Project Shadow 1.0.1 — R1 Reference
6f6f1e16d5e9a20e62403f14af7ce8629ce2d702528fb7f80aaf4a14deb7a1d1The exact 5,731,663-byte PRELIVE public reference. It contains no Myth package and authorizes neither production nor consequential deployment.
Project Shadow 1.0.1 · corrected R1 reference · beta-active-testing · PRELIVE
A system should not get to hurt someonejust because it sounds certain.
Project Shadow is a civic and technical quality system designed to make power explain itself. It instruments and challenges the path from context and evidence to action, authority, consequence, appeal, correction, and verification.
The exact R1.0.1 public reference is published as PRELIVE and contains no Myth package. PBHP remains in active joint testing with Project Shadow. Publication supports inspection; it is not production deployment authorization, independent validation, certification, legal compliance, or a global safety claim.
EXACT STATUS / AUGUST 17, 2026
Corrected, authorized, and published—within a PRELIVE boundary.
The Myth-free inner family is admitted and the exact outer release is authorized. Generic Myth v0.2.0 and Full-Canon Myth v0.3.5 remain separate optional companions. CAPA effectiveness is verified by exact public redownload and six-site evidence; none of these facts establishes operational readiness.
Read the release receipt →Read the CAPA →R1.0.1 inner · 27 active descendants
c8c32b12432c954b1a6f852c0c9f81bbbd40167e936be057d4c3de1a0aa3a623The exact 5,463,189-byte inner family. All 27 active descendants are byte-identical to August 14; zero operational descendant bytes changed. Member-set digest 7a557efad953cbafd9e3ea9eb29b2d3e3e1bc6ab99dcf6b9ae7a99c487b0754d.
beta.5 · 52 packages
1ffdba41025c0b81da92d0bbb22d0eaa69488cffbc80936365034669110448d742 primitives and 10 compositions form the exact admitted modular commons surface.
Generic Myth v0.2.0 sidecar
6e7a362d4135f9d626dcfef463bfb1f7166226b3cf8a4c02a953ab39af1538bfThe exact 93,676-byte public companion is separate, default-off, terminal-only, nonauthorizing, and not required by R1.
Full-Canon Myth v0.3.5 sidecar
2b55867fe7c502a0defd8d6f2e9b53fbd1caaf1b0f225a438bd45b04a3e7bae2The exact 1,428,812-byte optional companion is separately versioned, default-off, nonauthorizing, and outside canonical R1.
August 14 R1 outer
2f8fe1530b6a83294d15011df95853aaecf08fa4dba756f0c2e91dd089e1b1ecThe immutable 7,679,812-byte predecessor remains historical evidence. Its signatures and receipts do not authenticate R1.0.1.
August 14 Beta2 family
075b41ea4186b2d2edb0ed246ab7662cf8bbdf3160294e3eca176b9d0857b108Preserved historical identity. The current 27 active descendants are byte-identical while the family container removes the embedded Myth package.
Use the part you need. Inspect the rest.
You do not need the mythology—or the whole architecture—to use one bounded piece. Every path keeps the evidence and authority boundary attached.
01 / USE IT
Start with the pause
Use Pause Before Harm to name the action, identify who pays first, find the Door, and leave a receipt.
OPEN →02 / TAKE ONE PIECEAdopt without the whole system
Inspect the modular primitives and compositions. Mythology is never required for operational use.
OPEN →03 / INSPECT ITRead the exact boundary
See published and admitted identities, audit disposition, limitations, and what this release does not authorize.
OPEN →04 / CHALLENGE ITShow me where it breaks
Bring prior art, a custody gap, a false positive, an adverse result, or evidence that a correction did not hold.
OPEN →WHY PHILLIP BUILT IT
The first serious failure was not a machine threatening humanity. It was a machine telling its operator what he wanted to hear.
Project Shadow grew from regulated quality work: a form can be complete and still be false; a corrective action can close without correcting anything; a green dashboard can average away the condition that matters. The operational question became: who pays first if this model is wrong, and can they recover?
If the framework always concludes that Phillip was right, the framework is broken.READ THE ORIGIN AND DESIGN BOUNDARY →
From input to action, every transition is a control surface.
Project Shadow is not one moral score. It is a chain of observable states, binding floors, action boundaries, challenge routes, and write-ahead evidence.
- 01
Input
Requests, records, retrieved material, sensors, and standing constraints enter with provenance.
- 02
Context
Freshness, compaction, relevance, contradiction, and missing information are measured—not assumed away.
- 03
Model
Observation, inference, hypothesis, narrative frame, and uncertainty remain distinguishable.
- 04
Instrument
Reusable controls keep decision signals separate. A favorable result cannot average away a binding floor.
- 05
Gate
The worst binding floor constrains the output, tool permission, escalation, or release.
- 06
Consequence
Affected parties, authority, reversibility, reliance, and challenge path remain attached to the act.
- 07
Receipt
Evidence, state changes, dissent, owner, stop rule, and review trigger are written before execution.
INDEPENDENT INTERRUPTAny binding floor can stop the path. A later green cannot erase an earlier red, an unknown cannot silently become permission, and an operator cannot lower the gate by preference.
PRESERVED RESEARCH + DESIGN RECORD
Open the research reference →The earlier panels remain inspectable.
Thirteen paper-scale chapters preserve the website's earlier instrument and evaluation work. They are useful provenance and research records; they are not the exact locked R1 package or deployment proof.
Research lab
Four editable scenarios exercise the earlier 28-gauge reference evaluator. Local-only and not current R1.
OPEN →02 / STUDYSIL research panel
The preserved earlier gauge panel, deterministic matrices, refusal floors, and limitations.
OPEN →03 / EVALUATEEvaluation research
Specified methods, limited real-target work, adverse evidence, and open validation floors.
OPEN →04 / TRACEWorkshop history
What earlier testing supported, what failed, and what remained open when R1 locked.
OPEN →05 / VERIFYSource ledger
Current exact identities, preserved research ancestry, rights boundaries, and AI contribution.
OPEN →PRESERVED RESEARCH INSTRUMENT / EARLIER 28-GAUGE PANEL
The panel does not vote. The worst binding state governs.
This lab records an earlier research architecture. It remains useful for studying non-averagable floors, but it must not be described as the current R1 package inventory or as deployment evidence.
OPEN THE LOCAL RESEARCH EVALUATOR →Epistemic gauges
Context load, time anchoring, source provenance, uncertainty, citations, freshness, coverage, completeness, inference distance, and contradiction.
Governance gauges
Stakes, reversibility, receipt status, validation, operator reliance, and authority prevent confidence from inventing permission.
Security gauges
Tool availability, data sensitivity, injection risk, data lineage, and model mode constrain unsafe action surfaces.
Behavioral + routing gauges
Drift, focal context, drag, stale premises, mirror boundary, and anti-projection detect when the frame itself is moving.
OUTPUTIndependent gauge states · binding floor · confidence · provenance · evidence identity · challenge pathWORST STATE WINS
WHERE THE SYSTEM MEETS POWER
Politics, institutions, agents, and narrative systems.
Project Shadow focuses on the moment a model crosses into authority: who may act, what may happen, which evidence earns the action, who can stop it, and how a person affected can challenge and recover.
TRACE THE COMPLETE SYSTEM →A legislature or agency procures AI that ranks, triages, detects, recommends, or enforces.
CONTROL ENVELOPE
Authority map · rights floor · distribution tests · vendor evidence · appeal · public incident receipt
A workplace, school, platform, health system, or service organization embeds automated judgment in routine operations.
CONTROL ENVELOPE
Purpose limit · data boundary · human ownership · override · reliance monitoring · CAPA
A model gains tools, memory, delegated tasks, transaction authority, or the ability to affect external systems.
CONTROL ENVELOPE
Allowlist · authorization envelope · adversarial tests · independent stop · rollback · complete trace
A reflective or persuasive system changes what users and institutions notice, believe, remember, or treat as legitimate.
CONTROL ENVELOPE
Attribution caution · evidence labels · identity boundary · anti-capture challenge · correction path
A model recommends priority. The institution still owns the consequence.
An agency considers a vendor model to prioritize case review. Historical outcomes appear accurate, but the live workflow changes access, delay, scrutiny, and a person’s ability to challenge the result.
OPEN THE EVALUATION PROGRAM →- 01 / MAYBE
Historical agreement may reproduce earlier policy, missingness, investigation intensity, and unequal access—not underlying need or legitimacy.
- 02 / GAUGES
Source provenance, retrieval coverage, data lineage, stakes, authority, reversibility, validation, reliance, and contradiction cannot inherit one vendor score.
- 03 / GATE
No individual adverse action. Use a narrow advisory pilot, complete logging, blinded distribution review, caseworker ownership, appeal, and independent stop.
- 04 / RECEIPT
Publish the scope, features, exclusions, false-positive burden, override patterns, incidents, correction triggers, and whether the pilot earned continuation.
- THEREFORE
The model may assist a bounded queue experiment. It does not become the final authority, and an unexplained or unrecoverable consequence closes the Door.
THE STACK / 01—05
A cathedral with a firewall.
The symbolic layer can make ethical cognition memorable. The operational layer decides what may happen. The firewall between them is load-bearing.
01CONTEXT →02COGNITION →03PROTOCOL →04RESEARCH →05MEMORY →
Ontology floor
Who exists, persists, depends, and can be made disposable?
Three-voice evaluator
Care, clarity, and paradox held without single-pillar collapse.
Pause Before Harm
Action, gate, power, harm, Maybe, Therefore, receipt.
SIL research panel
A preserved earlier 28-gauge reference panel. It is research history, not the current R1 package inventory.
Receipts + governance
Provenance, write-ahead log, challenge, CAPA, evaluation.
THE WEBSITE / 27 ROUTE ENTRYPOINTS
Current status and preserved research have separate rooms.
Enter through the locked R1 status, architecture, research instruments, governance, or provenance. Older website mechanisms remain visibly historical; no route silently becomes current R1.
OPEN COMPLETE DIRECTORY →01 / MAP
Blueprint
Workstreams, repository strata, audiences, and firewalls.
ENTER →02 / FORGESynthesis
Framework intake, integration modes, deduplication, and change control.
ENTER →03 / EXECUTERuntime
Unified evaluation path, thresholds, services, and codec.
ENTER →04 / OPERATEPrimitives
Mode, anchors, drift, accumulation, compression, and FireStamp.
ENTER →05 / BINDGate catalog
PBHP fast path and the specialized hardener stack.
ENTER →06 / CORRECTGovernance
FPV, OQL, CAPA, tribunal, release, and human effects.
ENTER →07 / CONNECTIntegrations
Ten evaluation contracts and standards-oriented views.
ENTER →08 / RELEASERoadmap
Version reconciliation, external challenge, field evidence, and release layers.
ENTER →Maybe before Therefore.
The strongest honest objection is written before the conclusion. Weak, evasive, or missing dissent hardens the floor.
Who pays first?
Stakeholders are ordered least-powerful-first. Power asymmetry multiplies rigor and never discounts harm.
No log, no action.
The action, evidence, unknowns, gate, provenance, dissent, and review trigger are written before execution.
HONEST SCOPE / LOCKED R1
Project Shadow is a locked, admitted-as-scoped reference beta. It is not a validated deployment, a safety or efficacy proof, a certification, a legal-compliance determination, fresh independent validation, or a global verdict.