INTELLIGENCE MAP
RESEARCH → INTELLIGENCE → OUTPUT
Foundry Overview
SOURCE MIX
What’s in the archive
ACTION QUEUE
What the system wants to do
SOURCE RECOVERY
Archive → living research
v4.1 OPERATING LOOP
The Foundry can now carry contract evidence from signature readiness through performance, compliance and renewal decisions
v4.1 turns ACTIVE contract obligations into reviewed performance/compliance evidence, immutable variance and expiry risk, lineage-preserving amendment changes, and frozen renewal decision packages—without declaring legal breach, sending notices, executing amendments, renewing/terminating agreements or authorizing payment.
| Priority | Source | Category | Action | Recovery | Then→Now | Project fit |
|---|
Turn dead/syndicated links back into evidence
Fetch saved pages, follow safe public redirects, extract canonical/title/publisher/date metadata, and optionally search exact titles for surviving or original sources.
Work the highest-value queue
QUEUE
Unresearched sources
Claim → evidence → current verdict
Research a saved headline against the live web, prefer primary sources, preserve counterevidence, and convert the result into Then→Now, build and content intelligence.
Spend research only on the winners
Deep research is intentionally opt-in. The local triage queue decides what deserves a model/web-search call.
RESEARCH QUEUE
Highest-value unresearched claims
EVIDENCE GRAPH
Connected intelligence
Browse the durable relationships between sources, claims, explicit evidence/counterevidence, entities, technologies, signals, projects and production ideas.
STORY CLUSTERS
Syndication / duplicate story view
Conservative clustering only: titles must normalize to the same story fingerprint. The Foundry does not merge vaguely similar headlines.
What changed since the last research?
Sources with a Deep Research baseline enter this queue. A refresh re-researches the same claim, compares the two immutable reports, scores the delta, and preserves every material change.
Material deltas, not wording noise
High-value changes include verdict/status reversals, new or removed findings, primary-source changes, buildability shifts, new opportunities and resolved/new watch questions.
REFRESH QUEUE
Research with a baseline
WATCH QUESTIONS
Unresolved future hinges
Signal → editorial campaign
Create a canonical flagship plus channel-native Medium, supporting, counterpoint, Then→Now, LinkedIn, newsletter, social, Build This and video assets — all attached to the same evidence package.
Campaigns start NEEDS_REVIEW. Low evidence readiness is a research flag, not permission to fill gaps.
Build idea → MVP experiment
Turn a BUILD idea into a problem statement, hypothesis, target user, revenue model, minimum architecture, success metrics and a first smoke-test experiment.
Completing an experiment can automatically create a transparent first-party case-study campaign.
EDITORIAL CAMPAIGNS
Research → publishing system
BUILD PROJECTS
Hypothesis → experiment → result
First-party results become new intelligence
A build experiment can generate a case study; asset performance and customer feedback can be logged; those outputs become graph nodes and feedback records. The Foundry can then update the thesis instead of treating publishing and building as endpoints.
RELEASE QUEUE
Approved asset → destination
Every queue item carries its destination variant, canonical URL, UTM tracking URL, scheduled time, export history and publication state.
DESTINATIONS
Channel adapters
Manual export is the safe default. Optional webhook destinations can hand approved payloads to your own publishing automation without pretending a handoff is a confirmed publication.
PERFORMANCE INTELLIGENCE
What actually earned attention or outcomes?
LEARNING LAYER
Descriptive insights, not fake causality
Prepared ≠ published. Correlation ≠ causation.
The Foundry tracks PREPARED / EXPORTED / DISPATCHED / PUBLISHED separately and only learns from performance you actually record. Channel and topic summaries are descriptive until enough controlled evidence exists to justify stronger conclusions.
EXPERIMENT LAB
Variant → observation → evaluation
Run explicit title, hook, CTA, format, channel or timing tests. The Foundry stores the design assumptions, normalized observation window, minimum sample, practical-effect threshold and statistical signal separately.
NEXT BEST TEST
Recommendations from actual performance
Recommendations are proposals, not automatic decisions. They can suggest a title/hook test, CTA test, replication, or baseline collection depending on the evidence available.
Controlled vs. quasi
CONTROLLED means you actually assigned comparable traffic/cohorts to arms. QUASI means the variants were released separately or under conditions the Foundry cannot randomize. Both can be useful, but only controlled tests can support causal language here.
Stop double-counting snapshots
Performance records can be marked as snapshot/cumulative or incremental. Evaluations use the latest cumulative value plus incremental events inside each arm’s configured observation window.
Winning once is not a rule
A directional winner becomes a replication recommendation. The Foundry separates practical lift, statistical separation, sample sufficiency and experiment design so one lucky post does not silently become “what the market wants.”
PATTERN LIBRARY
What has replicated?
Completed Growth Lab experiments are aggregated across channel, topic, format and audience scopes. Every pattern keeps the exact contributing experiments and evaluations as provenance.
EXPECTED-VALUE QUEUE
What should we test or use next?
Expected-value scores shrink observed lift by replication depth and confidence. They rank attention; they do not promise future performance.
Emerging → Moderate → Durable
A pattern graduates only as evidence accumulates: repeated tests, directional consistency, controlled-test share, sample depth and statistical separation. Mixed evidence stays visibly mixed.
Pattern ≠ universal rule
Institutional patterns are scoped claims about our observed tests. Human review can ACCEPT, CHALLENGE or REJECT a pattern, and later experiments can weaken or reverse it.
Every rebuild leaves a snapshot
The Foundry keeps pattern snapshots over time, so we can see a belief strengthen, fragment or reverse as new evidence arrives instead of overwriting institutional memory.
DECISION PORTFOLIO
Exploit / Explore / Challenge
Institutional patterns, research gaps, unpromoted ideas and NOW_BUILDABLE findings become comparable decisions. Scores rank attention; they do not guarantee outcomes.
CAPACITY PLANNER
Spend attention intentionally
Capacity points represent relative attention/cost, not dollars. The planner reserves separate budgets for exploitation, exploration and belief-challenging work.
TREATMENT TAXONOMY
What actually changed?
Growth tests are decomposed into interpretable treatment features such as numbers, question framing, outcome language, CTA specificity, channel changes and timing changes.
Ranges, not promises
Where replicated experiment evidence exists, Portfolio Intelligence shows an empirical/planning range around the confidence-shrunk expected lift. The range is deliberately labeled as planning evidence rather than a forecast interval.
High information value means a decision is both uncertain and worth learning about. High expected value means the current evidence favors action. Those are different reasons to spend capacity.
Know enough → exploit. Know too little → explore. Belief weakening → challenge.
The Foundry now separates the value of acting from the value of learning. That lets scarce research, build and publishing capacity go where it can either produce an outcome or materially improve the next decision.
EXECUTION QUEUE
Approved decisions → governed work
Each portfolio allocation becomes one idempotent work item assigned to the appropriate specialist agent. Materialization does not automatically spend API credits, publish content, or mark work complete.
PLAN ACTIVATION
Materialize approved capacity
Approve a Portfolio plan first. Materializing it creates the work queue and preserves the exact candidate/allocation provenance. Re-running materialization creates no duplicate work items.
AGENT REGISTRY
Specialists responsible for the work
Prepare ≠ execute ≠ accept
RESEARCH can prepare portable packets without an API key and only spends model/search resources when you explicitly execute it. CONTENT, BUILD and GROWTH_TEST create reviewable downstream artifacts; their existing approval gates remain intact.
Completing a Work Item closes the associated portfolio allocation, but it does not bypass the approval/status rules of the campaign, build project, experiment, or evidence record it spawned.
Decision → owner → prerequisites → artifact → result → portfolio
The Foundry can now explain not only why a decision was selected, but who/what was assigned to execute it, which prerequisite blocked it, what artifact it created, and how the result flowed back into the evidence and planning system.
PROJECT PORTFOLIO
Objectives → adaptive reviewed DAGs
A strategic project can use the fixed v1.2 template or an adaptive plan assembled from objective-specific branches. Adaptive plans must be reviewed before executable Project Nodes exist. During execution, completed branch evidence can propose reviewable Plan Change Requests; high-risk findings may pause unfinished downstream work without silently rewriting the approved DAG.
Promote a larger objective
Adaptive creation proposes a reviewable plan first and creates zero executable Project Nodes until you approve that plan. Fixed mode preserves the v1.2 seven-node template.
Parallel discovery, gated production
Coordination does not erase gates
Completing a research branch does not automatically approve its conclusion. Completing an MVP branch does not make a product LIVE. Content still requires asset approval, and validation still requires an approved artifact and reviewable measurement design.
The Project Director rolls branch state upward and exposes the next dependency. It does not grant itself authority to accept consequential outcomes.
Objective → assumptions → validation portfolio → learning sequence → evidence → decisions → reviewed replanning
The Project Director still composes objective-specific branches, while v1.7 ranks the assumptions worth learning about next. The cheapest high-information test can move ahead of a much more expensive build when it protects more downstream capacity, and every measured result still flows through the existing assumption, decision, dependency, and replanning controls.
FORECAST CALIBRATION
How well did our ex-ante probabilities perform?
Only expected-learning analyses created before a validation run began are eligible for scoring. Later analyses cannot backfit the outcome.
STRATEGY OPTIMIZER
Which learning policies have actually been efficient?
Historical modifiers are deliberately small and sample-size shrunk. Realized learning efficiency is descriptive—not proof that one strategy causes better outcomes.
FORECAST LEDGER
Prediction → observed validation result
Improve future forecasts without rewriting old ones
The original probability vector, model version, timestamp and eventual validation result remain immutable. Future forecasts may apply scoped calibration corrections only when enough comparable history exists.
Brier score and calibration error measure forecast quality. They do not measure whether an opportunity itself was “good” or “bad.”
Learn about opportunities—and learn how well the Foundry learns
v2.0 closes the meta-learning loop: forecasts become scoreable claims, scoreable claims become calibration evidence, and calibration evidence can cautiously improve the next expected-learning strategy while preserving the full historical decision record.
DECISION DISCIPLINE
Did we choose well with what we knew then?
Do not confuse hindsight with a counterfactual
Ex-ante regret compares the selected strategy with alternatives actually available in the same immutable analysis. Realized outcome quality records what happened under the chosen strategy. Retrospective policy regret uses repeated historical policy performance with shrinkage.
Foundry does not claim to know what would have happened in this exact project under an unchosen strategy.
DECISION LEDGER
Choice set → selected strategy → realized learning
Forecast quality and decision quality are different problems
A calibrated forecast can still lead to a poor choice, and a good decision can still have a bad outcome. v2.1 keeps process quality, outcome quality and retrospective policy evidence separate so the Foundry can improve how it chooses without rewarding luck or rewriting history.
PROSPECTIVE POLICY TESTS
Test how the Foundry chooses
Pre-register competing decision policies before eligible projects are assigned. Balanced randomized assignment is the strongest available design; rotating quasi-experiments remain explicitly directional.
Pre-register before assignment
Creation does not assign a project. Human approval is required before prospective assignment begins.
Historical correlation → prospective policy test → bounded evidence
A strategy can look good historically because it was chosen for easier projects. v2.2 lets us prospectively assign comparable future analyses across policies, record deviations, and evaluate realized learning efficiency without pretending one experiment establishes a universal decision rule.
CONTEXTUAL POLICY MAP
Which decision policy works where?
Policy-experiment outcomes are grouped by context captured at assignment time: project stage, dominant uncertainty, downstream risk, test cost, regulatory intensity, physical complexity and evidence maturity. Sparse slices remain non-actionable.
Inspect a project in its current context
Choose a project to see its current context, matched policy slices and bounded strategy modifiers.
These modifiers influence strategy utility only. They do not change branch probabilities, approve a strategy, or create work.
Universal winner → conditional policy map
A strategy that performs well for early digital market-risk projects may not be the right policy for late-stage regulated physical projects. v2.3 preserves assignment-time context, uses prospective experiment evidence first, shrinks small samples toward neutral, and promotes only sufficiently replicated contextual signals.
ACTIVE POLICY LEARNING
Where is the policy map weakest?
Rank policy-learning gaps using contextual coverage, completed randomized depth, in-flight assignments, project value, strategy leverage and expected information gain. Active experiments count as coverage in progress rather than completed evidence.
EXPERIMENT PROPOSALS
Highest-value gaps → reviewable tests
Accepting a proposal does not approve an experiment. Materialization creates a separate DRAFT Decision Policy Experiment with context eligibility and its own human approval gate.
Context map → coverage gap → prospective experiment → stronger context map
Foundry can now identify where its own decision-policy knowledge is sparse, avoid duplicating experiments already in flight, and recommend the prospective comparison with the highest expected learning value. It still cannot enroll a project, approve a policy experiment, or promote a winner without the existing review and evidence boundaries.
META-LEARNING PORTFOLIO
Which policy experiments fit the exploration budget?
Allocate finite relative capacity across policy-learning proposals using expected map improvement, policy-pair novelty, contextual coverage gap, and a redundancy penalty. Points represent relative coordination/test burden, not dollars.
Budget approval is not experiment approval
A portfolio chooses where exploration capacity should go. Approving it does not approve any underlying Policy Learning proposal, create an experiment, enroll a project, or execute a validation. Those review gates remain independent.
Selection uses a novelty reserve first, then marginal learning value per capacity point after penalizing context/policy-pair redundancy.
Create a portfolio after Policy Learning has generated reviewable coverage-gap proposals.
Coverage gaps → experiment candidates → diversified exploration portfolio → stronger policy map
The system now optimizes the combination of meta-experiments, not merely the next one. It preserves budget, novelty, redundancy and governance provenance so later we can explain why one policy-learning experiment was funded while another was deferred.
ADAPTIVE META-PORTFOLIO REBALANCING
Reallocate exploration capacity as evidence changes
Scan an approved or active Meta-Learning Portfolio against the newest Policy Learning coverage. Materialized experiments stay locked; completed or no-longer-actionable planned allocations can release rolling capacity; new high-marginal-value gaps may be proposed as replacements.
Version the allocation — never silently rewrite it
A scan creates an immutable DRAFT rebalance proposal. Human approval is required before application. Applying it creates a successor Meta-Learning Portfolio and closes the prior version; it does not approve any underlying proposal or policy experiment.
Materialized experiments are preserved. Completed work may free rolling capacity. New replacement allocations are READY only at the portfolio layer and retain all downstream review gates.
REBALANCE LEDGER
Proposed and applied portfolio revisions
Allocate → experiment → learn → release/lock capacity → rescan coverage → propose successor portfolio
The exploration portfolio can now adapt to actual learning progress without erasing the original allocation plan. Rebalancing remains a decision-support layer: no experiment is cancelled, approved, enrolled or executed merely because its marginal learning value changed.
UNIFIED CAPACITY + OPPORTUNITY COST
What deserves the next capacity point?
Compare operating work, project validation, approved distribution opportunities, and meta-policy learning under one finite relative-capacity budget.
One budget does not mean one approval
Unified approval accepts only the relative allocation. It does not approve a Portfolio candidate, validation design/run, publication destination, or policy-learning experiment.
Validation receives a protected reserve; meta-learning is capped; remaining capacity competes globally on marginal utility per point after diversification penalties.
Build a portfolio to inspect cross-system opportunity costs.
Do → validate → distribute → improve the decision system — under one finite capacity model
The Foundry can now compare the opportunity cost of spending scarce capacity on an opportunity versus testing it, shipping an approved asset, or improving the policy map itself.
UNIFIED CAPACITY REBALANCING
Reallocate capacity as organization-wide work changes state
Build a state-based resource forecast for approved/active Unified Capacity portfolios, then propose a versioned successor allocation. Active downstream commitments stay locked; completed/dropped work can release capacity; still-uncommitted work remains reviewable.
Capacity states — not invented finish dates
Forecasts separate LOCKED_ACTIVE, MOVABLE_READY, FREED_NOW, and currently unallocated capacity. Scenario views show what is available now, after movable-work review, and after active commitments eventually complete.
These are capacity scenarios, not calendar forecasts. No completion date is inferred unless another workflow already has one.
REBALANCE LEDGER
Reviewable successor organizational portfolios
Allocate → execute → complete/invalidate → release capacity → rescore opportunities → propose successor portfolio
The Foundry can now adapt the organization-wide resource allocation without erasing the original decision. Rebalancing remains advisory until a human approves and applies the successor portfolio.
FORWARD RESOURCE DEMAND
What capacity could active projects demand next?
Project unmaterialized DAG work is separated into near-term pipeline demand and contingent downstream demand behind assumptions/decision gates.
Supported floor → expected gate survival → all survive
The model compares current reserved capacity with forward project demand under transparent scenarios. Existing Expected Learning probabilities are reused when already available; otherwise conservative assumption/decision-state heuristics are shown.
No calendar dates are invented. Gate-survival probabilities are planning estimates, and capacity points are relative organizational units—not dollars, labor-hours, or promises that projects will survive.
Build a forecast to inspect contingent demand and recommended holdback.
STRESS LEDGER
Scenario pressure against finite capacity
Current allocation → project gates → contingent demand → stress scenarios → reserve capacity before the bottleneck arrives
The Foundry can now see beyond today’s free capacity and identify how much resource demand may be waiting immediately behind unresolved project decisions.
MULTI-DIMENSIONAL CAPACITY
Which capability bottlenecks first?
Decomposes current allocation and gate-aware project demand into distinct capability pools instead of treating every capacity point as interchangeable.
Can the organization absorb this opportunity?
Admission is advisory and reviewable. Expected and stress demand are incremental capability needs for the proposed project.
Scalar capacity → capability pools → bottleneck stress → admission recommendation → human review
Research capacity is not CNC capacity. Content capacity is not sales-validation capacity. v3.0 models those constraints separately before admitting more work.
CAPABILITY-AWARE SCHEDULING
Can sequencing remove the bottleneck?
Turn an admission bottleneck into a relative-wave schedule. Waves are sequencing stages, not invented dates or durations.
Break the constraint without hiding the tradeoff
Compare sequencing, scope reduction, automation/process redesign, subcontracting, internal cross-training, and dedicated capacity acquisition.
Cost, delay, relief, and irreversibility values are relative planning heuristics until real quotes, staffing data, equipment specifications, or work-package estimates are supplied.
REMEDIATION LEDGER
Reviewed bottleneck-response plans
Capability forecast → admission bottleneck → sequence or acquire → review tradeoffs → select planning action
A selected option does not hire, subcontract, buy equipment, change the schedule, or mutate the capability profile. It records the reviewed strategy for removing the constraint.
CAPACITY INVESTMENT PORTFOLIO
Which bottleneck investment deserves funding first?
Consolidate project-level bottlenecks by capability and compare portfolio subcontracting, temporary rental, equipment/capacity acquisition, internal capacity growth, and process automation.
Relative first. Cash only from explicit assumptions.
Every candidate gets a relative value/cost multiple using existing planning scores. Cash payback, horizon value and ROI remain blank until you enter real financial assumptions such as an equipment quote, recurring cost, and value per capacity point per cycle.
Relative cost points are not dollars. Cash models are conditional scenarios, remain separately reviewable, and never authorize a purchase, hire, rental, subcontract or automation commitment.
Repeated bottlenecks → pipeline value → competing capacity strategies → portfolio allocation → reviewed break-even assumptions
A project-level remedy answers how to unblock one opportunity. The investment portfolio asks whether the same capability constraint appears across enough valuable opportunities to justify a shared organizational investment.
SOURCE EVIDENCE
Attach real quotes, bids, rentals, staffing or automation offers
Source terms are stored as entered evidence. Review confirms the terms are useful for planning; it does not verify the vendor or authorize procurement.
NORMALIZED TCO
Compare heterogeneous sourcing choices
Only options in the selected comparison currency are normalized. No exchange rate is silently assumed.
Investment candidate → sourced terms → review → same-currency TCO → preferred planning source
A source can be a quote, bid, offer, listing, budgetary quote, or internal estimate. TCO separates economic ownership cost from financing cash outlay and never converts currencies, verifies vendors, or executes purchases/contracts.
RFQ PACKAGE
Turn a capacity investment into a supplier-ready specification
Build an internal RFQ with technical, quality, capacity, delivery, commercial and evidence requirements. Exporting creates a file only; nothing is sent.
QUALIFICATION PIPELINE
Qualify before TCO promotion
Supplier capability evidence is separate from the commercial response. Hard qualification failures block shortlist; incomplete responses surface negotiation issues before promotion into v3.3 TCO.
Capacity requirement → RFQ → supplier qualification → response completeness → shortlist → reviewed offer → v3.3 TCO
The Foundry can generate and export an RFQ package, but it does not contact suppliers, transmit the RFQ, negotiate, contract, buy, rent, hire or subcontract.
SUPPLIER PORTFOLIO
Compare award structures, not just single bids
Use reviewed qualified RFQ responses to compare single-source, dual-source, balanced and resilience-oriented allocations.
NEGOTIATION INTELLIGENCE
Targets from reviewed bid differences
Cost, lead-time, reliability and capacity targets are planning anchors generated from the best reviewed same-currency alternatives. They are not supplier messages.
Qualified bids → source scenarios → concentration/resilience tradeoffs → negotiation targets → reviewed preferred award scenario
Scenario selection is planning-only. The Foundry does not contact suppliers, negotiate terms, award business, sign contracts, or issue purchase orders.
NEGOTIATION PLAYBOOK
Lock positions before revised bids
Convert an approved supplier portfolio with a selected scenario into supplier-specific target/reservation positions, BATNAs, concession ladders and conditional award branches.
AWARD DECISION INTELLIGENCE
Revised bids → conditional award recommendation
Revised supplier responses are evaluated against locked targets and reservation boundaries. Recommendations remain planning-only until separately reviewed.
Portfolio → BATNA / target / reservation → concession ladder → revised bid rounds → conditional award branch → reviewed recommendation
No supplier is messaged, counteroffered, awarded, contracted, paid or issued a purchase order. A selected or approved award recommendation is still planning intelligence only.
INTERNAL COMMITMENT PACKAGE
Award recommendation → terms / controls / approval checklist
Build only from an independently approved v3.6 award recommendation. The package compares proposed commitment terms against RFQ scope and locked negotiation reservation positions.
COMMITMENT DRIFT
Catch contract terms that exceed negotiation authority
Reservation breaches, award-share drift and missing scope/protection terms surface as durable risk flags. Internal approval remains separate from qualified legal/commercial review and signature authority.
Reviewed award recommendation → internal commitment terms → drift controls → approval checklist → external-review-ready package
The Foundry does not provide legal advice, send a contract, sign, award business, issue a purchase order, authorize payment, or bind any party. READY_FOR_EXTERNAL_REVIEW means only that the internal decision package is ready for qualified human/legal/commercial review.
ACTUAL DRAFT EVIDENCE
Supplier contract draft → clause / obligation / redline matrix
Paste the actual supplier draft text. A revised document creates a new immutable version; it does not overwrite prior contract evidence.
REDLINE / OBLIGATION LEDGER
Approved deal vs actual draft
Deterministic extraction flags reservation breaches, missing protections and unexpected obligations. Findings are decision-support signals for qualified legal/commercial review—not legal conclusions.
Approved award package → actual draft → extracted clauses/obligations → redline findings → revised draft history → legal-review handoff
The Foundry does not provide legal advice, interpret enforceability, approve legal sufficiency, send redlines, negotiate contract language, sign, award, issue a purchase order, or bind any party. READY_FOR_LEGAL_REVIEW means only that the internal business/negotiation comparison is ready for qualified legal/commercial review.
QUALIFIED REVIEW EVIDENCE
External legal / commercial review → durable review record
Record the reviewer conclusion as supplied evidence. The Foundry does not generate or substitute for the external legal/commercial judgment.
EXECUTION READINESS
Final draft → authority / identity / review controls
Build from the final v3.8 draft after external review evidence begins arriving. Automatic checks derive from evidence; human authority checks cannot be auto-overridden.
External reviews
AUTHORIZED SIGNATURE GATE
Review evidence → MUST checks → readiness package
Legal/commercial review evidence → issue reconciliation → final-document controls → signature-authority checklist → READY_FOR_AUTHORIZED_SIGNATURE
READY_FOR_AUTHORIZED_SIGNATURE is only a governed readiness state. The Foundry does not provide legal advice or legal approval, determine enforceability, sign or accept a contract, communicate acceptance to a supplier, issue a purchase order, authorize payment, or bind any party.
EXECUTED AGREEMENT EVIDENCE
Externally executed agreement → final-draft comparison
Record a contract that was executed outside the Foundry. The text is compared to the v3.9 signature-ready draft before any lifecycle obligations can activate.
CONTRACT REGISTER
Execution drift → review → activation
Exact matches still require an explicit evidence review. Any changed execution text or missing execution metadata creates a blocking drift finding.
ACTIVE OBLIGATIONS
Owners, deliverables, deadlines & recurrence
LIFECYCLE RISK
Upcoming due dates, notices & overdue items
READY_FOR_AUTHORIZED_SIGNATURE → externally executed evidence → execution-drift reconciliation → reviewed evidence → ACTIVE obligation register → lifecycle risk
ACTIVE means the Foundry is tracking supplied contractual evidence internally. It does not mean the Foundry validated enforceability, signed or accepted the agreement, sent a notice, created an invoice or purchase order, communicated with a supplier, or authorized payment.
PERFORMANCE EVIDENCE
Obligation → metric → observed result → reviewed variance
Performance observations remain unreviewed until a human accepts the supplied evidence. A reviewed miss creates an immutable variance finding instead of silently rewriting the obligation.
COMPLIANCE EVIDENCE
Artifact → review → validity window → expiry risk
The Foundry tracks supplied certifications, insurance, audits, licenses and similar evidence. Human review accepts the artifact into the evidence set; it does not certify regulatory sufficiency.
AMENDMENT INTELLIGENCE
External amendment evidence → reviewed obligation changes
Only externally executed amendment evidence can be recorded. Applying an approved amendment updates the internal obligation register; it does not sign, accept, transmit or determine enforceability.
RENEWAL DECISION INTELLIGENCE
Frozen evidence snapshot → recommendation → human decision
Renewal packages freeze the evidence that existed at the time of the recommendation. Human approval selects an internal course of action only; no notice, renewal, termination or negotiation is sent.
Performance / compliance / lifecycle risk
ACTIVE obligation → reviewed performance/compliance evidence → variance/expiry risk → reviewed external amendment → frozen renewal recommendation → human decision
The Foundry can now explain why a supplier appears healthy, risky, amendable or renewal-ready from recorded evidence. It still does not determine legal breach or compliance, negotiate or execute amendments, send notices, renew/terminate contracts, withhold payment, issue invoices/POs, or communicate a decision externally.
SUPPLIER INVOICE EVIDENCE
Invoice → reviewed evidence → reconciliation
An invoice is only supplied evidence until reviewed. Reconciliation checks duplicate numbers, arithmetic, executed-contract rates/Net terms, and delivery/acceptance support.
VALUE-LEAKAGE FINDINGS
Evidence mismatch → quantified review item
Direct duplicate/rate/arithmetic exposure can be quantified. Potential SLA/remedy recovery is surfaced for review but is not invented as a credit without contractual evidence.
PAYMENT READINESS
Reconciled invoice → MUST checks → authorized-process readiness
Invoice evidence → commercial reconciliation → leakage findings → human authority checks → PAYMENT_READY_FOR_AUTHORIZED_PROCESS → STOP
PAYMENT_READY_FOR_AUTHORIZED_PROCESS is an internal readiness classification only. The Foundry does not accept an invoice, post to accounts payable, claim a supplier credit, withhold funds, approve a bank transaction, transmit money, or mark an invoice paid.
EXTERNAL AP EVIDENCE
Payment-ready package → supplied AP record
The Foundry only records evidence that an external AP system created an entry. It never creates the liability itself.
EXTERNAL SETTLEMENT EVIDENCE
Reviewed AP evidence → payment/settlement evidence
Partial and multi-payment evidence is supported. Only human-approved external SETTLED evidence contributes to realized spend.
SETTLEMENT RECONCILIATION
Expected amount ↔ unique approved settlement evidence
Duplicate payment references, AP drift, currency mismatch, overpayment and partial settlement are surfaced without initiating any recovery action.
REALIZED VALUE EVIDENCE
Credit / refund / avoided leakage → reviewed realization
Value is recognized only from separately supplied evidence. An unpaid or outstanding balance is never automatically called savings.
REALIZED COMMERCIAL OUTCOME
Reviewed settled spend + separately evidenced value realization
PAYMENT_READY → external AP evidence → external settlement evidence → reconciliation → SETTLEMENT_EVIDENCE_RECONCILED → spend/value intelligence
The Foundry observes and reconciles externally executed financial events. It does not create AP entries, banking instructions, payment submissions, transfers, reversals, refunds, credits, deductions, supplier claims, or proof beyond the reviewed source evidence.
BUDGET EVIDENCE
Management envelope → reviewed planning evidence
Budget envelopes are supplied management evidence. They do not reserve funds or authorize spend.
FORECAST ASSUMPTIONS
Expected future cash demand → reviewed assumption
Forecast assumptions remain visibly separate from invoices, liabilities and contract commitments.
COMMERCIAL POSITION
Known evidence + probability-weighted planning forecast
IMMUTABLE MANAGEMENT SNAPSHOTS
What the evidence showed at a point in time
Evidence → commercial position → explicit planning assumptions → immutable portfolio snapshot → decision support → STOP
The Foundry can aggregate spend, concentration, value realization and forecast cash requirements. It does not create liabilities, reserve budgets, authorize purchasing, approve invoices, initiate payments, modify contracts, or represent a planning assumption as a contractual/accounting fact.
COUNTERFACTUAL SCENARIO
Frozen baseline → explicit stress assumptions
Scenarios transform a reviewed immutable v4.4 snapshot. They never mutate the snapshot, contracts, suppliers, budgets, accounting records, or payments.
SCENARIO COMPARISON
BASE / UPSIDE / DOWNSIDE / STRESS
Enter scenario IDs built from the same frozen baseline. Comparison review is analytical review only; it does not select a course of action.
RESILIENCE MODEL
Cash stress + supplier exposure without invented facts
Reviewed frozen snapshot → explicit counterfactual assumptions → scenario → comparison → human decision support → STOP
Supplier outage is exposure, not automatic termination or savings. Replacement cost is calculated only when explicitly supplied. Scenario budget shortfall is planning intelligence, not a liability, funding request, budget reservation, sourcing change, contract amendment, AP action, or payment instruction.
MITIGATION CANDIDATE
Costed internal resilience alternatives
Every impact is an explicit planning assumption. The Foundry does not infer supplier availability, replacement capacity, pricing, feasibility or realized savings.
MITIGATION PORTFOLIO
Compare alternatives under explicit constraints
v4.6 compares alternatives independently; it does not silently combine overlapping effects or select a best course.
CONTINGENCY TRIGGER
Condition truth → internal review, never automatic action
DECISION PACKAGE
Readiness for an authorized human decision
CONTINGENCY READINESS
Residual risk + prerequisites + decision history
Reviewed stress → costed alternatives → prerequisites → portfolio → trigger/readiness → READY_FOR_AUTHORIZED_MITIGATION_DECISION → STOP
READY means only that an authorized human has enough governed internal evidence to decide whether to launch a separate action path. The Foundry does not contact suppliers, qualify alternates externally, move awards, amend contracts, approve budgets, buy inventory/capacity, create POs/AP entries, or move money.
GOVERNED EXECUTION HANDOFF
Selected mitigation → frozen downstream handoff
A v4.7 handoff can only trace to an append-only v4.6 SELECT_CANDIDATE / SELECT_PORTFOLIO decision. READY means an authorized human may release the packet through a separate external process; the Foundry never transmits it.
EXTERNAL ACTION EVIDENCE
Record what actually happened outside the Foundry
Action evidence requires an AUTHORIZED_FOR_EXTERNAL_HANDOFF record. Review admits supplied evidence only; it does not imply the Foundry performed or legally validated the action.
OUTCOME EVIDENCE
Observed result metrics, separately reviewed
Outcome evidence requires reviewed COMPLETED external-action evidence. Only supplied metrics are compared; missing values are not invented.
RESILIENCE OUTCOME VERIFICATION
Expected mitigation ↔ actual evidence
Verification freezes expected-vs-observed calibration and durable variance findings. It never silently rewrites the original candidate assumptions or launches corrective action.
v4.7 CLOSED-LOOP RESILIENCE
Decision → handoff → external evidence → actual outcome → calibration
Selected mitigation → READY_FOR_AUTHORIZED_EXTERNAL_HANDOFF → recorded external action evidence → reviewed outcome → RESILIENCE_OUTCOME_VERIFIED → STOP
The Foundry prepares and records. It does not transmit handoffs, contact suppliers, issue RFQs/POs, amend contracts, buy inventory/capacity, authorize budgets, execute operational changes, move money, or silently retrain future assumptions from one observed outcome.
VERIFIED-OUTCOME COHORT
Build an immutable calibration snapshot
Only human-approved RESILIENCE_OUTCOME_VERIFIED evidence is eligible. Incompatible target metrics, units or currencies are excluded rather than silently pooled.
CONTINUITY PLAYBOOKS
Historical ranges → reviewable planning pattern
Playbooks require a human-approved snapshot whose cohort meets its minimum sample threshold. They are reference evidence only, not execution instructions.
v4.8 LEARNING BOUNDARY
Verified outcomes → compatible cohort → frozen calibration → reviewed guidance → continuity playbook
Historical learning may inform a future assumption, but it never becomes that assumption automatically
v4.8 preserves the original v4.6 model and v4.7 observed outcome. Learning snapshots, recommendations and playbooks are new evidence objects. Referencing one from future planning records provenance only; it does not inject values, contact suppliers, change contracts or budgets, buy capacity/inventory, or move money.
SAFE EXECUTIVE
Advance policy-safe work
The coordinator may synchronize dependencies and prepare reversible local artifacts within its capacity budget. Human review, API spend, external dispatch, publishing, verification and consequential acceptance remain explicit gates unless the active policy is changed.
AUTONOMY POLICY
What the coordinator is allowed to do
External dispatch and auto-drafting policy fields are reserved for later executive adapters; v1.1 never auto-runs those action classes.
WORK DAG
Dependency-aware execution steps
ESCALATION INBOX
What needs a human decision
OPT-IN TRIGGERS
Scheduled refresh & signal watch
No background daemon ships enabled. Create a trigger explicitly, then use the one-shot tick command from Task Scheduler/cron if you want unattended cadence.
EXECUTIVE RUNS
Audit trail
A completed executive cycle means the coordinator finished its permitted steps. It does not imply downstream content was published, research was VERIFIED, a build was approved, or an experiment succeeded.
Plan → decompose → run safe steps → retry → escalate → review
Autonomy is a policy-controlled execution mode, not permission to erase approval boundaries. The coordinator can move routine internal work forward while preserving the exact point where money, external action, accepted knowledge, publication or human judgment becomes necessary.
SAFARI / NOTES
Ingest Markdown
Use Safari Copy Links → paste into Apple Notes → Export as Markdown. Drop the resulting .md file here.
WHAT HAPPENS NEXT
Foundry pipeline
- IngestExtract title, raw URL and source group.
- DedupeNormalize tracking parameters while keeping original evidence.
- TriageTopic, source type, credibility, risk and opportunity scores.
- RecoverFetch metadata, redirects and original/surviving source candidates.
- Deep ResearchSearch current evidence, prefer primary sources, capture supporting and counterevidence.
- Human VerifyAgent conclusions stay NEEDS_REVIEW until explicitly accepted.
- Then→NowMark VALIDATED / OVERTAKEN / NOW BUILDABLE / etc.
- ProduceEvidence-backed campaigns, channel assets, build projects and experiments.
- DistributePrepare destination variants, canonical/UTM links and release queue items.
- MeasureCapture observed views, clicks, leads, conversions and revenue.
- TestCompare explicit variants with normalized windows, sample thresholds and practical-effect scoring.
- LearnTurn observations and experiments into cautious next-test recommendations.
- Generalize carefullyAggregate replicated tests into scoped, provenance-backed institutional patterns with maturity and expected-value scores.
- AllocateCompare exploit, explore and challenge opportunities by expected value, information value, upside and relative execution cost.
- OrchestrateMaterialize approved allocations into specialist work items with dependencies, outputs, review gates and portfolio completion feedback.