PUBLIC AUTHORITY MUST BE TRACEABLE
Consequential power needs a lawful, bounded mandate and an accountable owner.
Project Shadow 1.0.1 contains no Myth package. Generic Myth v0.2.0 and Full-Canon Myth v0.3.5 are separate optional companions; both default off, neither is required by R1, and neither can authorize action or change an R1 result. Civic QA remains self-service and unofficial; publication transfers no authority and is not operational deployment, efficacy, safety, certification, a legal-compliance finding, permission to act, or a global green.
One civic lifecycle / six distinct public doors
Project Shadowinstruments + testsPBHPpausesCivic QAinspectsARMrepairsThe RecordremembersALMSIVIpreserves optional originQA-06Method + integrity
Civic QA is an opinionated public-accountability method. It aims for evidence discipline, reproducibility, challengeability, and correction—not a false performance of view-from-nowhere neutrality.
These commitments shape what Civic QA notices, how gates are ordered, and what may count as closure. They do not predetermine a factual finding or excuse weak evidence.
Consequential power needs a lawful, bounded mandate and an accountable owner.
Inspection begins with people least able to refuse, wait, appeal, exit, or recover.
Unknown, inference, hypothesis, and verified fact do not collapse into one confident voice.
Notice, explanation, privacy, challenge, remedy, and recovery must work in practice.
Implementation, training, publication, or a high score cannot close a finding by themselves.
Material counterevidence reopens the work; prior editions and changed reasoning remain visible.
Claim labels reveal the decision weight a statement has earned. They also show where another reviewer can challenge or strengthen the work.
Directly present in the inspected record or event.
Confirmed against an independent attributable source.
A reasoned conclusion with confidence and alternatives visible.
A plausible explanation that still needs testing.
A gap that could change the finding or action.
Publish the scope, selection logic, source-search method, exclusions, reviewers, AI assistance, scoring rationale, unresolved disagreement, corrections, and what evidence would change the finding.
State where AI helped search, summarize, compare, draft, or score. Verify consequential claims against attributable sources. Do not treat model agreement as independent verification.
Explain why this subject, period, population, source set, and standard were chosen—and what remains outside the inspection boundary.
Preserve materially different reviewer judgments and the reasoning behind them. Consensus is not evidence when everyone shares the same frame or source lineage.
Never silently overwrite. Preserve the prior state, trigger, date, changed reasoning, and downstream findings that need to reopen.
These limits increase the duty to show scope and preserve challenge; they are not a reason to abandon inspection.
A structured score does not make the reviewer neutral or the conclusion objective.
Source volume does not establish source quality, coverage, independence, or relevance.
Legal compliance does not establish fairness, effectiveness, legitimacy, or absence of harm.
A model card, audit, benchmark, or vendor certification does not establish safe live operation.
Public data may reproduce reporting gaps, enforcement intensity, administrative choices, and historical exclusions.
A finding can be correct in one scope and wrong when generalized to another population, period, or decision path.
A challenge can dispute scope, source coverage, evidence state, rationale, scoring, hard-stop handling, burden, correction, or disclosure. It does not need to agree with this method's values, but it must identify what should change and why.
Name the exact page, self-service report ID or reviewed edition, claim, and evidence cutoff you are challenging.
Provide an attributable source with publisher, date, stable link or reference, and exact locator; separate fact from interpretation.
State the proposed correction, why it changes the scope, rationale, gate, finding, or remedy, and any affected downstream claim.
Send the packet to projectshadowqa@protonmail.com without personal, confidential, medical, or legally protected information. There is no response-time SLA.
For your own self-service draft, preserve the prior export, revise locally, export a new edition, and describe what changed. Only governed review can change an official Civic QA edition or mark.