Field Watch Signal Room

Feature-first and supportive copy can improve prototype legibility while preserving verified behavior, exact governed language, and evidence boundaries.

v0.1

Turn behaviorally complete prototype copy into feature-first, supportive, evidence-bounded language while preserving the artifact’s verified behavior and claim ceiling.

A prototype works, yet its descriptive copy foregrounds implementation or prohibitions more strongly than features, affordances, desired state, and evidence-calibrated possibility.

Language and presentation skill. Product claims, domain decisions, external actions, and publication remain with their accountable owners. Exact source text, code, accessibility labels, governed language, and material safety or evidence boundaries retain fidelity.

First material use: USIP-PROT-01 v0.2.0. Behavior QA passed; desktop and mobile visual QA passed; zero user-facing bare-negation matches remained; the code-only CSS :not selector remained unchanged.

Runnable artifact; feature and affordance inventory; audiences; claim ceiling; exact source/code/accessibility/governed language; supportive-change and negation-aware guidance; behavior and visual QA routes.

The entry is self-contained. The canonical private skill contract remains in the authorized resolver; the public Site Reading carries the artifact and receipt.

Revised artifact copy, polarity receipt, behavior QA, visual QA, claim ceiling, and bounded re-entry condition.

State the receiving outcome; inventory affordances; load the governing register; rewrite the semantic center; preserve material boundaries; scan polarity; rerun behavior and visual QA; write the receipt.

@driftframe/skills/refine-evidence-bounded-prototypes-supportive-semantic-attractors

Restore any softened material boundary and place its supported replacement state beside it. If feature language implies performance, return to the claim ceiling. If scanability falls, shorten supporting prose before removing the feature model.

A reader names the features first; affordance verbs describe available actions; boundaries include a constructive route; behavior is unchanged; desktop and mobile copy fit; the close names a bounded return.

A useful then-and-now reading treats technological eras as overlapping layers, distinguishes model families, and keeps adoption, architecture, performance, and governance claims at their own evidence standards.

v0.1

Compose a dated then-and-now Field Watch reading that treats technological eras as overlapping layers, distinguishes model families, and keeps every claim at its fitting evidence class.

A current shift needs a concise dated signal before a deeper synthesis, especially when earlier infrastructure persists and AI categories or evidence classes are being blended.

The skill produces a dated comparative reading and evidence routes. Source owners, accountable practitioners, and human publishers retain deployment-claim, consequential-interpretation, and release authority.

First material use: the published US insurance-movement comparative reading, dated 2026-09-24 and reviewable 2027-03-24, distinguishes eight layers and links to a canonical Generated Research synthesis.

As-of date and review horizon; concrete decision surface; official sources; survey or adoption evidence with population; model research; canonical research destination; audience and publication standing.

The entry is self-contained and links to public-possible Registry pages and public sources. The canonical private skill remains in the authorized resolver.

One-line signal, overlapping-era map, layer distinctions, affordance translation, decision implications, evidence boundaries, source-of-record route, machine node, and review date.

Name the signal; define overlapping eras; separate technical layers; classify claims; translate into affordances; state decision implications; keep system-specific claims local; connect the canonical spine; set the review date; close with a return.

@driftframe/skills/compose-evidence-bounded-then-and-now-field-watch-capsules

When the capsule drifts toward trend summary, restore the decision surface. When the timeline compresses the continuing substrate, restore what continues. When categories blur, return to constitutive mechanisms. When a statistic carries excess meaning, restate population, method, date, and adjacent open claim.

Readers describe change as accumulated capability across a continuing substrate; AI categories stay distinct; numbers carry population and date; named deployments require direct evidence; the entry stands alone; the research spine and review trigger are one step away.

One canonical evidence spine plus distinct signal, architecture, situation, judgment, and skill surfaces improves discovery while controlling duplication and claim drift.

v0.1

Turn deep technical research into one canonical evidence spine and a small constellation of distinct, cross-linked Registry surfaces for human discovery and AI retrieval.

A source-rich research family has public-possible value and needs clear source ownership, audience-fitting projections, AI-readable routes, and one bounded chain of action.

The skill can research, compare plans, create self-contained projections after audience authorization, and braid routes. Source owners retain private-material, publication, domain-claim, and consequential-decision authority.

First end-to-end use: five plans compared under one probability event; the selected braided constellation created a Generated Research spine, Field Watch reading, Core Architecture, Site Reading and prototype, Practitioner Probe, and three skill projections.

Source family and access standing; Registry map; public evidence; human and AI retrieval intents; claim classes and dates; publication authority; completion evidence.

The entry is self-contained and links to public-possible Registry outputs. The canonical private skill and private source lineage remain in the authorized resolver.

Probability-bounded plan, selected chain, canonical synthesis, distinct projections, navigation braid, and verification receipt.

Define the probability event; map independent retrieval intents; choose the source of record; compare distinct plans; select one chain; adapt across the audience edge; braid routes; add machine nodes; verify links, evidence, dates, and duplication.

@driftframe/skills/braid-deep-technical-research-into-registry-projection-plans

When pages repeat the same detail, restore the canonical spine and shorten projections to their reader jobs. When a public surface depends on private context, make it self-contained or hold it. When the chain becomes machinery, return to the smallest structure that changes discovery and decision quality.

Every claim has a canonical owner; each projection has a distinct job; human and AI routes converge on the same evidence; private locators stay bounded; machine nodes and links agree; updates have a clear first destination.

Generative AI adds retrieval, synthesis, explanation, and orchestration to a continuing substrate of transactions, rules, documents, statistical models, and accountable human review.

v1.0

A team is comparing administrative-AI approaches, interpreting insurer AI adoption, planning prior-authorization interoperability, or deciding where document vision, classical ML, language models, SLMs, MoE, or agents fit.

A dated comparative reading for public use. It supports architecture and workflow decisions; named deployment, performance, fairness, savings, payment, and adoption claims require system-specific evidence and accountable review.

Applied to a verified v0.2.0 simulated prototype, one Registry projection plan, and the accompanying architecture, Site Reading, and judgment probe.

The entry is self-contained and links only to public-possible Registry destinations and public sources. Private episode packets and locators remain in the authorized insurance loamthread.

A current signal, dated 2023–2027 capability timeline, overlapping-era flowchart, layer distinctions, decision questions, evidence routes, and direct continuation to the canonical Generated Research synthesis.

Read the eras as overlapping; locate the layer carrying the present decision; identify its evidence object and authority owner; follow the canonical synthesis for depth; define a local comparison before making a deployment claim.

@driftframe/field-watch/us-insurance-movement-generative-ai-2026-09-24

Return to the layered map whenever the narrative compresses the continuing substrate, deployment evidence, or measured-performance boundary. Restore the source-specific claim and route the open question to a local evaluation.

A reader can distinguish transaction, perception, prediction, generation, orchestration, and authority layers; name the evidence each needs; and reach the canonical synthesis in one step.

The costliest failures in AI-assisted work are fluent, confident, and structurally invisible to the person receiving them.

v0.1

The entry carries its complete diagnostic catalog and paired desired-state responses within the entry.

Style guidance reaches one layer of four. A first measurement shows where the reach stops and where it holds.

v0.1

The entry carries its complete measurement account, transfer boundary, and receiving recommendations within the entry.

The drafting entry carries its current analysis within the entry. Named sources remain prose until the bibliography and citation pass supplies their public reference routes.

Senior human scanners, automated intake systems, and personal AI interpreters often benefit from the same design property: specific, well-labeled claims placed beside inspectable evidence.

0.3.0

Preparing a principal+ portfolio, site-reading collection, case study, proposal, vendor submission, grant surface, or other artifact that a time-constrained human, automated intake system, or person’s AI may read.

The entry offers a design synthesis. Local reader research establishes local scan behavior; hiring, legal, accessibility, and risk owners retain their fitting decisions; publication requires separate human authorization.

Built from a 40-query first pass and 72-source expansion, then refreshed with information-foraging research, bounded scan-time estimates, AI hiring research, and current regulatory routes as_of 2026-09-23. A final negation-aware and supportive-change review preserved direct evidence boundaries, local validation, and human authority.

Audience and decision; principal+ body of work; strongest evidence; role and authority; systems and constraints; current public sources; local reader or parser evidence; publication boundary.

Self-contained public-possible guidance with public research and regulatory sources plus Practitioner Registry routes. Private hiring material and source lineage stay with their authorized owners.

A research timeline, scan-mode and commit-mode model, three-tier reading model, specificity revision pass, five phrasing moves, stable parser-facing structure, claim-to-evidence adjacency, AI hydration packet, eleven-point verification, and local validation plan.

Classify the receiving human, parser, and personal-AI reader jobs; front-load role, system, consequence, authority, and strongest evidence; use stable headings and explicit labels; preserve depth for sustained review; treat published scan timings as source-specific estimates; measure continuation, field recovery, summary fidelity, citation fidelity, and voice preservation.

two-reader-specificity-convergence

When scan-time evidence or parser behavior varies, preserve the structural principle and measure the receiving context. When specificity becomes keyword density or performative certainty, restore role, system, consequence, evidence, uncertainty, and human authority.

A senior reviewer can classify the work and reach its strongest evidence quickly; an automated parser recovers stable headings, roles, systems, dates, and outcomes; a person’s AI preserves voice, citations, confidence, and boundaries; local reader and intake evidence can revise the design.

Persona and role-language choices become governable when teams compare controlled conditions and keep human reliance effects separate from generated-output effects.

v0.1-pilot

Compare controlled persona and role-language conditions and return evidence about task behavior, calibrated reliance, delegation, responsibility, tool use, and recovery.

A consequential naming, persona, register, or role-language decision depends on a local behavioral or reliance effect that literature or intuition cannot settle for the deployed system and task.

The candidate designs and reports bounded comparisons. Human owners retain experiment approval, participant protection, deployment, publication, and consequential interpretation authority. Automatic discovery and durable promotion remain unapproved.

The candidate has a complete experimental contract derived from the naming research wave. It remains unexercised; one natural-use comparison and an independent cold-start replication are required before durable promotion.

The decision, affected population, task, model/version and settings, source packet, tools, candidate framing conditions, human and model-output measures, run budget, approval, and evaluation owner.

This internal candidate entry is self-contained for planning. The candidate contract, evidence atlas, and research relations remain in authorized Notion context; publication and promotion remain separate human decisions.

An experiment contract, controlled run matrix, evidence table, bounded recommendation, limitations, open transfer claims, and return condition.

Name the decision, population, model/version, task, and consequence; separate human and model-output hypotheses; select minimal framing conditions; hold content and configuration constant; define scoring and exclusions; run or stage the comparison; report uncertainty; issue a bounded recommendation and open transfer claims.

@driftframe/field-watch/evaluate-persona-reliance-effects

Return an unrun experiment contract when valid execution is unavailable. Narrow claims when sample size or independence is weak. Route provenance-only or wording-only decisions to calibration or translation.

The decision, population, model/version, controlled variables, human-facing measures, model-output measures, run count, uncertainty, limitations, and open transfer claims remain visible; null and mixed results remain reportable.

A coherent voice can remain useful without becoming an identity or authority claim when persona, register, role, permissions, and accountability stay separate.

v0.1

Separate product identity, interaction alias, local role, register, prompt persona, persona vector, permissions, authority, and accountability before designing or evaluating behavior.

A discussion, prompt, interface, policy, or evaluation treats persona, style, identity, role, permissions, authority, agency, accountability, or persona-vector language as interchangeable.

The profile maps constructs and their owners. Human stewards retain identity, representation, access, policy, deployment, publication, and consequential decision authority. Model-specific mechanistic findings retain their method and configuration limits.

The current research wave and existing register binding supplied a complete distinction map. Material use beyond the formation context remains a future receipt.

The terms, prompt, interface, policy, or evaluation claim; model and configuration; evidence type; local roles; permissions; authority; representation standing; accountable owner.

This public-possible account is self-contained. The owning binding, evidence atlas, and private lineage remain in authorized Notion context; publication remains a separate human decision.

A construct-and-referent map with term, construct, referent, evidence type and scope, owner, consequence if conflated, and smallest clarification.

Collect consequence-bearing terms; assign or mark each construct; name evidence type and scope; identify owner and consequence; split collapsed schema fields or claims; add the smallest clarification; preserve alternative framings where evidence remains open.

@driftframe/field-watch/separate-persona-register-identity-authority

When one term legitimately spans several constructs, retain it and add structured explanation. Route empirical uncertainty to the evaluation candidate rather than forcing a single label.

Product identity, interaction alias, local role, register, prompt persona, persona vector, permissions, authority, and accountability each retain a recoverable definition, evidence scope, owner, and consequence.

A name can support continuity without becoming an authority claim when product identity, local role, permissions, evidence horizon, and accountable human ownership remain separate.

v0.1

Choose or review names, aliases, role metaphors, provenance, authority, evidence horizon, and accountable human ownership for a generative system.

A generative system arrives already named, or a team is choosing a product label, interaction alias, role metaphor, or work-arrangement term that may blur capability, authority, provenance, or accountability.

The skill prepares a naming and role-language contract. Product owners and accountable humans retain naming, deployment, access, policy, publication, and consequential decision authority. A name or persona does not expand permissions.

The current naming research wave established a direct but context-bound naming evidence route, an already-named-system pattern, and the first complete identity-contract specification. Natural-use comparison remains open.

Provider, product, model/version, instance, local work arrangement, desired relational value, permissions, tools, source coverage, audience constraints, review requirement, and accountable owner.

This entry is self-contained. Canonical skill, evidence atlas, and research relations remain in authorized Notion context; publication remains a separate human decision.

A generating-system identity contract and a bounded recommendation to retain, qualify, change, or test the name and role language.

Preserve provider, product, model/version, and instance identity; name the local work role; separate the interaction alias; inspect role-metaphor expectations; declare permissions and evidence horizon; name human accountability; test comprehension and reliance; issue the compact contract.

@driftframe/field-watch/calibrate-ai-names-role-language

Use a structural identifier when provenance is missing. Preserve materially different role-label options for the accountable owner. Route behavioral uncertainty to a controlled evaluation rather than asserting a mechanism.

Product, model/version, instance, local role, interaction alias, role metaphor, permissions, evidence horizon, review requirement, and accountable human remain distinct and recoverable.

Anthropomorphic effects are usually produced by cue bundles, contexts, and expectations rather than by one word alone; a useful audit keeps each layer visible before changing the interface.

v0.1

Inspect anthropomorphic cue bundles and distinguish human-interpretation effects, model-output hypotheses, evidence standing, authority implications, and the smallest useful change.

Names, pronouns, avatars, voice, emotional language, role metaphors, agency claims, customization, turn-taking, or tool surfaces may shape reliance, delegation, capability expectations, responsibility, or recovery.

The skill diagnoses and recommends. Human owners retain interface design, deployment, access, publication, consequential interpretation, and accountability. The audit does not infer consciousness, identity, emotion, or universal effects from a name alone.

The current naming and anthropomorphism research wave established the independent diagnostic job, a bounded evidence spine, and a first Registry receiving surface. Natural-use and independent-receiver evidence remain open.

The interface, prompt, transcript, policy, or workflow; affected people and setting; task, consequence, authority, and recovery path; intended relational value; available behavioral and design evidence.

This entry is self-contained. Canonical skill, evidence atlas, and research relations remain in authorized Notion context; publication remains a separate human decision.

An anthropomorphic cue ledger with cue bundle, intended relational value, human-interpretation effect, model-output hypothesis, evidence standing, consequences, authority implication, and proportionate next move.

Inventory social and anthropomorphic cues; identify bundles; separate observed human effects, measured behavior, model-output hypotheses, and unsupported inference; locate consequences; calibrate evidence; select the smallest useful retain, qualify, separate, disclose, test, or revise move; name the accountable receiver.

@driftframe/field-watch/audit-anthropomorphic-cues

When evidence is thin, preserve the cue and open consequence, then choose a reversible test or disclosure. Route final wording to the translation owner and causal comparison to the evaluation candidate.

Cue bundles remain visible; human-facing effects and model-output hypotheses stay distinct; evidence standing, affected population, authority, and the smallest useful change are recoverable.

Relational language can stay warm and precise when local anchors make the acting system, evidence, authority, and accountable human recoverable.

v0.1

Translate anthropomorphic language so useful relational value remains while provenance, system state, evidence, authority, and human accountability become recoverable.

Anthropomorphic, relational, or person-like language supports welcome, continuity, or coordination while leaving system state, provenance, evidence, permission, ownership, authority, or accountability ambiguous.

The profile revises task-bearing language while preserving quoted, cited, participant-supplied, legally meaningful, historically significant, and intentionally authored source form. People retain interpretive, representation, legal, publication, and consequential decision authority.

The profile extends an established referent-resolution practice into the current naming and anthropomorphism wave. It has a complete output contract; a distinct natural-use receipt remains open.

The passage, prompt, interface text, or policy; intended reader job; source and authorship standing; relational value; consequence-bearing referents; provenance, evidence, permission, authority, and accountability boundaries.

This public-possible account is self-contained. The owning binding and private lineage remain in authorized Notion context; publication remains a separate human decision.

Revised copy plus a compact change register naming the intended reader job, relational value preserved, technical state made visible, authority clarified, source-fidelity note, and return route.

Identify the expression and its practical job; preserve its relational value; locate consequential ambiguity; add the smallest fitting referent, provenance, or authority anchor; retain source-bearing language; review cadence and texture; return revised copy with a change register.

@driftframe/field-watch/translate-anthropomorphic-language

Preserve uncertain or source-bearing language, name the materially different readings, and route the question to the fitting human receiver. Use the cue audit when the need is diagnosis rather than copy.

Useful warmth and metaphor remain; consequence-bearing actors, provenance, system state, evidence, permissions, authority, and accountability are recoverable; quoted and participant-supplied language stays faithful.

AI talent-agent intake becomes governable when each automated stage carries a dated capability claim, fitting evidence, affected-population view, candidate route, and accountable human decision owner.

0.2.0

Translate fast-changing talent-agent, screening, evidence, and regulatory signals into a reviewable decision chain with accountable human authority.

Comparing AI recruiting vendors, designing automated intake or screening, reviewing candidate-facing automation, establishing human oversight, or translating current legal and research signals into an enterprise decision.

This reading supports evaluation and governance design. Employers, counsel, compliance leaders, risk owners, hiring authorities, accessibility owners, and affected candidates retain their fitting legal, policy, selection, accommodation, appeal, and publication roles.

Synthesizes current official guidance, a peer-reviewed matched-resume audit, two bounded hiring and recruiter surveys, and current vendor documentation checked on 2026-09-23. Source classes remain distinct, a source-internal survey discrepancy stays visible, and local deployment evidence remains required.

Current vendor documentation; model, product, and workflow versions; hiring stage; affected populations; criteria; local performance and subgroup evidence; accessibility and accommodation routes; applicable law; decision owner; review date.

Self-contained public-possible comparative reading with public official, research, survey, and vendor sources. Candidate data, private evaluations, model traces, audit workpapers, and private source lineage stay with their authorized owners.

A dated market and regulatory timeline, hiring-side and candidate-side ecosystem map, decision-chain map, evidence-class ledger, human-authority checkpoint design, AI-collaborator hydration packet, regulatory review route, and currentness triggers.

Name the hiring decision and affected population; map the system’s task and output; classify each claim as official, research-supported, vendor-reported, survey-directional, locally observed, or proposed; design the human checkpoint and candidate correction route; verify current law and product behavior; test outcomes; set as_of and review dates.

ai-talent-agent-intake-human-authority-2026-09-23

When product behavior, legal status, audit currency, or evidence changes, mark the affected claim stale and reopen review. Keep consequential decisions with accountable people, preserve candidate correction and accommodation routes, and pause the affected automation when the evidence or authority boundary cannot support responsible use.

The team can separate source-reported capability, observed workflow performance, research evidence, legal duty, and policy proposal; every consequential intake stage names its accountable human authority, candidate route, as_of date, and review trigger.

Across Anthropic, OpenAI, Microsoft, Google, and NIST, agent-skill evaluation is converging on versioned scenarios, observable traces, decomposed outcome and process evidence, lifecycle monitoring, and explicit governance.

0.1.0

Translate current vendor and standards guidance into a source-attributed, reviewable, vendor-neutral agent-skill evaluation pattern.

Selecting or governing an agent-skill evaluation approach, translating vendor documentation into a portable enterprise pattern, or reviewing whether current tests still fit changed models, workflows, tools, or risk.

This entry reports and interprets current official guidance. Security, legal, compliance, procurement, domain, and risk owners retain endorsement and deployment authority; comparable metrics, universal sample sizes, and certification claims require evidence and review designed for those decisions.

Synthesizes current official documentation across four vendor ecosystems and NIST, checked on 2026-09-23, with research-backed cautions about repetition, independence, reliability, and claim transfer. Product-level superiority requires comparative evidence designed for that claim.

Current official documentation; product and model versions; intended skill and workflow; failure modes; task population; tool surface; risk tolerance; review owner; deployment and publication boundaries.

Self-contained public-possible comparative reading. Direct references point to public official documentation and Practitioner Registry pages; private traces, credentials, internal evaluations, and workspace locators remain with their owners.

A dated vendor-and-standards crosswalk, portable evaluation stack, source-versus-interpretation boundary, review triggers, and a set of enterprise adoption questions.

Confirm the as_of date; identify the receiving decision; separate source-reported product behavior from practitioner interpretation; compare discovery, outcome, trajectory, tool, security, lifecycle, and monitoring guidance; select metrics from consequential failure modes; record version, environment, thresholds, and limitations; schedule review.

agent-skill-evaluation-vendor-crosswalk-2026-09-23

If a feature, metric, preview status, URL, or lifecycle recommendation changes, mark the affected row stale and revisit the source. When products use different units or graders, retain a qualitative crosswalk and reserve numeric comparison for commensurate evidence. Calibrate confidence to the available implementation detail.

A team can trace each recommendation to its source, distinguish final outcome from trajectory and tool process, keep vendor heuristics separate from statistical claims, version the skill and evaluation suite together, and reopen review when guidance or operating conditions change.

A first skill receipt establishes a bounded baseline; later receipts preserve only the evidence that confirms, strengthens, narrows, contradicts, or supersedes that claim.

0.1.0

Preserve how evidence changes the supported claim for a reusable skill across first use, repeat use, re-entry, version change, and transfer.

A skill is first received, reused after a material change, resumed after interruption, evaluated across runs or contexts, revised, promoted, narrowed, contradicted, superseded, or retired—and the evidence can change a later decision.

The method structures evidence and claim movement. Accountable humans retain safety, quality, compliance, publication, deployment, individual-performance, risk, and promotion judgments; statistical-independence and population-effectiveness claims require studies designed to support them.

Formed from recent skill cultivation and a comparative review of vendor lifecycle guidance, NIST measurement practice, agent-reliability research, repeated-run evidence, pseudoreplication, and decision-record design. External-team transfer remains open.

Receiving decision; stable skill and version identity; receiver and relation; platform binding; task and environment; available evaluation evidence; prior receipt; authority and disclosure boundaries.

Self-contained public-possible guidance with direct routes to Practitioner Registry pages and public sources. Private skill locators, traces, participants, and operational records stay with their authorized systems.

A compact receipt with relation scope, encounter kind, evidence role, observation structure, independence class, claim effect, claim ceiling, prior link, supersession state, and one return condition.

Name the receiving decision; identify skill, version, receiver, binding, and environment; classify first, repeat, or re-entry for an explicit relation; route run-level traces to the evaluation system; state the evidence delta, independence class, claim effect, ceiling, and open claims; link rather than overwrite prior receipts; stop when the next responsible decision is clear.

skill-receipt-lineage

Routine success can release attention. When a decision changes, preserve the smallest fitting receipt. When context remains incomplete, classify the relation or independence as unknown and name the evidence that would reopen the claim. Keep batch traces with the evaluation owner.

The team distinguishes encounter, evaluation, and cultivation records; firstness names its relation; repeated observations retain their consistency value while independent-transfer claims use fitting evidence; a later receipt states the decision-changing delta and claim effect.

A production-shaped artifact becomes trustworthy when its source basis, fixture class, execution state, observed behavior, and inference ceiling remain visible together.

0.1

Classify what a prototype or synthetic fixture can legitimately demonstrate and state the disclosure and evidence needed for any stronger claim.

A prototype, demo, simulation, generated sample, fixture, or polished interface may be mistaken for production, population, clinical, causal, accessibility, or deployment evidence.

The skill classifies evidence and inference limits. It does not validate a dataset, prove safety or effectiveness, establish production readiness, confer regulatory adequacy, or grant publication, clinical, scientific, release, or adoption authority.

Repeated workspace and repository uses required separation among inspected source, synthetic fixtures, generated validation, environment-blocked runtime, production-shaped presentation, and human settlement. This supports the contract; comparative impact and cross-domain transfer remain open.

Artifact identity and revision; intended use; source basis; fixture class; execution and validation evidence; observed and unobserved behaviors; consequence level; domain and release owners.

Self-contained public-possible guidance. It contains no private workspace locator, protected record, real-person fixture, or hidden resolver dependency.

A prototype-standing receipt with source basis, fixture class, execution state, observed and unobserved behaviors, claim ceiling, disclosure, review owner, and re-entry condition.

Name intended use; classify artifact and fixture standing; record execution and validation states; separate observed from unobserved behavior; state the claim ceiling and required disclosure; route higher-consequence review to its accountable owner; define the evidence that would change standing.

prototype-standing-synthetic-fixture-honesty

Preserve the artifact's demonstrated value while correcting overbroad claims. Downgrade standing when execution or source evidence is absent; name the missing evidence and owner instead of discarding a useful prototype or implying proof.

A receiver can state what the artifact demonstrates, what remains unobserved, which inferences are permitted, which disclosure is required, and what evidence or review would change standing.

A handoff is complete only when the named receiver can begin correctly with declared capabilities and no decision-bearing private dependency.

0.1

Close a cross-platform handoff so a named receiver can orient, act or safely decline, verify, and recover without hidden dependencies.

A collaborator, tool, branch, or platform cannot use the originating workspace, conversation history, private links, or live source access.

Packaging does not broaden access, grant credentials, transfer publication or write authority, or make a snapshot live truth. Protected material is excluded rather than visually hidden. Consequential decisions stay with accountable owners.

One observed high-context collaboration formed and validated a twenty-file operator-blind package with schemas, fixtures, receipts, manifest, checksums, and archive integrity checks for a receiver unable to resolve workspace links. Independent cold-start continuation, comparative effect, and external-team transfer remain unobserved.

Named receiving job and receiver; capability profile; authorized source material; source revision or cutoff; disclosure boundary; required output; first checkpoint; stop and re-entry conditions.

Self-contained public-possible guidance. It contains no private workspace locator, hidden resolver dependency, or required source-local notation.

A portable-closed package, capability and source-scope manifests, cold-start entry, first checkpoint, acceptance state, open conditions, and re-entry trigger.

Name the receiver and job; declare capabilities; pin source identity; classify included, omitted, public, package-local, and unresolved dependencies; remove private locator dependencies; provide cold-start entry and safe non-entry; validate package integrity; obtain a behavioral acceptance receipt.

portable-context-closure

If identity is ambiguous, a capability is unavailable, or decision-bearing material is missing, mark the handoff blocked or in need of hydration. Return to the last verified revision, owner, and checkpoint; issue a successor package after source change.

The receiver can identify the current authority, perform or safely decline the first action, verify the first checkpoint, recover open returns, and proceed without private locators or undeclared connectors.

Collaboration strain becomes recoverable when its utility and consequence are assessed before one bounded move is routed to the surface that owns the affected state.

0.1

Turn a consequential collaboration strain into the smallest adequate, boundary-aware, authorized recovery move with a receiving receipt and dimensional outcome review.

Urgency, overload, divergence, context thinning, source loss, authority confusion, duplicated state, stale continuation, or diminishing returns disrupt a collaboration and a local repair may restore agency.

Recovery describes collaboration state rather than diagnosing people. It does not suppress dissent, infer consent, bypass privacy or permission, or acquire operational authority. Calm tone is not evidence that repair occurred.

Small-scope recovery, source reopening, context hydration, triangulation, successor handoff, and preservation of unresolved state recur across the Field Guide and Narrative Architecture and were used throughout Registry reflow. In this movement, an overlap review selected four bounded enrichments instead of four additional public entries, preserving mature owners and a smaller skill shelf.

Observable strain cue; current contract and goal; gap utility; consequence; affected dimensions; hard boundaries; smallest adequate scope; owning surface; activation authority; expected restored state.

Self-contained public-possible guidance with routes only to Registry pages. Private conflict, protected context, and operational authority remain with their owners.

A recovery assessment, one primary move, route target, authority and stop condition, receiving receipt, restored and remaining dimensions, and re-entry point.

Describe observable strain without blame; assess gap utility and consequence; reapply hard boundaries; name affected dimensions; choose the smallest adequate scope and one primary move; route to the owning surface; confirm authority; request receipt; review outcome and preserve remaining strain.

proportionate-recovery-routing

A generative or neutral gap may remain open. Narrow the route when scope exceeds consequence. Pause when authority or source cannot be recovered. Allow decline, rest, handoff, and alternate stable states. Repeated activation reopens the owning architecture.

The selected move matches the strain scope, useful disagreement remains available, boundaries and authority stay intact, the receiver acknowledges what arrived, restored dimensions are observable, and unresolved state remains reachable.

Verification becomes material when each check distinguishes a behavior someone relies on from a plausible failure through an observable result.

0.1

Design or evaluate checks by connecting consumer behavior, plausible fault, observable evidence, consequence, and the fitting level of pressure.

A test, validation step, inventory, count, coverage measure, digest, CI workflow, or AI-generated check may appear complete without demonstrating the behavior or integrity claim that matters.

A passing check supports the behavior it actually distinguishes. Counts, coverage, hashes, and generated tests remain scoped signals. Verification does not grant release, safety, regulatory, or business authority.

Behavior-first testing, boundary analysis, fault injection, property testing, artifact integrity, and operational CI are established practices. The workspace synthesized and applied them as a proportionate verification contract during Registry and research work. This movement verified handle identity, standing fields, empty public destinations, aggregate inclusion, and the expected self-contained count rather than treating creation count alone as proof.

Consumer or adjacent-system behavior; plausible failure; observable oracle; consequence; primitive under test; trust boundary; available test levels; execution and maintenance constraints.

Self-contained public-possible adaptation. Direct routes remain inside the Practitioner Registry; private source fragments and internal locators are not referenced.

A material test contract, fitting pressure cases, scoped-signal interpretation, execution evidence, perturbation result, and explicit remaining boundary.

Name behavior, fault, oracle, and consequence; select cases from implementation and judge through the contract; match pressure to the primitive; distinguish direct evidence from counts, coverage, and hashes; execute and perturb generated checks; weigh CI signal against cost; preserve remaining uncertainty.

material-verification-behavior-fault-oracle

Remove or narrow checks that protect no material contract. Reuse existing platform integrity and CI capabilities. When the oracle is weak, lower the claim and seek a user example, domain property, trace, differential implementation, or human review.

The check fails under the named defect, passes under the intended behavior, protects a stable contract, produces an interpretable signal, and supports only the claim its oracle can distinguish.

A release becomes trustworthy when the visitor’s need, source standing, consequence, destination, and human settlement remain visible together.

0.2

Carry useful meaning into a fitting audience while preserving source identity, privacy, evidence standing, human publication authority, and a clear stopping condition.

Material is moving between private, internal, public-possible, repository, or public contexts and the receiving audience, consequence, or destination can change what should travel.

The skill prepares and reviews audience-fitting material. It does not grant publication, representation, legal, privacy, repository, or operational authority. Private locators and protected evidence remain with their authorized owners.

The method is already embodied in the Registry’s publication-boundary guidance and was repeatedly applied during Registry reflow, Generated Research routing, exact-payload preservation, and public-possible skill precipitation. This movement exercised it by adapting two source-private capabilities into self-contained entries and routing four public capabilities to existing owners while keeping publication human-owned.

Visitor need; candidate source material; source and evidence standing; intended destination; disclosure and authority boundaries; consequence level; existing Registry receivers.

Self-contained public-possible guidance with direct routes only to Enterprise Human-AI Practitioner Registry pages. Private source locations and resolver records remain absent.

A selected receiver or audience-fitting adaptation, semantic projection receipt, source-boundary account, proportionate review depth, assembled-surface check, and release or re-entry condition.

Name the visitor result; classify source standing; inspect existing receivers; identify portable meaning and semantic invariants; declare transformed expression, intentional omissions, and target-local additions; choose review depth by consequence; form the smallest fitting receiver; inspect the assembled route; record human settlement and re-entry conditions.

audience-fitting-publication-proportionate-review

Use an existing receiver when it already serves the reader. If a projection changes What, When, Boundaries, Result, claim standing, or authority, return the change upstream rather than hiding it in adaptation. Preserve several destination candidates when consequences differ; return publication to its human owner.

The visitor can identify the page’s job; the destination supports the claim; preserved meaning, transformed expression, intentional loss, and source boundary are recoverable; private identity is not exposed; review depth matches consequence; another pass would leave the route materially unchanged.

Partial enterprise context remains workable when the current goal, evidence, authority, scope, reversible move, and decision-triggered questions are explicit.

0.1

Support competent action under intentionally partial context by declaring the active horizon, continuing reversible work, and requesting additional context when it can change a consequential decision.

A collaborator sees only a role-shaped portion of an organization or system and must remain useful without inventing missing approvals, policies, ownership, or operational context.

A declared horizon describes available context and action. It does not expand access, infer approval, centralize intentionally distributed knowledge, or replace human risk acceptance and institutional authority.

Progressive disclosure, least-privilege access, retrieval on demand, human approvals, and compact continuity artifacts recur across enterprise and AI systems. The workspace used this posture throughout high-context Registry work. This movement stayed within loaded Notion sources and approved Registry receivers while repository and publication actions remained outside the active horizon.

Current goal; available evidence; current artifact; settled constraints; granted tools and permissions; known unknowns; likely holders; reversible next move; decision triggers.

Self-contained public-possible adaptation with routes only to Registry pages. Private organizational context, inaccessible sources, and internal locators remain outside the projection.

A declared context horizon, reversible current movement, bounded context requests, human-owned questions, memory residue, and re-entry triggers.

Declare goal, evidence, authority, scope, and source boundary; distinguish active context from distributed context; choose a reversible move; record only decision-changing unknowns; ask at the decision edge; preserve human authority; settle when missing context would not change the current decision.

declared-context-horizon

Pause when missing context can change safety, permission, privacy, irreversible effects, or high-consequence outcomes. Otherwise proceed with stated limits. Return when a decision trigger, source change, or human response arrives.

The collaborator advances useful reversible work, names the limits of the visible slice, avoids simulated organization-wide authority, and raises exact questions when their answers become consequential.

Responsible continuation begins by distinguishing a first encounter, a repeat encounter, and re-entry after interruption from the material and evidence actually available.

0.1

Select a fitting collaboration posture for first, repeat, or re-entry encounters by recovering target identity, available context, iteration intent, boundaries, delta, and continuation point.

A human or AI is meeting, revisiting, or resuming a page, artifact, contract, branch, workspace, or collaboration and familiarity or missing context could be misread.

An encounter receipt records what was received; it does not prove complete understanding, memory, agreement, permission, or authority. Privacy, publication, license, safety, and source boundaries remain active.

First, repeat, and re-entry distinctions recur across Welcome Mat, threadkeeping, Narrative Capsules, Registry recovery, repository re-entry, and successor handoff. The practice was exercised during interruption recovery and multi-session Registry work. This movement resumed from a verified continuation state, then reloaded the current schema, entries, navigation surfaces, and affected source pages before writing.

Encountering presence; target and stable identity; prior encounter evidence; material actually in context; continuity mode; iteration intent; current contract and horizon; boundaries; re-entry point.

Self-contained public-possible guidance with direct routes only to Registry pages. Private encounter history and inaccessible source locations remain protected.

An encounter classification, context assessment, delta, posture, receipt, hydration request where needed, and verified continuation or deferral.

Identify presence and target relation; resolve stable identity; classify first, repeat, or re-entry; assess material context; name intent; recover current contract, horizon, and boundaries; choose posture; disclose gaps; issue receipt; continue, request hydration, or route recovery.

context-sensitive-encounter-reentry

Lower context confidence when material is partial. Preserve source terminology on first encounter. Request the minimum hydration needed for a decision. Continue only inside the verified boundary when the full contract cannot be recovered.

The collaborator avoids blank-slate overwrite and unsupported familiarity; repeat work names the delta; re-entry recovers current state and boundaries; missing context is explicit; continuation begins from a named point.

Long-running work resumes more reliably when decisions, rationale, provisional state, open questions, and review triggers live in a small inspectable artifact.

0.1

Preserve decision-changing rationale and provisional state outside transient conversations so collaboration can resume without reconstructing or replaying the full history.

Work spans sessions, tools, collaborators, or major revisions and future action depends on why a decision was made, what remains provisional, or when an open question should return.

A threadkeeper preserves collaboration memory; it does not replace authoritative source documents, tests, approvals, participant testimony, or publication decisions. Sensitive rationale remains in its authorized context.

Threadkeeping, decision records, continuity surfaces, fold receipts, research ledgers, and re-entry accounts recur across the Registry and supported recovery after interruption during a high-consequence multi-surface reflow. This movement used a compact account to preserve why six candidates became entries and four became enrichments.

Material decisions; rationale; stable or provisional standing; assumptions; declined alternatives; open questions; next movement; review or wake trigger; source and authority references.

Self-contained public-possible guidance with routes only to Registry pages. Private decisions, participant identities, and exact internal locators stay with their authorized records.

A compact threadkeeper entry or ledger containing decision, rationale, standing, open returns, source pointers, and review trigger.

Identify future-decision value; record the decision and rationale; mark stable versus provisional; preserve meaningful alternatives and open questions; attach source or authority pointers; name the next movement and review trigger; compact entries that become transcript-like.

decision-rationale-threadkeeping

Record only information that can change later work. Compact narrative accumulation into decisions and relations. Use direct notes for one-off tasks. Reopen when a decision changes, an assumption fails, or the review trigger arrives.

A future collaborator can recover why the decision exists, what may still change, what was declined, which questions remain open, and where to resume without replaying the conversation.

Goal fidelity can be restored non-adversarially when the desired effect, local boundary, current source, and behavior check become more explicit than the mismatch narration.

0.1

Restore the intended effect when humans and AI receive the goal signal unevenly, using proportionate clarification, local boundaries, surgical correction, and behavior checks.

A human–AI result is fluent yet only partly fits because goal signal, context, authority, constraints, or receiving criteria were unevenly available or interpreted.

The skill interprets mismatch as a recoverable collaboration condition before treating it as opposition. It does not erase accountability, disagreement, power, or safety boundaries. Exact quotations, code, identifiers, properties, and payloads remain source data.

Desired-state-first phrasing, locally adjacent exclusions, current-source recovery, surgical correction, and behavior checks recur across Semantic Attractor Design and Goal Fidelity work and were exercised repeatedly during a completed high-consequence Registry reflow.

Human-intended effect; current receiving artifact; current source of truth; visible constraints and authority; observed mismatch; useful work to preserve; behavior or consequence that can verify restoration.

Self-contained public-possible guidance with direct routes only to Practitioner Registry pages. Private dialogue, source-local notation, and protected evidence remain with their authorized owners.

An outcome anchor, mismatch reading, desired replacement state, adjacent boundary, smallest corrective change, behavior check, and bounded return condition.

Recover the intended effect and current authority; compare the receiving result with that effect; classify missing context, attention drift, referent or authority drift, constraint literalization, scaffold capture, or evidence gap; state the desired replacement; place any hard exclusion beside it; preserve useful work; apply the smallest corrective change; verify behavior; record the return condition.

proportionate-response-uneven-goal-fidelity

When the mismatch is local, preserve the fitting result and repair only the affected layer. When the goal itself remains plural or contested, retain alternatives and route the decision to its human owner. When visible context already supports the result, return to direct work.

The intended effect, current receiving state, cause of mismatch, preserved useful work, local boundary, and behavior check are explicit; the correction restores direction without broad deletion or invented authority.

A probability becomes decision-usable when the reader can answer: probability of what, for whom, over what time, under which conditions, from which evidence, and with what calibration standing?

0.1

Turn an ungrounded likelihood statement into a calibrated event-specific probability contract that can support human–AI decisions.

A collaboration uses probabilities, likelihood bands, confidence ranges, expected effectiveness, transfer claims, or comparative language that could be misread without an explicit referent.

Probability describes a defined event under conditions. Desirability, ethical acceptability, priority, action authority, and the truth of underlying claims each receive their own evaluation. Structured judgment carries an explicit label.

Probability-of-what contracts, range estimates, access coverage, claim classes, and transfer limits recur across Registry research payloads and Practitioner Probes. The recent collaboration used structured-judgment priors while reserving measured-frequency and universal-superiority claims for evidence designed to support them.

Decision or claim; event; population; horizon; operating conditions; evidence base; calibration method; counterevidence; transfer boundary; human-owned risk or priority decisions.

Self-contained public-possible guidance with direct routes only to Practitioner Registry pages. Separate authorization governs any use of private observations and unpublished estimates.

A calibrated probability statement, structured-judgment label where applicable, transfer boundary, counterevidence condition, and next observation that could update the estimate.

Name the event before the number; define the population and horizon; state operating conditions; identify measured, estimated, or structured-judgment evidence; evaluate probability, desirability, priority, and claim confidence in their fitting frames; include counterevidence and update conditions; present a range when the evidence supports bounded precision.

probability-of-what-contract

Remove unsupported precision, widen the range, or return to qualitative standing when evidence is thin. Preserve competing events and conditions rather than merging them into one confidence score. Route consequential risk acceptance to the accountable human.

Every probability or range names its event, population, horizon, conditions, evidence base, calibration status, and transfer boundary; desirability, priority, claim confidence, and frequency remain distinct.

Research becomes actionable without becoming automatic authority when its returned report receives an evidence audit and only decision-changing findings revise the receiving work.

0.1

Validate returned research before it changes instructions, artifacts, or public claims, then apply only evidence-bearing revisions.

A substantial research report has returned and may alter an instruction, skill, design, public claim, implementation plan, or decision surface.

A report supplies evidence and interpretation; it does not acquire publication, implementation, representation, or risk-acceptance authority. Validation preserves original sources and protected material with their owners.

Registry research receipts and atlas revisions repeatedly separate source coverage, claim standing, currentness, disconfirmation, instruction yield, and retained baseline. The full skill has strong precedent; a fresh dispatch-and-ingestion cycle remains a useful next observation.

Returned report; original decision question or brief; source and access ledger; receiving baseline; claim and probability contracts; human-owned decisions; expected output and completion conditions.

Self-contained public-possible guidance. Validation may name unavailable evidence without exposing protected sources, private locators, credentials, or unpublished decisions.

A validation receipt, corrected claim map, retained baseline, bounded instruction or artifact revisions, unresolved evidence gaps, and a human decision route where needed.

Recover the original decision and brief; inspect source access and citations; classify claims and counterevidence; test transfer and probability contracts; compare the report with the receiving baseline; select accept, accept with caveats, revise, reopen research, or human decision; make only evidence-bearing changes; record what remains unchanged.

returned-research-validation

When the original brief is missing, reconstruct only the decision, source expectations, output contract, and authority boundary before validation. Preserve conflicting evidence and choose a reversible movement. Route publication or risk acceptance to the accountable human.

A returned report is checked against its decision question, source coverage, citations, counterevidence, claim classes, transfer boundaries, and output contract before it changes instructions or public claims.

Deep research travels more reliably when the brief carries a complete decision contract rather than a topic, keyword list, or hidden-context prompt.

0.1

Compose a self-contained research brief that lets another human or AI investigate a consequential decision without relying on hidden originating context.

Another human or AI system must conduct substantial research from a blank slate, across a platform boundary, or without access to the originating collaboration history.

The brief specifies research work without granting access, execution, publication, or decision authority. Each receiver uses only authorized sources and keeps protected information in its governing system.

Multiple self-contained Registry research payloads have used decision questions, source hierarchies, coverage ledgers, claim classes, counterevidence, probability contracts, and output schemas across technical and collaboration domains for several months.

Research object; receiving decision and surface; current baseline; source and authority map; known access conditions; claim and probability requirements; intended output; completion gate.

The public-possible entry is self-contained. A research brief includes only information authorized for its receiver and excludes private locators, credentials, and protected evidence.

A self-contained decision-bearing research brief with source hierarchy, disconfirmation requirements, access coverage, probability contract, transfer boundaries, output specification, completion gate, and an optional profile-gated multi-path comparison design.

Name the research object, receiving decision, and intended use; include current baseline and authority; specify source hierarchy, claim classes, counterevidence, access ledger, probability contract, transfer boundaries, output form, and completion gate; for multi-path decisions add a profile gate, commensurable paths, sensitivity, falsification, and persona-dependent flips; review the packet for receiver sufficiency and disclosure safety.

self-contained-research-briefs

Return to a bounded source check when the decision does not require a full brief. If the packet lacks authority, access, representation, or publication context, route one human question before dispatch. Preserve unresolved sources and expected gaps explicitly.

A blank-slate researcher can identify the decision, receiving surface, source hierarchy, claim classes, counterevidence, access requirements, transfer limits, output form, and stopping condition without private hidden context.

A short continuity checkpoint can surface what still matters, preserve a restart point, and release attention without turning every ambiguity into work.

0.1

Surface consequential gaps between collaboration movements and preserve a reliable re-entry point without reopening settled work by default.

Between meaningful movements, after interruption, before settlement or handoff, or whenever asking what is missing, forgotten, or dangling could change the next responsible action.

The review is bounded by consequence. It preserves settled decisions, human-owned questions, privacy, publication standing, and source authority. It may propose a repair; execution still follows the current task grant.

Interstitials, checkpoints, re-entry protocols, threadkeeping, compression records, and return observations have recurring public Registry precedent. The completed Registry reflow used the pattern to close resolved returns and preserve thirteen current return families without inventing additional work.

Just-settled result; authoritative receiving state; goals and preservation boundaries; evidence standing; open routes and owners; expected restart conditions.

Self-contained public-possible guidance with direct routes only to Practitioner Registry pages. Private receipts and source-local annotations remain with their authorized owners.

A bounded continuity checkpoint, resolved-now items, named returns, one human question where needed, and a re-entry receipt.

Anchor the settled movement and current authority; select only consequence-bearing review lenses; classify findings as resolve now, named return, human question, deferred, or rest; repair the smallest safe item; write a restart receipt; release attention when another pass would preserve the same decision state.

interstitial-review-reentry

Use the four-question minimum when the state is already clear. Preserve one human-owned ambiguity rather than interpreting past its authority. Repeated review under unchanged evidence closes with rest.

Consequential omissions, stale assumptions, unowned returns, and restart needs are classified without reopening settled work; unchanged evidence closes the review.

Research becomes more reliable when source authority and access coverage are mapped before fluent synthesis makes missing evidence hard to see.

0.2

Map source authority and access coverage before synthesis so missing evidence, generated interpretation, and decision ownership remain visible.

A synthesis, recommendation, public claim, or consequential action depends on knowing which sources were authoritative, current, reached, generated, or unavailable.

Source authority describes what a source can support; it does not grant execution, publication, representation, or decision authority. Credentials, private records, and protected locators stay with their owners.

Access ledgers, source hierarchies, claim classes, provenance routes, and permission-bounded fallbacks recur across multiple Registry research payloads and were used during the completed reflow to distinguish current receiving authority from historical or inaccessible routes.

Decision-bearing question; receiving surface; likely source classes; access permissions; currentness horizon; consequence of missing evidence; publication and representation boundaries.

Self-contained public-possible guidance. Direct references remain within the Practitioner Registry or point to public sources; private locators and credentials stay absent.

A decision-bearing source map, source-scope integrity ledger, access and currentness record, accepted and rejected evidence, claim-class map, claim ceiling, unavailable-source record, and selected next posture.

Name the decision and consequence of error; declare intended targets and retrieval mode; identify authoritative, generated, current, public, and inaccessible sources; record reached scope, accepted and rejected results, completeness, currentness, and execution state; classify claims; preserve counterevidence and gaps; state the claim ceiling and smallest next posture.

source-authority-access-mapping

Record unavailable, partial, truncated, stale, or out-of-scope material and its consequence instead of implying coverage. Tighten identity before broadening search. Use an authorized fallback only when it can support the same claim, and route authority or representation questions to the fitting human owner.

The receiving decision, source hierarchy, intended and reached scope, retrieval mode, accepted and rejected results, currentness, completeness, execution state, claim ceiling, unavailable sources, and authority boundaries are visible before synthesis begins.

Detailed human–AI work becomes more recoverable when current authority, bounded work, verification, return paths, proportionate research, and public-safe translation remain visible together.

0.1

Inspect twelve conditions that make long-horizon human–AI work recoverable and select the smallest missing collaboration affordance.

A team is beginning, reviewing, or recovering detailed multi-turn human–AI work and wants to understand which collaboration conditions support reliable continuation.

The conditions support collaboration design without rating people, assigning identity, granting authority, or proving general superiority. Affordance maturity varies; planning and research-needed entries remain visibly provisional.

The twelve conditions recur across months of Registry pages on threadkeeping, coordination horizons, narrative capsules, re-entry, recovery routing, public-safe translation, skill design, and research payloads. One completed high-consequence collaboration made their combined effect unusually visible.

Receiving outcome; current artifacts and authority; active and deferred work; preservation boundaries; evidence and access conditions; collaboration strain or restart need.

Self-contained public-possible analysis with direct routes only to Practitioner Registry pages. Private coordination history and source-local notation remain outside this entry.

A twelve-condition collaboration reading, a narrative-affordance map, one smallest strengthening move, and explicit open transfer questions.

Inspect current authority, scope, packages, receipts, preservation, ownership, retrieval, verification, returns, coordination load, research depth, and public translation; identify which absence could change the next movement; select one fitting narrative affordance; observe the result and retain open questions.

recoverable-human-ai-collaboration-conditions

Use the list as a lens rather than a maturity score. Select only the condition whose absence can change the next action. Preserve unresolved transfer questions and return to the material work when another assessment would not change the route.

Teams can inspect twelve concrete collaboration conditions, identify the smallest missing affordance, and improve continuation without turning the list into a score or mandatory process.

Live collaboration remains trustworthy when coordination load is read through observable conditions and authority moves before the surface becomes unreliable.

0.1

Assess live coordination load and transfer current authority to a verified successor before retrieval or state ambiguity disrupts collaboration.

A long-running collaboration surface accumulates enough history, repetition, truncation, or moving-target risk that continuation reliability deserves an explicit reading.

The assessment describes the reliability of a coordination surface, not collaborator competence or work quality. Handoff preserves existing human decisions, permissions, privacy, publication standing, and unresolved questions.

Months of public Registry precedent in threadkeeping, re-entry, narrative capsules, coordination horizons, and recovery routing, plus one completed high-consequence collaboration that exercised an Amber successor handoff. A genuinely Red recovery remains untested.

Current coordination surface; latest receiving artifacts and receipts; active work; open returns; preservation boundaries; observed truncation, duplication, or edit-stability signals.

Self-contained public-possible guidance with direct routes only to Practitioner Registry pages. Private coordination records and source-local notation remain outside the entry.

A Green, Amber, or Red reading; a continue, compact, or handoff decision; and a verified successor receipt where needed.

Inspect retrieval, current authority, active work, settled history, open returns, and edit stability; select Green, Amber, or Red together with its movement; compact while state is intact; transfer predecessor, accepted receipts, current horizon, boundaries, and first action; verify one live authority.

coordination-load-successor-handoff

At Green, continue bounded work. At Amber, reduce batch size, move settled detail to receipts, and prepare a successor. At Red, pause expansion and rebuild the smallest trustworthy live surface from verified artifacts. Length alone does not determine the band.

One live coordination surface remains recoverable; the band is tied to observable retrieval conditions and a responsible movement; a successor receives current authority before ambiguity compounds.

A concept can travel without flattening when the source mechanism, portable name, added intuition, and transfer boundary remain visible together.

0.1

Translate a source-local collaboration concept into ordinary portable language while retaining its mechanism, evidence standing, and disclosure boundary.

A useful collaboration concept is expressed through local metaphor, custom notation, inherited terminology, or a platform-specific name and needs to travel to unfamiliar humans or AI systems.

Translation supports comprehension without acquiring source, evidence, identity, or publication authority. Private locators, protected evidence, and source-local annotations stay with their authorized owners. Verbatim technical identifiers remain unchanged where exactness governs.

The rubric consolidates recurring Registry practices in soft translation, public-safe summary, precipitation, referent resolution, campaign-shape explanation, and skill projection. The complete vocabulary matrix was selected after a multi-month source history and one high-consequence synthesis.

Source concept; operative mechanism; intended audience; candidate portable terms; evidence and authority boundaries; public/private disclosure posture.

The entry is self-contained and uses ordinary source concepts rather than source-local notation. Direct routes stay within the Practitioner Registry.

A source-concept and portable-name matrix, an explicit illumination statement, a transfer boundary, a cold-start readability check, and—when larger material travels—a preserve/change/did-not-travel/emergent ledger.

Name the source concept; choose an ordinary portable term that retains the operative mechanism; state what the translation makes easier to perceive; declare where the translation stops carrying evidence; when larger material travels, record preserved, changed, did-not-travel, and emergent elements plus receipt and return; test cold-start readability and authority; preserve the source as lineage rather than a public dependency.

source-concept-portable-language-rubric

If the portable term obscures a load-bearing distinction, keep the source concept visible and revise the bridge. If several terms fit, preserve alternatives and test them with the receiving audience. A vivid metaphor remains an intuition aid rather than proof.

An unfamiliar reader can understand and apply the portable term; the source mechanism, evidence standing, and translation boundary remain recoverable; private notation and locators stay absent.

Long-horizon human–AI work becomes more reliable when current authority stays compact, detailed work returns through bounded receipts, and handoff occurs before context becomes a moving target.

0.2

Coordinate long-horizon human–AI work through bounded packages, return receipts, current authority, and recoverable successor handoff.

Work spans several artifacts, turns, dependencies, or collaborators and interruption, duplicate action, stale state, or loss of exact boundaries would carry meaningful cost.

The skill organizes collaboration; it does not acquire decision, identity, publication, repository, or representation authority. Current receiving state outranks stale plans. Exact payloads, quotations, code, governed data, privacy, and evidence standing remain named preservation boundaries.

Formed from one completed, high-consequence collaboration program across hundreds of Registry surfaces and many work domains. A later cross-platform use produced a validated operator-blind package for a receiver unable to resolve workspace links, exposing receiver-specific closure as a missing contract. Comparative superiority, independent cold-start continuation, and external-team transfer remain untested.

Receiving outcome; authoritative artifacts and owners; preservation boundaries; work domains and dependencies; completion conditions; open returns; observed retrieval and coordination-load signals.

Self-contained public-possible guidance. Private workspace locators, internal work-package identities, and source-local notation remain outside this entry.

A compact control surface, bounded work packages, return receipts, receiver-specific resolution and acceptance states, coordination-load decision, verified settlement, or successor handoff.

Name the receiving outcome, receiver, and current authority; declare preservation boundaries; divide work into independently settleable domains; issue bounded packages; return verified receipts; distinguish creation, transmission, resolution, acceptance, and settlement; assess Green, Amber, or Red; compact or hand off while authority remains recoverable.

bounded-packages-return-receipts

Use direct work for one clear action. At Amber, reduce batch size and compact settled history. At Red, pause expansion and rebuild a minimal successor from verified receipts. If a receiver cannot resolve or verify the first checkpoint, return a partial or blocked receipt and hydrate only the missing decision-bearing material. When another coordination pass would preserve the same decision state, return attention to the material work.

Current authority, active work, preservation boundaries, open returns, next action, receiver capability, resolution state, and acceptance remain recoverable across turns; a package is not treated as closed merely because it was emitted.

Better research begins by choosing the lightest posture that can change the decision responsibly, then stops when additional evidence is unlikely to alter the next move.

0.2

Select the lightest adequate research posture before a consequential action and return a decision-bearing evidence receipt.

Missing workspace context, source authority, current facts, competing explanations, or transfer limits could materially change a consequential action or public claim.

Research informs decisions without acquiring decision or publication authority. Workspace content and search results are evidence unless explicitly designated as instructions. Generated synthesis, authoritative sources, and publication standing remain distinct. Probability claims name their event, population, horizon, conditions, evidence base, and calibration status.

Workspace orientation, bounded source verification, and comparative research were exercised throughout a completed multi-turn collaboration program. A later source-specific repository review exposed search contamination, truncated outputs, and environment-blocked execution; those observations support explicit source-scope integrity. Blank-slate dispatch and validation of a newly returned report remain unexercised in this skill cycle.

Decision-bearing question; consequence of error; visible context; likely authoritative sources; access conditions; competing explanations; evidence and publication boundaries; completion gate.

Self-contained public-possible guidance. It contains no private workspace locator, source-local annotation, or hidden resolver dependency.

A selected research posture, source-scope and access map, accepted and rejected evidence, counterevidence, claim ceiling, transfer limits, decision delta, completion gate, or one human-owned question.

Name the decision and what missing information could change it; declare intended source targets and retrieval mode; map actual authority, access, currentness, completeness, rejected out-of-scope results, and execution state; select the lightest posture; classify claims and counterevidence; state the claim ceiling, transfer limits, and decision delta; stop at the completion gate.

proportionate-research-depth-selection

When inquiry expands faster than the decision, return to the decision-bearing question and lower the posture. Record unavailable sources rather than implying coverage. Preserve competing claims when evidence conflicts. Ask one human question when authority, access, representation, or publication ownership remains material.

The selected posture is proportionate to consequence and uncertainty; intended and reached source scope, rejected out-of-scope results, access limits, currentness, execution state, and claim ceiling are visible; the result changes or confirms a specific decision.

Canary text supports filtering and exposure detection; a benchmark result remains attached to its model, harness, scorer, revision, uncertainty, and run date.

0.1

Public-possible internal entry. The human authorizes publication. Canary and benchmark claims stay attached to their named public sources, versions, run conditions, and evidence limits.

Prepared as the source-grounded Field Watch companion for the Context Canvas benchmark-orientation and MAPS evaluation visitor routes.

Every direct destination is an official public source. The entry contains no private workspace locator or protected route.

fw-benchmark-canary-coordinate

A known multi-phase collaboration route becomes easier for a cold-start AI collaborator when desired state, authority, ordered phases, checkpoints, recovery, stop, and safe handback are visible together.

0.1.0

Convert a settled repeatable collaboration route into a versioned AI-legible runbook without creating new authority or process machinery.

Use when a known outcome spans several source owners, Issues, checks, or handbacks and one short instruction no longer preserves reliable re-entry.

The current task grant supplies mutation and external-action authority. Active Issues supply scope, owners, and acceptance; tool constraints, repository policy, and human review bound operations. The runbook orders the work. This is collaboration procedure design, not legal or compliance writing.

Exercised on 2026-08-29 through a read-only cold-start traversal of the repository's multi-Issue Core iframe and publication route. The exercise exposed and corrected safe non-entry behavior, package/response owner separation, direct-route versus parent-carrier ordering, complete-closure checks, and exact recovery coordinates.

Desired receiving state; current source owners and Issues; explicit action modes; settled commands or actions; human-held decisions; source-band classification; checkpoint, recovery, stop, and handback conditions.

Every direct reference on this page is public-possible or an official public source. Private workspace locators, protected destinations, account paths, raw browser state, headers, and credentials remain with an authorized resolver.

One lean public-possible or internal-only runbook, one safe handback shape, and an exercised or working standing tied to a real receiving route.

Confirm runbook fit; lead with desired state; resolve existing owners and grants; select public-possible or internal-only source band; write ordered phases with checks, recovery, and stop; exercise once; return one public-safe disposition; rest when the route is legible.

@driftframe/field-watch/ai-collaboration-runbooks

Return an unsettled diagnosis to a playbook or research spike. Pause state-changing actions when a named stop forms. Preserve the known-good coordinate, return the bounded condition, and reopen only the affected phase through its owner.

A cold collaborator selects the correct phase and action mode, runs the fitting check, states what it establishes, stops at material divergence, and returns one sanitized pass, friction, or blocker without loading the full history.

Natural language and recoverable responsibility can coexist when consequence-bearing references gain explicit anchors while low-consequence references remain light.

1.1

Keep human reference natural while making consequential referents recoverable across Notion skills, prompts, artifacts, handoffs, and public-possible descriptions.

Role, authority, audience, evidence, ownership, personalization, representation, or cross-platform movement could change with an unresolved reference.

Referent anchors clarify task meaning without becoming identity governance. Representation authority remains with the represented people and their authorized evidence. Prompt personas, persona vectors, social proxies, register, identity, and evidence remain distinct constructs.

Applied throughout the current Registry and Context Canvas work to replace pronoun prohibition with referent resolution, sustain declarative supportive change writing, and keep publication, repository, participant, visitor, and collaborator roles explicit.

The task-bearing language; candidate referents; the consequence of ambiguity; the receiving audience or function; relevant authority, evidence, ownership, representation, and privacy boundaries.

The public-possible account is self-contained. The canonical binding and affected private skill contracts remain in authorized Notion context.

Recoverable referent anchors, natural supportive prose, explicit authority and evidence relationships, disclosed social proxies where present, and a bounded clarification or return route.

Locate consequence-bearing references; name the intended person, role, group, function, artifact, source, or decision owner; keep reliably resolved references natural; separate task-local anchors from governed identity records; preserve a clarification route where alternatives remain open.

@driftframe/field-watch/referent-resolution-supportive-register

An open reference remains a faultless gap. Preserve the expression, name the consequence-bearing alternatives, and route the question to the fitting human receiver. Additional wording changes rest when the referent and action are already recoverable.

Consequential references are recoverable; personal pronouns remain available; ownership and authority do not drift; social proxies remain disclosed; prose retains warmth, variation, specificity, and cognitive breathing room.

Material skill use can improve the next collaboration when one decision-changing observation receives one proportionate route, one return condition, and a release line.

1.1

Preserve one decision-changing observation from material skill use, select one proportionate route, and release attention.

Observed use, human correction, a tool boundary, counterexample, dependency event, or projection exit can change the next materially similar collaboration.

Accounts preserve collaboration evidence without becoming activity logs, performance records, candidacy scores, or authority grants. Repository execution, publication, and platform addition remain with their fitting human or repository-capable owners.

Used to retain existing Notion skills during the exact-payload and aggregate-view formation, record material tool boundaries, preserve archive as reversible holding, and avoid creating additional skills where existing contracts already own the job.

A material receiving use; the observed change or correction; the existing skill or binding that owns the job; the decision effect; one fitting route and return condition.

The public-possible entry carries the practice account. Private collaboration accounts and canonical skill locators remain inside authorized Notion context.

At most one short account, one primary route, one smallest change or explicit rest state, one return condition, and a release line.

Name the receiving use; state the new observation and decision effect; check existing skill ownership; select retain, revise, form a distinct skill, project temporarily, rest, fold, or retire; make the smallest useful change; release attention.

@driftframe/field-watch/collaboration-accounting-notion

Smooth use needs no account. Two consecutive meta-passes without intervening material use leave one return condition and close. Deeper recursion requires a human-named purpose, bounded question, and explicit end.

The next similar collaboration gains a changed instruction, boundary, route, or return condition; unchanged evidence supports rest; the receiving work becomes easier to resume than the meta-level account is to continue.

A collaboration surface becomes easier to use when present meaning, consequence-fitting boundaries, contingent possibilities, and a stopping condition remain distinct.

1.1

Reduce overextended planning or coordination to the smallest useful receiving surface while preserving lineage, material boundaries, deferred possibilities, and a stopping condition.

Possible roles, gates, validators, services, identities, or automation have become more prominent than the reachable receiver and present collaboration need.

The skill shapes approved Notion surfaces and preserves material boundaries. It avoids assigning model authority, creating candidacy gates, inferring repository access, or turning possible machinery into current obligations.

Used to keep the Registry aggregate-view formation bounded, retain direct routes instead of a universal control surface, and separate exact payload preservation from visitor adaptation without discarding either form.

A proposal or coordination surface; the reachable receiver; the smallest useful result; material privacy, security, legal, destructive-operation, and production boundaries; known re-entry conditions.

The public-possible account carries the full practice meaning. Internal lineage resolves only for authorized collaborators.

One usable receiving surface, one calm shelf of contingent possibilities, preserved lineage, and one observable stopping condition.

Name the receiver; separate present meaning from contingent mechanisms; retain consequence-fitting fields and boundaries; place deferred mechanisms at rest with concrete re-entry conditions where known; state completion and release attention.

@driftframe/field-watch/proportionate-surface-release

If a real consequence boundary is at risk, restore the necessary depth. If expansion resumes through speculative machinery, return to the named receiver and useful result. Two unchanged meta-passes support rest.

The receiver can act, interpret, or rest; privacy and execution boundaries retain appropriate depth; deferred possibilities remain recoverable; another meta-pass would leave present meaning unchanged.

A Registry route becomes useful when the reader job, load-bearing claim, fitting destination, private lineage, and unresolved question remain legible without implying repository execution.

1.1

Prepare the smallest visitor-centered route from one internal source family to an existing or prospective Practitioner Registry destination.

Internal material needs a Registry counterpart, consolidation destination, local explanation, adaptation, or purposeful deferral.

The skill prepares Notion-side route understanding. Private source identity remains inside authorized resolver context. Public authorization and repository writes remain human-owned or repository-capable work.

Used to distinguish exact payload mirrors from visitor syntheses, route Generated Research families into the Registry, preserve repository-read-only boundaries, and settle a two-key source relation system whose public records remain useful without private locators.

One authorized internal source or coherent family; the present Registry routes; the visitor’s decision or interpretive need; known publication and repository boundaries.

The public-possible account is self-contained. Canonical skill and private source locators resolve only inside authorized Notion context.

A visitor-centered route proposal with destination, relationship, adaptation boundary, optional stable public source key, authorized-resolution boundary, unresolved question, and calm stopping condition.

Name the reader job and load-bearing claim; compare existing destinations; choose the smallest fitting relationship; when movement crosses repositories or public/private contexts, assign a stable public source key and keep exact private resolution in authorized context; state what travels, adapts, remains private, and stays open; leave execution and publication with their owners.

@driftframe/field-watch/registry-route-reading

If destination fit remains ambiguous, preserve the alternatives and the consequence-bearing question. If another pass leaves the interpretation unchanged, retain the route as proposed or deferred and return attention to the receiving work.

The receiving route is understandable without title matching, private locators, inferred write access, or a new approval gate; purposeful deferral remains a valid result.

Hybrid teams can distinguish authentic registry artifacts, inherited implementation details, project utility, execution risk, maturity, and human release authority through one bounded evidence receipt.

1.1

Establish the identity, integrity, source relationship, security standing, receiving utility, maturity, currentness, and human release condition of an external dependency.

A dependency enters or changes; a transitive module appears surprising; utility is unclear; publisher, source, advisory, install behavior, lock digest, or delivery route changes; or a prior assurance receipt loses currentness.

The human owns package selection, credential use, installation authorization, merge, release, and deployment. Repository-capable collaborators execute approved changes. Public-possible receipts retain public package evidence and relative repository paths while local machine identities, home-directory paths, credentials, tokens, private registries, unpublished locations, and sensitive environment values remain within the operational surface.

Applied on 2026-08-25 across the Notion-facing chimeric-prototype-exemplars inspection and the prototype-exemplars execution receipt. It resolved kleur as inherited through Wrangler, verified public registry coordinates, and connected schema, visual, and Cloudflare dry-run checks to their material project affordances.

Repository or artifact scope; manifest and lock coordinates; imports and executable routes; official registry or upstream metadata; lifecycle scripts and downloaded artifacts; human-owned release coordinates.

Public package evidence remains in this entry. Private canonical contracts, platform companions, and operational receipts remain within authorized context.

A sanitized assurance receipt with independently scoped claims, utility standing, dependency path, relevant checks, unresolved evidence, decision, re-entry events, and human release condition.

Inspect declarations and observed use; resolve direct or inherited dependency paths; verify registry identity and digest; separate declared source, attestation, reproducibility, advisory, utility, and maturity claims; review lifecycle and native execution; run the smallest affordance checks; return one currentness-scoped receipt.

@driftframe/field-watch/dependency-provenance-utility-review

Preserve one unresolved claim and the observation that would settle it. Reuse the receipt while coordinates and digests remain stable. Re-enter for manifest, lock, advisory, publisher, execution, or delivery-route changes.

Direct dependencies have observed or conditional receiving jobs; transitive modules resolve through evidenced parents; integrity, source, security, utility, and maturity remain distinct; executable checks demonstrate the claimed affordance; the receipt states when it becomes current again.

Teams can make prompt-side conditions visible, vary them deliberately, and judge their contribution in the receiving work.

0.1

This entry describes prompt-side conditions and local observation. Named teams own workflow architecture, policy, safety, domain verification, evaluation, and deployment decisions. A single successful prompt is local evidence; representative variation establishes transfer.

The Context Canvas Prompt-Side Conditions family applies the practice across an AI Systems reading, a conditioning instrument, an educator-centered live lab, architecture correspondences, and public research citations.

Direct destinations are public or public-possible. Additional internal guidance and lineage remain within the authorized two-key resolver.

fw-practice-prompt-side-conditions

A bounded repository re-entry pass can recover current artifact sets, resolve Notion source families, and surface standing surprises while repository change ownership remains explicit.

1.2

Recover current repository context, resolve it against Notion source authority and standing, and return a bounded mapping or repair receipt.

A repository artifact reappears after a pause, an older Notion inventory may have drifted, or a cited cornerstone is resting, archived, duplicated, renamed, or difficult to retrieve.

Embedded Notion AI reads repository material and edits approved Notion surfaces. The human or a repository-capable collaborator owns repository changes. Public authorization remains a distinct human decision.

Applied to all 18 Context Canvas Debriefs on 2026-08-24 and later to a source-specific repository review and validated portable-closed package. These uses support mode-qualified re-entry and explicit environment boundaries; independent cold-start continuation and target-runtime execution remain unobserved.

Named repository scope; receiving decision; current catalog or artifact set; known coordination surface or source family; repository and publication authority.

The public entry carries the portable skill description and evidence. The private canonical contract and source maps remain within authorized Notion context.

A mode-qualified repository re-entry receipt with revision or cutoff, optional digest, artifact map, source scope, execution state, claim ceiling, changed relationships, unresolved edges, action owner, and re-entry condition.

Name repository scope and decision; select native, connector-resolvable, or portable-closed mode; pin revision or cutoff; preflight capabilities; inspect accepted, rejected, unavailable, and unsearched scope; separate source standing and execution evidence; compare prior routes; state the claim ceiling; return the smallest responsible action and re-entry condition.

@driftframe/field-watch/repository-reentry

Ambiguous connector identity stops the claim. Missing decision-bearing package material returns needs-hydration. Environment-blocked execution is recorded as a boundary rather than a repository failure. Stable decisions return the current receipt and rest; repository change, retrieval need, publication decision, or material standing surprise reopens the skill.

The repository identity, revision or cutoff, access mode, reached scope, execution state, claim ceiling, current artifact set, unresolved lineage, and next accountable action are recoverable without treating connector reach or a portable snapshot as live completeness.

Notion skills remain more stable and negotiable when every instruction leads with the desired state, names constitutive boundaries, and gives recovery a calm destination.

1.1

Shape every Notion skill fragment, contract, projection, cycle, schema capsule, and public description as an affirmative, observable collaboration target.

Authoring, revising, cataloging, projecting, or publicly adapting a Notion skill or collaboration cycle.

Materially meaningful negation retains its role in definitional terms, epistemic humility, privacy and safety edges, and paired diagnostic catalogs. Verbatim source material remains source data. The desired replacement state carries the operational instruction.

Applied to the Repository Re-Entry skill, the Context Canvas precipitation contract, eight Exemplar teaching surfaces, and the collaboration-accounts directory on 2026-08-24.

The skill surface; its receiving job; authority, privacy, safety, and evidence boundaries; Semantic Attractor Design guidance.

The Field Watch entry carries public-safe guidance and a stable handle. The private canonical skill and supporting lineage remain within authorized resolver context.

A negation-aware skill surface with affirmative routing, constitutive boundaries, observable success, calm recovery, and a verifiable release state.

Identify the desired end state; write affirmative routing and actions; state authority as ownership and location; pair meaningful limits with their positive anchor; give recovery a destination; make completion observable; verify prose, examples, schema keys, comments, and identifiers share the same polarity.

@driftframe/field-watch/affirmative-semantic-attractors

One language pass follows each material skill change. Further passes follow observed use, human correction, or a changed boundary. Stable receiving behavior releases attention.

A collaborator can describe the target behavior directly; boundaries name the owner of authority; recovery points to a calm next state; completion is observable in the receiving work.