TechDex AI Framework ™ v1.1

Overview

Version 1.1 of the TechDex AI Framework ™ matures its governed cognitive architecture through stronger MiniBrain coordination, expanded language comprehension, more reliable conversational continuity, stricter source and identity boundaries, and improved business-facing deployment capabilities. It is the current published documentation line for the Framework. For the completed Version 1.0 history, see the Version 1.0 Patch Notes.

Key Enhancements

• Public Search and AI Discovery Restored - Replaced the live site's blanket crawler block with a selective policy that keeps the Framework homepage, sales pages, public documentation, sitemaps, and LLM guidance files discoverable while excluding application, API, administration, installation, membership, registration, CGI, and form-processing surfaces.
• Technical Wiki URL Repair - Corrected five structured-data references to the existing technical wiki and its current section anchors, and added a permanent redirect for the previously advertised nonexistent URL.

• Resumable Transactional Onboarding - Onboarding now consumes natural answers to the exact field it requested, completes its isolated account-or-no-account transaction, and returns any remaining thought to the normal cognitive pipeline. The required first name remains mandatory, optional email remains declinable, and onboarding gains no authority over the resumed conversation.
• Evidence-Aware Article Continuity - A qualified article object now retains its bounded excerpt and evidence primitives across turns. When asked what the source actually establishes, the configured intelligence can distinguish article identity and retained evidence from unverified body claims, figures, authorship, or publication metadata instead of collapsing into a generic refusal.
• Satisfied Conversational Cessation - Expanded Human Speech Assist with general spoken-English forms that indicate the person has enough for the present and is ending the current exchange. Expressions such as “I’ll leave it there for now” and “I’ve got enough to go on” can now conclude naturally without creating a market rule or allowing the language observation to decide the route, source, answer, or speaker.
• Common Speech Interpretation Baseline - Matured the existing Human Speech Assist within Pragmatics so ordinary spoken English can be represented as bounded observations before Semantics and MiniBrain determine meaning and action. The externally audited baseline covers conversational response tokens, satisfied closure, indirect advice requests, reformulation, vague collections, substitution, common problem language, and reasoning-oriented multiword verbs without creating a new routing or decision layer.
• Client-Controlled Onboarding Allowance - Each installation continues to choose immediate onboarding, a configurable number of anonymous substantive exchanges, or no automatic onboarding. In delayed mode, the configured allowance is now applied exactly: the separately paced name prompt becomes eligible after the final allowed answer without counting greetings, thanks, social acknowledgments, satisfied closure, or commands as questions.

User Experience Improvements

• Improved interpretation of declarative questions, compound requests, corrections and ordinary references while preserving productive disagreement and honest limitations.

• Complete Requests and Useful Disagreement - Improved preservation of compound requests and current needs. The Framework may question a factual premise while retaining the requested scope, ordering and deliverable.
• Conversation Continuity - Improved correction handling, recognition of the Framework's own disclosure within its limits, and persistence of explicit closure without disabling productive disagreement.

• Business-First Homepage Positioning - Reworked the public homepage around the needs of business owners and decision-makers evaluating a private intelligence service. The surrounding sales copy now leads with easier access to company knowledge, reliable answers, useful decision support, and reduced staff effort while preserving the detailed capability lists, technical comparison tables, architecture, governance distinctions, deployment facts, and documentation links as supporting proof. Search and AI-retrieval signals retain the full TechDex AI Framework entity name, Governed Tier IV Cognitive Intelligence classification, current software version, and a direct plain-language definition near the top of the page.

Content Interaction Enhancements

• Preserved structured source identity and partial-result availability through more response and session handoffs. First-party provenance is distinct from whether a requested field is available.
• Extended testing still finds incorrect source recovery after correction; source continuity is not yet accepted as reliable across extended research.

• Qualified Partial Results - Release 1.1.12 preserves currently qualified article titles and links when another requested attribute remains unverified. It states the limitation without promoting proposed evidence or presenting the unfinished task as complete. Live acceptance is pending.

Core Architectural Infrastructure

• Strengthened MiniBrain-governed operation contracts, evidence handoff, request roles, source establishment and committed undertaking lifecycle transitions. Specialists report signals; MiniBrain adjudicates state.

• MiniBrain Undertaking Coordination - Extended reconciliation to consult Quest Log history, interpret specialist repair signals, and distinguish unresolved work from stopped work. The Ledger remains an inert record; MiniBrain coordinates and Identity speaks.
• Shared Governance and Evaluation Philosophy - Recorded the shared governance commitment and the distinction between task authority and revisable informational claims. Conversation quality, task outcome, retained defects and successful repair are evaluated separately.

• Present-Turn Discourse Coordination - MiniBrain now authorizes the conversational-maintenance response path from the current utterance's resolved discourse act. Stale or inherited request state can no longer overrule a correctly represented accepted next action, while mixed plan-and-question utterances continue through the full cognitive pipeline. Identity retains exclusive speaking authority.
• Onboarding Transaction Boundary - The onboarding sandbox recognizes only its bounded name, optional-email, verification, and PIN transactions. Non-onboarding content is quarantined as a resumable remainder and cannot be interpreted, routed, answered, or stored as onboarding knowledge.
• Qualified Source Evidence Retention - Article promotion now carries the accepted source object's bounded excerpt, evidence primitives, and available metadata into Source Continuity. The retained object remains evidence rather than cognition and reports absent evidence as absent instead of manufacturing support.
• Registry-Governed Closure Observation - Satisfied cessation is represented through the shared Pragmatic Registry and evaluated against the complete utterance. A phrase that sounds final cannot close a turn when the person also supplies a continuing request, preserving MiniBrain’s authority over completion and keeping Human Speech Assist observational.
• Mention-versus-Request Semantic Contract - A noun such as email, author, date, location, or URL may now remain a mentioned subject unless its grammatical relationship actually asks the Framework to retrieve that attribute. Advice, comparison, confirmation, and conditional structures retain their complete meaning instead of being collapsed into a scalar lookup.
• Canonical Multiword Predicate Representation - Grammar now represents common phrasal and multiword verbs as one predicate, including bounded adverb and object insertion such as “focus next on,” “break this down,” and “walk me through.” The canonical predicate is supplied to Semantics and Pragmatics as structural evidence only and cannot select an operation, source, route, or answer.
• Post-Answer Onboarding Coordination - Delayed onboarding remains a separate interaction-control event and now returns to MiniBrain after the governed answer is complete. An active request keeps priority during cognition but no longer starves an already-due, separately paced onboarding prompt; pending, failed, or still-reconciling responses continue to block the event.
• Discourse-Aware Completion Coordination - MiniBrain now distinguishes active requests from acknowledgment, accepted next actions, intentional closure, and satisfied completion before deciding whether a response may conclude the current transaction. Open Quest Ledger observations preserve progress without forcing a person to continue an objective they have completed, deferred, or chosen to leave.
• Bounded Response Reconciliation Loop - A response candidate that genuinely conflicts with an earlier phase decision now receives one internal repair pass and returns to MiniBrain for a second completion review. Earlier route or evidence decisions remain important signals but cannot become permanent vetoes after the present response resolves the operation honestly.
• Constitutional Pipeline Enforcement - Activated MiniBrain's transaction authority across routing and response completion so the Framework's established governance is enforced as the operating architecture rather than interpreted as advisory guidance. Specialists remain capable coordinating partners, but only MiniBrain authorizes the whole cognitive action and only Identity speaks.
• MiniBrain-Owned Route Reconciliation - Analysis scope, first-party locator entitlement, generic-concept scope, contact evidence, article prerequisites, and tenant-topic relevance now arrive as bounded proposals. MiniBrain accepts or rejects them and publishes the only downstream-visible intent and route.
• Single Scoped Source Verdict - Source Governance now issues an integrity-protected verdict for the exact article object and request scope it qualified. Retrieval, recovery, metadata, synthesis, and presentation carry or validate that verdict instead of independently re-adjudicating relevance, while missing, stale, altered, or cross-scope verdicts fail inert.
• Pre-Commit Cognitive Boundary - MiniBrain and Identity now authorize a response before source state, answer projection, cache, interaction logs, or conversation history can observe it. An incomplete candidate keeps the transaction open and cannot be recorded as completed meaning.
• Universal Identity Projection - Every answer-generating model call receives the one-governed-speaker contract, and deterministic source-cardinality findings are returned to Identity for expression rather than speaking from Source Presentation. Pending model replies also use the governed response path instead of writing directly to transport.

Stability & Polish

• Versioned Documentation Separation - Separated the completed 1.0 release history from the current 1.1 documentation line, added dedicated 1.1 patch notes and changelog files, and retained the 1.0 files as historical records. This is documentation maintenance and does not create a new Framework runtime or database release.

• Framework 1.1.27 passed 48 runtime regression suites plus installer and updater contracts. These checks are separate from live conversational acceptance.
• A 15-substantive-turn conversation retained the revised goal, topic return and useful explanation, but failed to recover the requested original source. The research draft preserves both findings.

• Source and Disposition Hardening - Hardened repeat projection of governed source envelopes, rejected stale or contradictory source identities, and preserved present-turn corrections and stopping signals. These are capability repairs rather than scenario-specific answers.
• Conversation Acceptance and Judge Calibration - Updated simulator evaluation to distinguish useful repaired conversations from persistent failures. The 1.1.11 review covered 66 turns; five main conversations received three passes, one bounded pass and one partial. Remaining defects are retained separately. The 1.1.12 local build passes 32 Framework regression suites; live improvement remains unverified.
• Operation-Scoped Source Projection (1.1.8) - Released an initial MiniBrain operation-scoping repair. Local regressions passed, but the exact live replay on all four updated sites reproduced the compound-request source defect. The fresh randomized run also found conversation and source-handling failures. This release has not passed acceptance for closing the source-projection repair; later releases continued this repair; the current release and acceptance status are recorded below.
• Architecture Documentation Alignment - Reconciled contributor guidance, architecture references, and current status notes with the existing MiniBrain coordination, scoped evidence qualification, and Identity-only expression boundaries. Updated public definitions and the generated Framework Self-Reference package. This is documentation maintenance, not a new runtime release.

• Natural Onboarding Answer Recovery - Corrected repeated name and optional-email prompts when a person supplied the requested answer together with a continuing statement. The transaction now consumes only its own answer and resumes the remainder once.
• Progress Acknowledgment Recovery - Corrected a downstream override that could turn a valid accepted-next-action observation into a generic source refusal. MiniBrain's present-turn coordination now reaches the response boundary without allowing Language or the Ledger to authorize completion.
• Precise Article Evidence Limits - Article follow-ups can now state exactly what the retained qualified object supports and what still requires verification, preserving useful continuity without converting a title, link, or excerpt into proof of the full article.
• ASCII Common-Speech Coverage - Added a dedicated suite of ordinary keyboard-English closure and continuation fixtures, including contractions, thanks, coordinated requests, and topic-preserving contrasts. Typographic apostrophes and dashes are normalized to the same observation, but are not required for comprehension. The complete 29-file Framework regression catalog passes.
• Conversational Advice Classification - Ordinary questions such as whether to focus on email and SMTP setup rather than changing a form now remain advisory decisions with their condition and contrast intact. Explicit requests for an email address, phone number, author, date, location, URL, headline, lede, or byline continue to resolve as attribute requests.
• Language-Layer Coverage Validation - Added a dedicated Common Speech regression suite covering the tenant-neutral registry, authority boundaries, multiword predicates, mention-versus-request pairs, and the original SMTP advisory failure. The complete 28-file Framework regression catalog passes.
• Natural Social Acknowledgments - Short acknowledgment forms are recognized through the shared Grammar registry as conversational maintenance, preventing a completed explanation from being unnecessarily reopened with repeated advice or warnings.
• Internal Control-Signal Isolation - The pre-commit boundary no longer turns preserve_and_replan into repetitive user-facing speech. Reconciliation remains internal; if bounded repair cannot complete the operation, Identity expresses only the actual evidence limitation in natural language.

Ready for Client Deployment

• Framework 1.1.27 is centrally published and owner-confirmed installed. Extended source-object reliability remains under development.

• Framework 1.1.12 Published - The update and fresh installer are available following local regression and package verification. Owner installation and live conversation testing are reserved for the next session. No database schema change is required.

TechDex AI Framework ™ v1.1 is the current published framework line. Framework 1.1.27 is centrally published and owner-confirmed installed; extended source-object reliability remains under development and is not represented as complete acceptance.

Whats Next

Continue MiniBrain-governed requested-source-object establishment and preservation across discovery, correction, topic return, delivery and completion. Retain the separate hypothetical/request-executability and author/date limitations, and assign repairs from the first demonstrated divergence. Microphone continuity, expanding input and thinking indicators remain future interface work. Existing longer-term work includes capability and market packs, broader work planning, governed BigQuery evidence, report downloads, user-owned files and sharing, employee messaging and profiles, and the future Customer Service module.

Release Information

• Session closed under the owner-designated date 09-12-2026. Releases 1.1.13 through 1.1.27 were published during this session; the current release is 1.1.27. No additional runtime release was created at closeout.

Framework application 1.1.12 is published as release 112 with verified update and fresh-installer packages. It follows the 1.1.10 coordination and 1.1.11 scope/continuity releases. The owner confirmed installation of the preceding build before the 1.1.11 acceptance run; installation of 1.1.12 and its live replay/fresh-conversation acceptance remain pending for the next session. Local validation includes 32 Framework suites, 24 installer checks and 153 updater checks. No database schema changes are involved. All work in this closeout belongs to the owner-designated September 11 session.

Framework application 1.1.7 supersedes 1.1.6 before client rollout with one source-continuity refinement: authorized full-article analysis remains the first path whenever the article body is available. Identity expresses the retained qualified excerpt and exact verification limits only when that authorized body recovery fails, so evidence continuity improves without reducing existing article-analysis capability. No database schema version changes.

Framework application 1.1.6 repairs the downstream contracts isolated by Human Simulator Test Thirty-Five without expanding the Language registry. The onboarding sandbox now consumes natural name and optional-email answers, preserves any non-onboarding remainder, and resumes it only after the isolated transaction completes. MiniBrain's present-turn discourse decision governs whether Identity may acknowledge an accepted next action, preventing inherited request state from forcing a source refusal while preserving mixed plan-and-question obligations. Qualified article anchors retain bounded excerpts and evidence primitives so Identity can state exactly what the source establishes and what remains unverified. The first-name requirement remains mandatory once onboarding activates, optional email remains optional, the Quest Ledger remains observation-only, Identity remains the sole speaker, and neither database schema version changes.

Framework application 1.1.5 expands Human Speech Assist with a reusable action-ready commitment family. First-person future forms such as choosing a next step or stating an immediate plan are now represented as progress observations instead of being mistaken for unsupported knowledge requests. The implementation recognizes the grammatical and pragmatic construction rather than enumerating task-specific phrases; complete-utterance guards preserve coordinated questions and conditional requests, while stative prefaces and third-party future statements remain distinct. The Conversation Quest Ledger may retain the progress signal, but it receives no authority to decide, route, complete, or speak. MiniBrain remains the cognitive coordinator, Identity remains the sole speaker, market packs remain deferred, and neither database schema version changes.

Framework application 1.1.4 expands the existing Human Speech Assist with tenant-neutral satisfied-cessation language and dedicated ASCII-only regression coverage. The Framework can now recognize that expressions such as “I’ll leave it there for now” end the present exchange while preserving a request that continues in the same utterance. Smart punctuation normalizes to the same observation, but ordinary ASCII keyboard input is the tested baseline. This remains a Pragmatics observation with no route, source, truth, completion, answer, or speech authority; MiniBrain still coordinates completion and Identity remains the sole speaker. Market packs remain deferred, mg retrieval is intentionally unchanged for this isolation test, and neither database schema version changes.

Framework application 1.1.3 matures the existing Human Speech Assist rather than adding another cognitive authority. A tenant-neutral Pragmatic Registry now supplies bounded observations for ordinary spoken-English structures, while Grammar represents common multiword predicates and Semantics distinguishes an attribute being mentioned from that attribute actually being requested. Advice, contrast, condition, reformulation, closure, substitution, vague collections, and common problem language can therefore inform the complete structural model of the utterance without choosing intent, route, source, action, answer, or speaker. General English remains in the core Framework; industry language remains reserved for future market packs, and installation-specific language remains tenant-owned. Identity remains the sole speaker, MiniBrain remains the cognitive coordinator, and neither database schema version changes.

Framework application 1.1.2 coordinates the configurable onboarding gateway with the matured MiniBrain response lifecycle. Installations retain independent control over immediate, delayed, or disabled onboarding and the number of anonymous substantive exchanges permitted in delayed mode. The configured allowance now becomes due after the final allowed answer rather than one turn later; greetings, thanks, brief acknowledgments, satisfied closure, and commands do not consume it; and an active request no longer starves the separately paced event after its governed answer has completed. Pending, failed, or unresolved responses still block onboarding, the first-name requirement remains mandatory once activated, optional email remains declinable, and short social acknowledgments now close naturally without repeated advice. Identity remains the sole speaker, MiniBrain remains the event coordinator, and neither database schema version changes.

Framework application 1.1.1 matures the active MiniBrain completion boundary exposed by Human Simulator Test Twenty-Nine. MiniBrain now reconciles the current discourse act, the person's objective disposition, the response candidate's declared coverage, remaining obligations, and earlier phase decisions before authorizing final projection. Satisfied closures and accepted next actions can conclude naturally without allowing open Quest Ledger observations to control the conversation; active first-pass conflicts receive one bounded internal repair and a second MiniBrain review; and preserve_and_replan can no longer leak into Identity speech. Earlier grounding decisions remain phase-specific coordination signals rather than permanent vetoes, specialists remain bounded partners, the Ledger remains observation-only, Identity remains the sole speaker, and neither database schema version changes.

Framework application 1.1.0 activates the constitutional cognitive transaction. Pre-routing heuristics now report bounded proposals for MiniBrain to reconcile; Source Governance issues one integrity-protected, request-scoped article verdict that later consumers may validate but not repeat; and the shared response path obtains MiniBrain completion authorization plus Identity projection before it can update source state, memory projections, cache, logs, or conversation history. Vetoed candidates retain the open objective instead of being recorded as complete, answer-generating model calls carry the one-governed-speaker contract, deterministic source constraints return to Identity for expression, and pending replies no longer bypass the governed response boundary. Specialists retain authority over their bounded work, the Quest Ledger remains observation-only, and neither database schema version changes.