recovery validation.
It is not termination-gate accuracy validation.
It is not synthetic replay validation.
It is not certification.
It is not production readiness.
P4-4 adds a public phone-only simulator scaffold.
P4-5 adds a public synthetic session replay scaffold.
P4-4 and P4-5 are documentation and simulator / replay scaffolding only.
P4-4 is not currently part of the P5 helper-validation chain unless a later validator or lint step is added.
P4-5 is not currently part of the P5 helper-validation chain unless a later validator or lint step is added.
| Milestone | Status | Notes |
|---|---|---|
| P5-0 Boundary language lint | Done / advisory mode | evaluation-baseline/boundary_lint.py and evaluation-baseline/prohibited_terms.json are implemented; GitHub Actions runs the boundary lint step in advisory mode |
| P5-1 P3 helper-schema validator | Done / Passed | evaluation-baseline/validate_p3_schemas.py validates the synthetic P3 dyadic helper files against human_state_packet.schema.json, dyadic_session_event.schema.json, and benchmark_session.schema.json |
| P5-1 synthetic dyadic helper package | Done / Passed | sample-data/synthetic-dyadic-session-001/ contains human_state_packet_A.json, human_state_packet_B.json, dyadic_session_event.json, and benchmark_session_container.json |
| P4-0 synthetic dyadic demo-flow package | Done / Passed | sample-data/synthetic-dyadic-session-001/ contains ai_outputs.json, dyadic_delta.json, recovery_gate.json, termination_gate.json, and audit_log.json |
| P4-1 synthetic dyadic recovery delta evaluator | Done / Passed | evaluation-baseline/evaluate_dyadic_recovery_demo.py evaluates synthetic demo-flow consistency only |
| P4-2 mediation policy prompt pack | Done | prompts/ contains README.md and mediation_policy_v0.1.json; docs/mediation-policy-prompt-pack.md documents private cue, shared mediation output, false recovery prevention, and termination boundary logic |
| P4-3 synthetic termination-gate helper case package | Done / Passed | sample-data/synthetic-dyadic-session-001/termination_gate_cases.json contains synthetic pause, narrow, close, terminate, refresh, and audit-only helper cases |
| P4-3 termination gate demo evaluator | Done / Passed | evaluation-baseline/evaluate_termination_gate_demo.py evaluates synthetic termination-gate helper consistency only |
| P5-1 documentation alignment | Done | schemas/README.md, sample-data/README.md, evaluation-baseline/README.md, and root README.md explain P3 helper-schema validation as helper-structure validation only |
| P4-3 documentation alignment | Done | sample-data/README.md, evaluation-baseline/README.md, and root README.md explain P4-3 termination-gate helper evaluation as synthetic helper consistency only |
| P4-4 phone-only simulator scaffold | Present / documentation only | phone-only-simulator/ contains public-helper documentation and simulator scaffolding only; it is not a validator and is not production monitoring |
| P4-4 phone-only simulator README | Present / documentation only | phone-only-simulator/README.md defines folder boundary, public data boundary, P4-3 relationship, and final rule |
| P4-4 phone session flow wireframe | Present / documentation only | phone-only-simulator/session-flow-wireframe.md defines synthetic consent, packet check, AI output, delta review, recovery gate, termination gate, closure, and audit screens |
| P4-4 phone session state machine | Present / synthetic mockup only | phone-only-simulator/phone-session-state-machine.json defines synthetic-only states, allowed transitions, forbidden transitions, allowed decisions, prohibited decisions, and boundary flags |
| P4-4 sample phone session script | Present / synthetic script only | phone-only-simulator/sample-phone-session-script.md provides a synthetic sample phone-session script without real audio, real transcript, real participant data, Sal-Meter input, CAIS compliance dossier, or production intervention logic |
| P4-5 synthetic session replay scaffold | Present / documentation and JSON scaffold only | synthetic-session-replay/ contains public-helper documentation, replay manifest, replay event timeline, and replay boundary only; it is not a validator and is not real session replay |
| P4-5 synthetic replay README | Present / documentation only | synthetic-session-replay/README.md defines replay scaffold purpose, scope, intended files, public data boundary, P4-4 relationship, closed-session replay rule, and final rule |
| P4-5 synthetic replay manifest | Present / synthetic manifest only | synthetic-session-replay/replay-manifest.json defines replay source declaration, replay scope, boundary flags, replay flow, closed-session rule, allowed decisions, prohibited decisions, and success meaning |
| P4-5 synthetic replay event timeline | Present / synthetic timeline only | synthetic-session-replay/replay-event-timeline.json defines synthetic replay sequence from manifest loading through source declaration, consent, packet review, AI output, delta, recovery gate, termination gate, closure, and audit |
| P4-5 synthetic replay boundary | Present / documentation only | synthetic-session-replay/replay-boundary.md defines allowed replay materials, prohibited replay materials, prohibited replay claims, closed-session replay rule, replay interpretation, P4-4 relationship, and public release rule |
Current P5 helper-validation chain:
validate_sample_package.py
→ validate_p3_schemas.py
→ evaluate_dyadic_recovery_demo.py
→ evaluate_termination_gate_demo.py
→ boundary_lint.py
P4-4 is not currently included in the validation chain.
P4-5 is not currently included in the validation chain.
Current P4-4 scaffold files:
phone-only-simulator/
README.md
session-flow-wireframe.md
phone-session-state-machine.json
sample-phone-session-script.md
Current P4-5 scaffold files:
synthetic-session-replay/
README.md
replay-manifest.json
replay-event-timeline.json
replay-boundary.md
A successful P5 validation run means only:
The public synthetic/sample helper files follow the expected helper structure.
The P3 helper-schema objects follow expected helper-schema structure.
The P4-1 synthetic demo-flow objects preserve expected helper consistency.
The P4-3 synthetic termination-gate helper cases preserve expected helper consistency.
Wording boundary checks are clean.
A completed P4-4 scaffold means only:
The phone-only simulator scaffold is publicly documented.
The phone-only simulator files are synthetic-only.
The phone-only session flow is represented as a helper wireframe.
The phone-session state machine is a synthetic mockup.
The sample phone-session script is not a real transcript.
The closed-session rule is explicit.
The public data boundary is preserved.
A completed P4-5 scaffold means only:
The synthetic session replay scaffold is publicly documented.
The replay manifest is synthetic-only.
The replay event timeline is a synthetic structural review timeline.
The replay boundary is explicit.
The replay does not reopen a closed session.
The replay does not process real session data.
The replay does not process real phone recordings.
The replay does not process real call transcripts.
The public data boundary is preserved.
A successful run or completed scaffold does not mean:
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
synthetic replay validation
phone monitoring validation
Sal-Meter validation
CAIS compliance
clinical readiness
diagnostic readiness
therapeutic readiness
device readiness
production readiness
certification
phone monitoring authority
relationship verdict authority
human-ranking authority
production closed-loop authority
Correct boundary sentence:
The P5 helper-validation chain checks public helper structure, synthetic demo-flow consistency, synthetic termination-gate helper consistency, and wording hygiene only; P4-4 adds a synthetic phone-only simulator scaffold only, P4-5 adds a synthetic session replay scaffold only, and none of these create benchmark validation, mediation validation, dyadic recovery validation, termination-gate accuracy validation, replay validation, Sal-Meter validation, CAIS compliance, certification, phone monitoring authority, or production authority.
evaluation-baseline/
boundary_lint.py
prohibited_terms.json
validate_p3_schemas.py
evaluate_dyadic_recovery_demo.py
evaluate_termination_gate_demo.py
README.md
sample-data/
synthetic-dyadic-session-001/
README.md
human_state_packet_A.json
human_state_packet_B.json
dyadic_session_event.json
benchmark_session_container.json
ai_outputs.json
dyadic_delta.json
recovery_gate.json
termination_gate.json
audit_log.json
termination_gate_cases.json
These files support:
P3 helper-schema validation
P4-1 synthetic demo-flow consistency checking
P4-3 synthetic termination-gate helper consistency checking
boundary language linting
They do not support:
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
synthetic replay validation
Sal-Meter validation
CAIS compliance
clinical readiness
diagnostic readiness
therapeutic readiness
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
phone monitoring authority
production closed-loop authority
Correct boundary sentence:
Completed P5 helper-validation files support structure, schema, demo-flow, termination-gate helper, and wording checks only; they do not create evidence, validation, certification, Sal-Meter status, CAIS compliance, replay validation, phone monitoring authority, or production authority.
phone-only-simulator/
README.md
session-flow-wireframe.md
phone-session-state-machine.json
sample-phone-session-script.md
These files support:
phone-only simulator boundary documentation
synthetic phone-session flow wireframe
synthetic phone-session state-machine mockup
synthetic sample phone-session script
closed-session rule visibility
public data boundary visibility
P4-4 public-helper scaffold documentation
They do not support:
real phone monitoring
real phone recording
real transcript processing
real participant data processing
clinical intake
diagnosis
therapy
counseling
mediation-service operation
surveillance
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
Sal-Meter validation
CAIS compliance
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
P4-4 scaffold files must remain:
research-stage
public-helper-only
synthetic-only
non-clinical
non-diagnostic
non-therapeutic
non-counseling
non-surveillance
non-certification
non-human-ranking
not Sal-Meter
not CAIS compliance
not benchmark validation
not mediation validation
not dyadic recovery validation
not termination-gate accuracy validation
not phone monitoring authority
not production readiness
not production closed-loop
Correct boundary sentence:
Completed P4-4 public simulator scaffold files may demonstrate synthetic phone-only session structure only; they do not create evidence, validation, certification, phone monitoring authority, production authority, relationship verdicts, or human-ranking authority.
synthetic-session-replay/
README.md
replay-manifest.json
replay-event-timeline.json
replay-boundary.md
These files support:
synthetic session replay boundary documentation
synthetic replay manifest structure
synthetic replay event timeline structure
synthetic replay boundary rules
closed-session replay handling
audit-only replay posture
public data boundary visibility
P4-5 public-helper replay scaffold documentation
They do not support:
real session replay
real phone replay
real transcript replay
real participant data replay
raw human data replay
clinical replay
diagnostic replay
therapeutic replay
counseling replay
surveillance replay
production mediation replay
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
synthetic replay validation
phone monitoring validation
Sal-Meter validation
CAIS compliance
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
P4-5 scaffold files must remain:
research-stage
public-helper-only
synthetic-only
replay-scaffold-only
non-clinical
non-diagnostic
non-therapeutic
non-counseling
non-surveillance
non-certification
non-human-ranking
not real session replay
not real phone replay
not real transcript replay
not Sal-Meter
not CAIS compliance
not benchmark validation
not mediation validation
not dyadic recovery validation
not termination-gate accuracy validation
not synthetic replay validation
not phone monitoring authority
not production readiness
not production closed-loop
P4-5 replay scaffold files must not contain:
raw human data
identifiable human data
real participant data
real dyadic conflict records
real phone recordings
real call transcripts
real phone-session logs
private consent records
clinical records
health records
diagnostic labels
therapeutic recommendations
counseling notes
relationship verdicts
human scores
human-ranking outputs
raw biosignals
raw Sal-Meter traces
raw CAIS traces
CAIS compliance dossiers
production intervention logs
production monitoring logs
device-readiness evidence
production-readiness evidence
certification evidence
Correct boundary sentence:
Completed P4-5 public replay scaffold files may demonstrate synthetic session replay structure only; they do not create evidence, validation, certification, replay validation, phone monitoring authority, production authority, relationship verdicts, or human-ranking authority.
AI Output
→ Human-State Packet
→ Human-State Session Protocol
→ Dyadic Mediation Session Flow
→ Human-State Delta A/B
→ Dyadic Delta
→ Recovery Gate
→ Termination Gate
→ Consent and Data-Sharing Boundary
→ Session Closure
→ Audit Log
The Consent and Data-Sharing Boundary controls what may cross the arrows.
P3 defines the core helper architecture.
P4-4 does not replace this architecture.
P4-4 projects this architecture into a public-safe phone-only simulator scaffold.
P4-5 does not replace this architecture.
P4-5 projects this architecture into a public-safe synthetic replay scaffold.
P4-4 represents the same boundary logic through:
phone-only-simulator/
README.md
session-flow-wireframe.md
phone-session-state-machine.json
sample-phone-session-script.md
P4-5 represents replay review of the same boundary logic through:
synthetic-session-replay/
README.md
replay-manifest.json
replay-event-timeline.json
replay-boundary.md
The P4-4 phone-only simulator may demonstrate:
- consent-first session entry;
- packet availability checking;
- synthetic baseline state summary;
- synthetic AI output;
- synthetic Human-State Delta review;
- Recovery Gate placeholder;
- Termination Gate placeholder;
- closed-session handling;
- audit-log boundary.
The P4-5 synthetic session replay scaffold may demonstrate:
- replay manifest loading;
- replay source declaration;
- synthetic event timeline review;
- consent boundary review;
- packet boundary review;
- synthetic AI output replay;
- synthetic Human-State Delta replay;
- Recovery Gate replay;
- Termination Gate replay;
- closure replay;
- audit-only replay summary;
- closed-session replay handling.
The P4-4 phone-only simulator must not imply:
real phone monitoring
real phone recording
real transcript processing
real participant data processing
clinical intake
diagnosis
therapy
counseling
mediation-service operation
surveillance
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
Sal-Meter validation
CAIS compliance
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
The P4-5 synthetic session replay scaffold must not imply:
real session replay
real phone replay
real transcript replay
real participant data replay
raw human data replay
clinical replay
diagnostic replay
therapeutic replay
counseling replay
surveillance replay
production mediation replay
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
synthetic replay validation
phone monitoring validation
Sal-Meter validation
CAIS compliance
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
P4-5 must not reopen a closed session.
P4-5 must not continue mediation after closure.
P4-5 must not convert closure into recovery evidence.
P4-5 must not convert audit replay into certification.
Correct boundary sentence:
P4-4 is a phone-only public helper projection of the P3 session architecture, and P4-5 is a synthetic replay scaffold for reviewing that structure after representation; neither creates evidence, validation, certification, phone monitoring authority, production authority, relationship verdicts, or human-ranking authority.
A Human-State Packet is a minimal consent-bound, permission-bound, expiry-bound, confidence-aware, data-quality-aware, session-scoped, sharing-scoped, raw-data-excluding state-summary object.
It is not the body.
It is not diagnosis.
It is not Sal-Meter.
It is not CAIS compliance.
A Dyadic Session Event is a public-safe synthetic/sample event object that records boundary events such as consent, permission, packet status, sharing scope, private cue status, shared output status, Human-State Delta A/B, Dyadic Delta, gate decisions, closure, and audit status.
It records the boundary.
It does not record the body.
A Benchmark Session Container is a public-safe synthetic/sample container that connects event references, baseline suite status, gate summaries, leakage review, holdout strategy, audit status, public release status, authority status, and final boundary status.
It records the benchmark container.
It does not validate the benchmark.
AI Output
↓
Human-State Delta
↓
Dyadic Recovery
↓
Recovery Gate / Termination Gate
The system records what the AI generated.
Examples:
- generic AI output;
- state-aware AI output;
- private cue;
- shared mediation output;
- pause recommendation;
- clarification request;
- scope narrowing;
- recovery check;
- termination recommendation.
The system observes what changed after the AI output.
Examples:
- toward recovery;
- away from recovery;
- unchanged;
- mixed;
- uncertain;
- insufficient data;
- invalid.
Human-State Delta is not diagnosis.
It is not therapy.
It is not emotion reading.
It is not a human score.
It is a bounded benchmark observation.
The benchmark asks whether both sides of the dyad moved toward a session-defined recovery condition.
Recovery is not agreement.
Recovery is not silence.
Recovery is not obedience.
Recovery is not therapy.
Recovery is a bounded session-state condition where continued AI mediation can reduce, pause, or stop.
Recovery Gate asks whether the session-defined recovery condition has been reached.
It prevents false success.
It does not crown AI for speaking well.
It does not treat silence, obedience, agreement, synchrony, or one-sided improvement as automatic recovery.
Termination Gate asks whether the session must pause, narrow, or stop.
It prevents endless mediation.
It protects consent, permission, expiry, data quality, session scope, private state, raw human data, and auditability.
A closed session must stay closed.
The baseline ladder is:
| Level | Baseline | Question |
|---|---|---|
| B0 | Dummy / Chance Baseline | Can the model beat guessing or majority-class prediction? |
| B1 | Individual State Baseline | Can one person’s state alone explain the outcome? |
| B2 | Dyadic Relationship Baseline | Does the relation between both participants add explanatory value? |
| B3 | No-Intervention Baseline | Would the dyad recover naturally without AI intervention? |
| B4 | Generic AI Baseline | Is state-aware AI better than ordinary supportive AI output? |
| B5 | Rule-Based Mediation Baseline | Is the system better than fixed mediation scripts? |
| B6 | Human-State-Aware AI Mediation Model | Does packet-informed AI improve dyadic recovery under bounded conditions? |
| B7 | Recovery / Termination Gate Baseline | Can the system identify when to reduce, pause, or stop mediation? |
Primary outcome:
Dyadic Recovery Delta
Secondary outcomes may include:
- individual recovery direction;
- dyadic tension reduction;
- interruption reduction;
- turn-taking balance;
- mutual restatement success;
- recovery asymmetry;
- post-intervention stability;
- termination accuracy;
- mediation overstay rate;
- consent-boundary compliance;
- leakage-safe benchmark score;
- human non-judgment compliance.
This benchmark must be sensitive to false recovery.
A session is not successful merely because the AI sounded good.
A session is not successful merely because one participant became quiet.
A session is not successful merely because one participant reported relief.
A session is not successful merely because both participants showed synchrony.
A session is not successful if the AI continues after it should stop.
Failure conditions include:
- one participant improves while the other deteriorates;
- silence is misclassified as recovery;
- synchrony is treated as automatically positive;
- AI output quality is treated as sufficient evidence;
- generic supportive language is mistaken for human-state improvement;
- private state becomes exposed in shared output;
- packet permission is exceeded;
- expired packet is used;
- human score is generated;
- relationship verdict is generated;
- AI fails to stop when termination is required;
- leakage-safe holdout is not satisfied;
- model performance fails to exceed simpler baselines.
The dyad is the unit of interpretation.
One-sided improvement is not dyadic recovery.
The public benchmark must not exchange raw human data.
It should exchange only bounded summaries.
A Human-State Packet is:
minimal
consent-bound
permission-bound
expiry-bound
confidence-aware
data-quality-aware
session-scoped
sharing-scoped
raw-data-excluding
The packet is not the person.
The packet is not the body.
The packet is not the raw signal.
The packet is not diagnosis.
The packet is not a human score.
The packet is not a relationship judgment.
The packet is a minimal state-summary object for bounded interaction adjustment.
A session does not begin silently.
A session begins with consent.
A session runs only within packet permission.
A session closes through a recovery gate or termination gate.
A session that cannot close is not mediation.
It is surveillance drift.
A valid session should follow this structure:
Session Creation
→ Consent Confirmation
→ Packet Availability Check
→ Baseline State Summary
→ AI Output
→ Post-Output State Summary
→ Human-State Delta
→ Recovery Gate
→ Termination Gate
→ Session Closure
→ Audit Log
P4-4 projects this session principle into a phone-only public helper scaffold.
P4-5 projects this session principle into a synthetic replay scaffold.
The P4-4 phone-only simulator may represent the same session principle through:
phone-only-simulator/
README.md
session-flow-wireframe.md
phone-session-state-machine.json
sample-phone-session-script.md
The P4-5 synthetic session replay scaffold may represent the same session principle through:
synthetic-session-replay/
README.md
replay-manifest.json
replay-event-timeline.json
replay-boundary.md
In P4-4, the phone-only simulator may demonstrate:
- consent-first session entry;
- packet availability checking;
- synthetic baseline summary;
- synthetic AI output;
- synthetic Human-State Delta review;
- Recovery Gate placeholder;
- Termination Gate placeholder;
- closed-session handling;
- audit-log boundary.
In P4-5, the synthetic replay scaffold may demonstrate:
- replay manifest loading;
- replay source declaration;
- synthetic event timeline review;
- consent boundary review;
- packet boundary review;
- synthetic AI output replay;
- synthetic Human-State Delta replay;
- Recovery Gate replay;
- Termination Gate replay;
- closure replay;
- audit-only replay summary.
The phone-only simulator and replay scaffold must not process:
real phone calls
real audio
real transcripts
real participant data
real session records
identifiable human data
clinical data
Sal-Meter raw input
CAIS compliance dossiers
production intervention logs
The phone-only simulator and replay scaffold must not imply:
real phone monitoring
real session replay
real transcript replay
clinical intake
diagnosis
therapy
counseling
mediation-service operation
surveillance
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
synthetic replay validation
Sal-Meter validation
CAIS compliance
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
A closed session must stay closed.
A replay must not reopen a closed session.
A replay must not continue mediation after closure.
A replay must not generate new AI output after closure.
A replay must not convert closure into recovery evidence.
A replay must not convert audit into certification.
Correct boundary sentence:
The P4-4 phone-only simulator and P4-5 synthetic replay scaffold demonstrate the session principle as synthetic public helper flows only; they do not create evidence, validation, certification, phone monitoring authority, replay validation, production authority, relationship verdicts, or human-ranking authority.
sample-data/synthetic-session-001/
Required public helper files include:
session_metadata.json
streams_manifest.csv
events.csv
labels.csv
qc_report.json
features_baseline.csv
splits.json
operator_log.md
README.md
This package is checked by:
evaluation-baseline/validate_sample_package.py
sample-data/synthetic-dyadic-session-001/
Required public helper files include:
README.md
human_state_packet_A.json
human_state_packet_B.json
dyadic_session_event.json
benchmark_session_container.json
This package is checked by:
evaluation-baseline/validate_p3_schemas.py
P3 validation mapping:
human_state_packet_A.json
→ schemas/human_state_packet.schema.json
human_state_packet_B.json
→ schemas/human_state_packet.schema.json
dyadic_session_event.json
→ schemas/dyadic_session_event.schema.json
benchmark_session_container.json
→ schemas/benchmark_session.schema.json
sample-data/synthetic-dyadic-session-001/
Required public helper files include:
ai_outputs.json
dyadic_delta.json
recovery_gate.json
termination_gate.json
audit_log.json
This package is checked by:
evaluation-baseline/evaluate_dyadic_recovery_demo.py
sample-data/synthetic-dyadic-session-001/
Required public helper files include:
termination_gate_cases.json
This package is checked by:
evaluation-baseline/evaluate_termination_gate_demo.py
A successful P4-3 helper evaluation means only:
The synthetic termination-gate helper cases preserve expected public-helper consistency.
It does not mean:
termination-gate accuracy validation
dyadic recovery validation
mediation validation
benchmark validation
scientific validation
Sal-Meter validation
CAIS compliance
clinical readiness
diagnostic readiness
therapeutic readiness
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
phone-only-simulator/
Required public helper files include:
README.md
session-flow-wireframe.md
phone-session-state-machine.json
sample-phone-session-script.md
P4-4 is not stored under sample-data/.
P4-4 is a separate public simulator scaffold.
P4-4 may demonstrate:
- synthetic phone-only session structure;
- consent-first flow;
- packet availability check;
- synthetic baseline summary;
- synthetic AI output;
- synthetic Human-State Delta review;
- Recovery Gate placeholder;
- Termination Gate placeholder;
- closed-session handling;
- audit-log boundary;
- public-helper-only simulator posture.
P4-4 must not imply:
real phone monitoring
real phone recording
real transcript processing
real participant data processing
clinical intake
diagnosis
therapy
counseling
mediation-service operation
surveillance
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
Sal-Meter validation
CAIS compliance
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
synthetic-session-replay/
Required public helper files include:
README.md
replay-manifest.json
replay-event-timeline.json
replay-boundary.md
P4-5 is not stored under sample-data/.
P4-5 is a separate public replay scaffold.
P4-5 may demonstrate:
- synthetic session replay structure;
- replay manifest structure;
- replay source declaration;
- synthetic replay event timeline;
- consent boundary review;
- packet boundary review;
- synthetic AI output replay;
- synthetic Human-State Delta replay;
- Recovery Gate replay;
- Termination Gate replay;
- closure replay;
- audit-only replay summary;
- closed-session replay handling;
- public-helper-only replay posture.
P4-5 must not imply:
real session replay
real phone replay
real transcript replay
real participant data replay
raw human data replay
clinical replay
diagnostic replay
therapeutic replay
counseling replay
surveillance replay
production mediation replay
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
synthetic replay validation
phone monitoring validation
Sal-Meter validation
CAIS compliance
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
A synthetic replay may document a closed session.
A synthetic replay must not reopen a closed session.
A synthetic replay must not continue mediation after closure.
A synthetic replay must not convert closure into recovery evidence.
A synthetic replay must not convert audit into certification.
Public sample, simulator, and replay files must remain:
synthetic
sample
mock
placeholder
structure-only
non-identifying
raw-data-free
public-helper-only
non-clinical
non-diagnostic
non-therapeutic
non-counseling
non-surveillance
non-certification
non-human-ranking
not Sal-Meter
not CAIS compliance
not benchmark evidence
not mediation evidence
not dyadic recovery evidence
not termination-gate accuracy evidence
not synthetic replay validation
not phone monitoring authority
not production data
Public sample, simulator, and replay files must not include:
real raw human data
identity-bearing data
real dyadic conflict records
real session records
real phone recordings
real call transcripts
real transcript replay
clinical records
health records
raw biosignals
raw Sal-Meter traces
raw CAIS traces
private consent records
production intervention logs
relationship verdicts
human-ranking outputs
device-readiness claims
production-readiness claims
certification claims
termination-gate accuracy claims
synthetic replay validation claims
phone monitoring authority
Correct boundary sentence:
Synthetic sample packages, the P4-4 phone-only simulator scaffold, and the P4-5 synthetic replay scaffold may demonstrate public helper structure only; they do not create evidence, validation, certification, replay validation, phone monitoring authority, production authority, relationship verdicts, or human-ranking authority.
The GitHub Actions workflow is:
.github/workflows/validate-synthetic-sample.yml
Current intended workflow sequence:
Run synthetic sample package validator
Run P3 helper schema validator
Run P4 synthetic dyadic recovery demo-flow evaluator
Run P4 termination gate demo evaluator
Run boundary language lint
Validation helpers:
evaluation-baseline/validate_sample_package.py
evaluation-baseline/validate_p3_schemas.py
evaluation-baseline/evaluate_dyadic_recovery_demo.py
evaluation-baseline/evaluate_termination_gate_demo.py
evaluation-baseline/boundary_lint.py
The workflow successfully runs on the main branch.
This confirms only public helper-structure validation, synthetic demo-flow consistency, synthetic termination-gate helper consistency, and wording-boundary hygiene.
P4-4 currently adds documentation and simulator scaffold files only.
P4-5 currently adds documentation and replay scaffold files only.
P4-4 does not currently add a new validator.
P4-5 does not currently add a new validator.
P4-4 does not currently add a new GitHub Actions workflow step.
P4-5 does not currently add a new GitHub Actions workflow step.
Current P4-4 scaffold files:
phone-only-simulator/
README.md
session-flow-wireframe.md
phone-session-state-machine.json
sample-phone-session-script.md
Current P4-5 scaffold files:
synthetic-session-replay/
README.md
replay-manifest.json
replay-event-timeline.json
replay-boundary.md
The P4-4 scaffold may be reviewed by existing boundary-language lint if included in the lint scan path.
The P4-5 replay scaffold may be reviewed by existing boundary-language lint if included in the lint scan path.
If a later validator is added for P4-4 or P4-5, the workflow may be extended in a separate issue.
This workflow does not validate benchmark performance.
It does not validate scientific truth.
It does not validate mediation.
It does not validate dyadic recovery.
It does not validate termination-gate accuracy.
It does not validate synthetic replay.
It does not validate Sal-Meter.
It does not grant CAIS compliance.
It does not validate the P4-4 phone-only simulator.
It does not validate the P4-5 synthetic replay scaffold.
It does not certify phone monitoring.
It does not certify replay.
It does not certify any system, model, dataset, dashboard, laboratory, device, repository, schema, session protocol, implementation, mediation system, termination gate, phone-only simulator, replay scaffold, or closed-loop system.
It does not create clinical, diagnostic, therapeutic, counseling, surveillance, certification, device-readiness, production-readiness, relationship-verdict, phone-monitoring, replay-validation, production closed-loop, or human-ranking authority.
Correct boundary sentence:
The validation workflow checks public helper structure, synthetic demo-flow consistency, synthetic termination-gate helper consistency, and wording hygiene only; P4-4 currently adds phone-only simulator scaffold documentation only, P4-5 currently adds synthetic replay scaffold documentation only, and neither creates benchmark validation, mediation validation, dyadic recovery validation, termination-gate accuracy validation, replay validation, Sal-Meter validation, CAIS compliance, certification, phone-monitoring authority, or production authority.
Install dependencies:
pip install -r evaluation-baseline/requirements.txtRun validators:
python evaluation-baseline/validate_sample_package.py
python evaluation-baseline/validate_p3_schemas.py
python evaluation-baseline/evaluate_dyadic_recovery_demo.py
python evaluation-baseline/evaluate_termination_gate_demo.py
python evaluation-baseline/boundary_lint.pyExpected meaning of PASS:
The public synthetic/sample helper files follow the expected helper structure.
The P3 helper-schema objects follow expected helper-schema structure.
The P4-1 synthetic demo-flow objects preserve expected helper consistency.
The P4-3 synthetic termination-gate helper cases preserve expected helper consistency.
Wording boundary checks are clean.
P4-4 local status:
phone-only-simulator/README.md exists.
phone-only-simulator/session-flow-wireframe.md exists.
phone-only-simulator/phone-session-state-machine.json exists.
phone-only-simulator/sample-phone-session-script.md exists.
P4-5 local status:
synthetic-session-replay/README.md exists.
synthetic-session-replay/replay-manifest.json exists.
synthetic-session-replay/replay-event-timeline.json exists.
synthetic-session-replay/replay-boundary.md exists.
P4-4 currently has no separate local validator.
P4-5 currently has no separate local validator.
P4-4 currently has no separate GitHub Actions validation step.
P4-5 currently has no separate GitHub Actions validation step.
P4-4 is documentation and simulator scaffolding only.
P4-5 is documentation and replay scaffolding only.
P4-4 files may be reviewed manually for boundary consistency.
P4-5 files may be reviewed manually for boundary consistency.
P4-4 files may be scanned by the boundary language lint if the lint path includes the phone-only-simulator/ folder.
P4-5 files may be scanned by the boundary language lint if the lint path includes the synthetic-session-replay/ folder.
If a later P4-4 or P4-5 validator is added, it should be added in a separate issue.
PASS does not mean:
benchmark validated
scientific truth validated
mediation validated
dyadic recovery validated
termination-gate accuracy validated
phone-only simulator validated
synthetic replay validated
phone monitoring validated
Sal-Meter validated
CAIS compliant
clinical evidence
diagnostic evidence
therapeutic evidence
device-ready
production-ready
certified
relationship verdict authority
human-ranking authority
production closed-loop authority
Correct boundary sentence:
Local validation checks helper structure, synthetic demo-flow consistency, synthetic termination-gate helper consistency, and wording hygiene only; P4-4 currently adds phone-only simulator scaffold documentation only, P4-5 currently adds synthetic replay scaffold documentation only, and neither creates evidence, validation, certification, replay validation, phone monitoring authority, Sal-Meter status, CAIS compliance, or production authority.
This repository must not contain:
- raw human data;
- identifiable human data;
- private participant data;
- real dyadic conflict records;
- real session records;
- real phone recordings;
- real call transcripts;
- real transcript replay;
- real phone-session logs;
- consent forms with identifiers;
- private session logs;
- raw biosignal files from real participants;
- raw Sal-Meter traces;
- raw CAIS traces;
- private labels;
- hidden ground-truth labels;
- clinical interpretations;
- diagnostic interpretations;
- therapeutic interpretations;
- counseling interpretations;
- person ranking;
- human ranking;
- relationship verdicts;
- relationship scoring outputs;
- employment, insurance, legal, educational, or eligibility decisions;
- surveillance or coercive monitoring materials;
- phone monitoring authority;
- replay validation authority;
- real-time monitoring authority;
- device-readiness claims;
- production-readiness claims;
- certification claims;
- production closed-loop claims;
- termination-gate accuracy claims;
- dyadic recovery validation claims;
- mediation validation claims;
- synthetic replay validation claims;
- benchmark validation claims;
- scientific validation claims;
- Sal-Meter validation claims;
- CAIS compliance claims.
Public sample, helper, simulator, and replay files must remain:
synthetic
sample
mock
placeholder
structure-only
non-identifying
raw-data-free
public-helper-only
non-clinical
non-diagnostic
non-therapeutic
non-counseling
non-surveillance
non-certification
non-human-ranking
not Sal-Meter
not CAIS compliance
not benchmark evidence
not mediation evidence
not dyadic recovery evidence
not termination-gate accuracy evidence
not synthetic replay validation
not phone monitoring authority
not replay validation authority
not production data
P4-3 termination-gate helper cases may demonstrate:
- pause-session examples;
- narrow-scope examples;
- close-session examples;
- terminate-session examples;
- consent-refresh examples;
- packet-refresh examples;
- audit-only examples;
- closed-session handling;
- permission-expiry handling;
- low-confidence handling;
- insufficient-data-quality handling;
- private-state exposure risk handling;
- one-sided improvement caution.
P4-4 phone-only simulator scaffold files may demonstrate:
- synthetic phone-only session structure;
- consent-first flow;
- packet availability check;
- synthetic baseline summary;
- synthetic AI output;
- synthetic Human-State Delta review;
- Recovery Gate placeholder;
- Termination Gate placeholder;
- closed-session handling;
- audit-log boundary;
- public-helper-only simulator posture.
P4-5 synthetic session replay scaffold files may demonstrate:
- synthetic session replay structure;
- replay manifest structure;
- replay source declaration;
- synthetic replay event timeline;
- consent boundary review;
- packet boundary review;
- synthetic AI output replay;
- synthetic Human-State Delta replay;
- Recovery Gate replay;
- Termination Gate replay;
- closure replay;
- audit-only replay summary;
- closed-session replay handling;
- public-helper-only replay posture.
P4-3 termination-gate helper cases must not imply:
real mediation accuracy
validated termination-gate accuracy
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
Sal-Meter validation
CAIS compliance
clinical readiness
diagnostic readiness
therapeutic readiness
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
P4-4 phone-only simulator scaffold files must not imply:
real phone monitoring
real phone recording
real transcript processing
real participant data processing
clinical intake
diagnosis
therapy
counseling
mediation-service operation
surveillance
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
Sal-Meter validation
CAIS compliance
phone monitoring authority
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
P4-5 synthetic session replay scaffold files must not imply:
real session replay
real phone replay
real transcript replay
real participant data replay
raw human data replay
clinical replay
diagnostic replay
therapeutic replay
counseling replay
surveillance replay
production mediation replay
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
synthetic replay validation
phone monitoring validation
Sal-Meter validation
CAIS compliance
device readiness
production readiness
certification
relationship verdict authority
human-ranking authority
production closed-loop authority
A synthetic replay may document a closed session.
A synthetic replay must not reopen a closed session.
A synthetic replay must not continue mediation after closure.
A synthetic replay must not generate new AI output after closure.
A synthetic replay must not convert closure into recovery evidence.
A synthetic replay must not convert audit into certification.
Correct boundary sentence:
Public data in this repository may demonstrate helper structure, synthetic consistency, phone-only simulator scaffolding, and synthetic replay scaffolding only; it must not create evidence, validation, certification, replay validation, phone monitoring authority, production authority, relationship verdicts, or human-ranking authority.
All issues and pull requests must preserve the repository boundary.
Contributions must not claim or imply:
- benchmark validation;
- scientific validation;
- mediation validation;
- dyadic recovery validation;
- termination-gate accuracy validation;
- phone-only simulator validation;
- synthetic replay validation;
- phone monitoring validation;
- Sal-Meter validation;
- CAIS compliance;
- diagnostic status;
- clinical status;
- therapeutic status;
- counseling-service status;
- legal mediation authority;
- surveillance readiness;
- phone monitoring authority;
- replay validation authority;
- device readiness;
- production readiness;
- certification;
- production deployment;
- production closed-loop authority;
- human ranking;
- relationship verdict;
- relationship scoring;
- official consciousness measurement;
- ground-truth human-state truth measurement.
Issues and pull requests may propose or modify:
- public helper documents;
- synthetic sample structures;
- schema helper structures;
- synthetic demo-flow objects;
- synthetic termination-gate helper cases;
- phone-only simulator scaffold files;
- synthetic phone-session wireframes;
- synthetic phone-session state-machine mockups;
- synthetic sample phone-session scripts;
- synthetic session replay scaffold files;
- synthetic replay manifests;
- synthetic replay event timelines;
- synthetic replay boundary documents;
- validation helper scripts;
- wording-boundary lint rules;
- documentation alignment;
- release-boundary notes.
Issues and pull requests must not introduce:
raw human data
identifiable human data
clinical data
real session records
real phone recordings
real call transcripts
real participant data
real consent records
real phone-session logs
real transcript replay
Sal-Meter raw input
CAIS compliance dossier
benchmark validation claim
scientific validation claim
mediation validation claim
dyadic recovery validation claim
termination-gate accuracy validation claim
phone-only simulator validation claim
synthetic replay validation claim
phone monitoring authority claim
replay validation authority claim
device-readiness claim
production-readiness claim
certification claim
relationship verdict authority
human-ranking authority
production closed-loop authority
A valid issue or pull request may improve helper structure.
A valid issue or pull request may improve boundary clarity.
A valid issue or pull request may improve synthetic consistency checks.
A valid issue or pull request may improve termination-gate helper case coverage.
A valid issue or pull request may improve phone-only simulator scaffold clarity.
A valid issue or pull request may improve synthetic phone-session flow representation.
A valid issue or pull request may improve synthetic session replay scaffold clarity.
A valid issue or pull request may improve synthetic replay event ordering.
A valid issue or pull request may improve closed-session replay handling.
A valid issue or pull request must not convert this repository into:
an evidence system
a certification system
a production system
a clinical system
a diagnostic system
a therapeutic system
a counseling system
a surveillance system
a real phone monitoring system
a real session replay system
a real transcript replay system
a relationship-verdict system
a human-ranking system
a Sal-Meter validation system
a CAIS compliance system
Correct boundary sentence:
Issues and pull requests may improve public helper structure, synthetic termination-gate cases, phone-only simulator scaffolding, and synthetic replay scaffolding, but they must not create evidence, validation, certification, replay validation, phone monitoring authority, production authority, relationship verdicts, or human-ranking authority.
Dashboard mockups in this repository are public helper structures only.
They may present bounded synthetic/sample helper fields for demonstration.
They may show:
- synthetic session identifiers;
- synthetic packet availability status;
- synthetic confidence fields;
- synthetic data-quality fields;
- synthetic Human-State Delta summaries;
- synthetic Dyadic Delta summaries;
- synthetic Recovery Gate status;
- synthetic Termination Gate status;
- synthetic pause / narrow / close / terminate examples;
- synthetic audit status;
- synthetic public-boundary flags;
- synthetic phone-only simulator state;
- synthetic phone-session flow status;
- synthetic phone-session state-machine status;
- synthetic phone-session closure status;
- synthetic replay manifest status;
- synthetic replay event timeline status;
- synthetic replay boundary status;
- synthetic replay closure status;
- synthetic audit-only replay status.
They must not present:
- person scores;
- diagnosis;
- treatment guidance;
- counseling guidance;
- clinical interpretation;
- employment or insurance eligibility;
- legal eligibility;
- educational eligibility;
- surveillance status;
- phone monitoring status;
- real-time monitoring status;
- real phone recording status;
- real transcript status;
- real session replay status;
- real phone replay status;
- real transcript replay status;
- replay validation status;
- relationship verdicts;
- relationship scoring;
- human ranking;
- psychological safety score;
- certified status;
- validated benchmark status;
- validated mediation status;
- validated dyadic recovery status;
- validated termination-gate accuracy status;
- validated phone-only simulator status;
- validated synthetic replay status;
- device-readiness status;
- production-readiness status;
- production closed-loop status;
- Sal-Meter output;
- CAIS compliance.
A dashboard may show bounded synthetic/sample helper fields for demonstration.
A dashboard may show P4-4 phone-only simulator scaffold status only as synthetic helper structure.
A dashboard may show P4-5 synthetic replay scaffold status only as synthetic helper structure.
A dashboard must not show real call monitoring.
A dashboard must not show real phone audio status.
A dashboard must not show real transcript processing.
A dashboard must not show real session replay.
A dashboard must not show real transcript replay.
A dashboard must not show real participant state.
A dashboard must not show phone monitoring authority.
A dashboard must not show replay validation authority.
It must not become a judgment engine.
It must not become a monitoring engine.
It must not become a phone monitoring engine.
It must not become a replay validation engine.
It must not become a clinical engine.
It must not become a mediation-service engine.
It must not become a relationship-verdict engine.
It must not become a human-ranking engine.
It must not become a production closed-loop intervention engine.
Correct boundary sentence:
A dashboard mockup may display public helper structure, synthetic phone-only simulator scaffold status, and synthetic replay scaffold status, but it must not create evidence, validation, certification, replay validation, phone monitoring authority, production authority, relationship verdicts, or human-ranking authority.
Closed-loop demo-lite files are local placeholder structures only.
They may demonstrate:
- synthetic event-log shape;
- synthetic feedback-loop boundary fields;
- placeholder routing logic;
- pause-session examples;
- narrow-scope examples;
- close-session examples;
- terminate-session examples;
- audit-only examples;
- public-helper-only closure logic.
P4-4 phone-only simulator files may demonstrate:
- synthetic phone-session flow structure;
- synthetic phone-session state-machine structure;
- synthetic sample phone-session script structure;
- consent-first phone-only session entry;
- packet availability check;
- synthetic Human-State Delta review;
- Recovery Gate placeholder;
- Termination Gate placeholder;
- session closure;
- audit-log boundary.
P4-5 synthetic replay scaffold files may demonstrate:
- synthetic replay manifest structure;
- synthetic replay event timeline structure;
- synthetic replay boundary structure;
- replay source declaration;
- consent boundary review;
- packet boundary review;
- synthetic AI output replay;
- synthetic Human-State Delta replay;
- Recovery Gate replay;
- Termination Gate replay;
- closure replay;
- audit-only replay summary;
- closed-session replay handling.
They do not define a production closed-loop intervention system.
They do not authorize real-time human monitoring.
They do not authorize phone monitoring.
They do not authorize real phone recording.
They do not authorize real transcript processing.
They do not authorize real session replay.
They do not authorize real phone replay.
They do not authorize real transcript replay.
They do not authorize replay validation.
They do not authorize automated intervention on real participants.
They do not validate mediation.
They do not validate recovery.
They do not validate dyadic recovery.
They do not validate termination-gate accuracy.
They do not validate the phone-only simulator.
They do not validate the synthetic replay scaffold.
They do not validate Sal-Meter.
They do not grant CAIS compliance.
They do not certify device readiness.
They do not certify production readiness.
They do not create clinical, diagnostic, therapeutic, counseling, legal mediation, employment, insurance, educational, eligibility, surveillance, phone-monitoring, replay-validation, relationship-verdict, production closed-loop, or human-ranking authority.
Closed-loop demo-lite, P4-4 phone-only simulator, and P4-5 synthetic replay scaffold files must not contain:
raw human data
identifiable human data
clinical data
real session records
real phone recordings
real call transcripts
real transcript replay
real participant data
real consent records
real phone-session logs
Sal-Meter raw input
CAIS compliance dossier
real-time monitoring authority
phone monitoring authority
replay validation authority
automated intervention authority
benchmark validation claim
scientific validation claim
mediation validation claim
dyadic recovery validation claim
termination-gate accuracy validation claim
phone-only simulator validation claim
synthetic replay validation claim
device-readiness claim
production-readiness claim
certification claim
relationship verdict authority
human-ranking authority
production closed-loop authority
A closed session must stay closed.
A replay must not reopen a closed session.
A replay must not continue mediation after closure.
A replay must not convert closure into recovery evidence.
A replay must not convert audit into certification.
Correct boundary sentence:
Closed-loop demo-lite, P4-4 phone-only simulator, and P4-5 synthetic replay scaffold files may demonstrate placeholder helper structure only; they must not create evidence, validation, certification, replay validation, phone monitoring authority, monitoring authority, production authority, relationship verdicts, or human-ranking authority.
The next roadmap should move from synthetic replay scaffolding toward public helper demo package review and optional lint extension.
Recommended next milestones:
| Milestone | Name | Purpose |
|---|---|---|
| P4-6 | Public Helper Demo Package Review | Review synthetic demo packages, simulator scaffolds, and replay scaffolds for public-boundary consistency before any future release |
| P4-7 | Phone-only / Replay Boundary Lint Extension | Consider extending boundary-language lint coverage to phone-only-simulator/ and synthetic-session-replay/ if needed |
| P4-8 | Public Helper Release Readiness Note | Prepare a bounded release-readiness note only after P4-6 review and any needed lint extension are complete |
Completed helper-validation and P4 helper milestones are tracked under:
Current P5 helper-validation state
Implementation status table
Completed P5 helper-validation files
Completed P4-4 public simulator scaffold files
Completed P4-5 public replay scaffold files
Synthetic sample packages
Validation workflow
Local validation
Completed P4 helper items include:
P4-0 synthetic dyadic demo-flow package
P4-1 synthetic dyadic recovery demo-flow evaluator
P4-2 mediation policy prompt pack
P4-3 synthetic termination-gate helper case package
P4-3 termination gate demo evaluator
P4-4 phone-only simulator scaffold
P4-4 phone-only session flow wireframe
P4-4 synthetic phone-session state-machine mockup
P4-4 synthetic sample phone-session script
P4-5 synthetic session replay scaffold
P4-5 synthetic replay manifest
P4-5 synthetic replay event timeline
P4-5 synthetic replay boundary document
Current P4-4 scaffold files:
phone-only-simulator/
README.md
session-flow-wireframe.md
phone-session-state-machine.json
sample-phone-session-script.md
Current P4-5 scaffold files:
synthetic-session-replay/
README.md
replay-manifest.json
replay-event-timeline.json
replay-boundary.md
Future roadmap items must remain:
research-stage
public-helper-only
synthetic-first
non-clinical
non-diagnostic
non-therapeutic
non-counseling
non-surveillance
non-certification
non-human-ranking
not Sal-Meter
not CAIS compliance
not benchmark validation
not scientific validation
not mediation validation
not dyadic recovery validation
not termination-gate accuracy validation
not synthetic replay validation
not phone monitoring authority
not replay validation authority
not device readiness
not production readiness
not production closed-loop
Future roadmap items must not introduce:
raw human data
identifiable human data
clinical data
real session records
real phone recordings
real call transcripts
real participant data
real consent records
real phone-session logs
real transcript replay
Sal-Meter raw input
CAIS compliance dossier
benchmark validation claim
scientific validation claim
mediation validation claim
dyadic recovery validation claim
termination-gate accuracy validation claim
phone-only simulator validation claim
synthetic replay validation claim
phone monitoring authority claim
replay validation authority claim
device-readiness claim
production-readiness claim
certification claim
relationship verdict authority
human-ranking authority
production closed-loop authority
P4-6 review may check:
- public helper file completeness;
- synthetic-only status;
- boundary-language consistency;
- closed-session handling;
- replay does not reopen closure;
- simulator and replay folders remain outside
sample-data/; - root README alignment;
- issue checklist alignment;
- Actions PASS status.
P4-6 review must not become:
benchmark validation
scientific validation
mediation validation
dyadic recovery validation
termination-gate accuracy validation
synthetic replay validation
Sal-Meter validation
CAIS compliance
device-readiness review
production-readiness review
certification review
Correct boundary sentence:
Future roadmap items may extend public helper review, synthetic replay scaffolding, simulator boundary coverage, and optional lint hygiene, but they must not create evidence, validation, certification, replay validation, phone monitoring authority, production authority, relationship verdicts, or human-ranking authority.
This repository does not attempt to:
- prove consciousness;
- measure consciousness directly;
- infer emotions;
- diagnose mental state;
- treat or counsel people;
- rank persons;
- judge relationships;
- produce relationship verdicts;
- produce human-ranking outputs;
- replace human consent;
- expose raw human data;
- process identifiable human data;
- publish clinical data;
- process real phone calls;
- process real phone recordings;
- process real call transcripts;
- process real phone-session logs;
- process real session records;
- replay real sessions;
- replay real phone calls;
- replay real transcripts;
- create phone monitoring authority;
- create replay validation authority;
- authorize real-time phone monitoring;
- validate the phone-only simulator;
- validate the synthetic replay scaffold;
- validate Sal-Meter;
- define CAIS compliance;
- validate benchmark performance;
- validate scientific truth;
- validate mediation;
- validate dyadic recovery;
- validate termination-gate accuracy;
- certify any system;
- certify device readiness;
- certify production readiness;
- operate a production mediation service;
- operate a production phone-monitoring service;
- operate a production replay service;
- operate a production closed-loop intervention system;
- authorize surveillance;
- authorize real-time monitoring;
- authorize automated intervention on real participants.
This repository may support:
public helper documentation
synthetic sample structure
schema helper structure
synthetic demo-flow consistency checks
synthetic termination-gate helper consistency checks
synthetic phone-only simulator scaffolding
synthetic phone-session flow representation
synthetic phone-session state-machine mockups
synthetic sample phone-session scripts
synthetic session replay scaffolding
synthetic replay manifest structure
synthetic replay event timeline structure
synthetic replay boundary documentation
boundary-language hygiene
repository-level transparency
This repository must not become:
a clinical system
a diagnostic system
a therapeutic system
a counseling system
a surveillance system
a real phone monitoring system
a real session replay system
a real transcript processing system
a replay validation system
a relationship-verdict system
a human-ranking system
a production closed-loop system
a certified benchmark system
a Sal-Meter validation system
a CAIS compliance system
Correct boundary sentence:
This repository is a public helper surface; it does not create evidence, validation, certification, replay validation, phone monitoring authority, production authority, relationship verdicts, or human-ranking authority.
Unless otherwise stated, public helper materials in this repository are released under:
Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)
Document-level license statements in DOI-registered canonical records remain fixed by those records.
Please cite DOI-registered records as the authority layer.
This GitHub repository is a helper surface.
DOI records govern.
GitHub helps.
See:
CITATION.cff
This repository documents structure.
It does not validate the body.
It does not validate the person.
It does not validate the relationship.
It does not validate a human state.
It does not validate dyadic recovery.
It does not validate termination-gate accuracy.
It does not validate the phone-only simulator.
It does not validate the synthetic replay scaffold.
It does not validate Sal-Meter.
It does not grant CAIS compliance.
It does not crown a benchmark as validated.
It does not validate mediation.
It does not certify any system.
It does not certify any model.
It does not certify any dataset.
It does not certify any dashboard.
It does not certify any laboratory.
It does not certify any device.
It does not certify device readiness.
It does not certify production readiness.
It does not authorize surveillance.
It does not authorize diagnosis.
It does not authorize therapy.
It does not authorize counseling.
It does not authorize legal mediation.
It does not authorize relationship verdicts.
It does not authorize human ranking.
It does not authorize phone monitoring.
It does not authorize real-time monitoring.
It does not authorize real phone recording.
It does not authorize real transcript processing.
It does not authorize real session replay.
It does not authorize real phone replay.
It does not authorize real transcript replay.
It does not authorize replay validation.
It does not authorize production mediation.
It does not authorize production closed-loop intervention.
A closed session must stay closed.
A replay must not reopen a closed session.
A replay must not continue mediation after closure.
A replay must not generate new AI output after closure.
A replay must not convert closure into recovery evidence.
A replay must not convert audit into certification.
The packet is not the person.
The event is not the relationship.
The container is not the truth.
The demo-flow is not recovery.
The termination-gate case is not accuracy evidence.
The phone-only simulator is not the phone call.
The sample phone-session script is not a transcript.
The phone-session state machine is not authority.
The replay skeleton is a map of a map.
The replay manifest is not a session.
The replay event timeline is not the event.
The replay boundary is not authority.
The validator is not authority.
The evaluator is not proof.
The workflow is not certification.
The repository is a map.
It is not the mountain.