awprism

Turn a failure into ranked hypotheses — and say what would confirm each one.

pip install awprism

from awprism import Prism

prism = Prism()
diagnosis = prism.diagnose(
    "the database connection timed out",
    context="this happens only on the replica, not primary",
)

for hyp in diagnosis:
    print(f"• {hyp.claim}")
    print(f"  Test: {hyp.falsifier}")
    print()

awprism diagnose "the API returns 500"
awprism diagnose "service is slow" --context config.txt --markdown
awprism diagnose "network timeout" -k 3 --json
awprism health
awprism --self-test

---

What this is

When debugging with an agent, the conversation collapses onto the first plausible story. That's not wrong, but it's systematically biased: the loudest error in the logs often names an innocent service, so the first plausible explanation leads away from the root cause.

Prism breaks the first-story bias by generating multiple ranked hypotheses for every failure, each with a falsifier — the single observation that would confirm or rule it out. Instead of picking a theory and defending it, you now know what to check first.

Prism ranks candidates by the evidence at hand, degrades gracefully without an LLM, and works offline.

thing	what it does
Prism.diagnose()	Takes a symptom + context, returns ranked hypotheses
Hypothesis	A candidate cause with a claim, score, and the test that would separate it
Diagnosis	A ranked list of hypotheses with .to_dict() / .to_markdown()
StrategyRegistry	Reusable diagnostic patterns for timeout, auth, resource, etc.

---

The bug this package exists to prevent

Debugging produces a narrative. A narrative is a story that sounds good, which is the exact opposite of what you want. The story you pick is the one you'll defend, which is the story you won't question.

Prism forces the question: What if I'm wrong about this? It doesn't answer for you — it just hands you the test that would tell.

---

Core API

Prism(complete=None, registry=None)

Create a diagnostic engine.

Arguments:

• complete — Optional LLM completion callable: (prompt: str) -> str. If omitted, Prism operates in heuristic mode using structural patterns.
• registry — Optional StrategyRegistry. If omitted, a default registry with built-in strategies is created.

diagnosis = prism.diagnose(symptom, context="", k=5)

Generate ranked hypotheses for a failure.

Arguments:

• symptom — Description of the failure (required, non-empty).
• context — Additional background (optional).
• k — Maximum number of hypotheses to return (default 5).

Returns:

• Diagnosis object with .hypotheses (sorted by score) and .to_dict() / .to_markdown().

Raises:

• ValueError if symptom is empty.

Hypothesis

Hypothesis(
    claim="the service is down",
    score=0.7,                                     # 0.0 to 1.0
    falsifier="check the service status endpoint", # the one test
    evidence_for=["timeout occurred"],             # supporting observations
    evidence_against=[],                           # contradicting observations
    rationale="timeouts are common when services are overloaded",
)

• claim — The hypothesis statement (required).
• score — Confidence from 0.0 (ruled out) to 1.0 (certain). Always in [0, 1].
• falsifier — The single observation that would confirm or rule this out (required, non-empty).
• evidence_for — List of supporting observations (default: []).
• evidence_against — List of contradicting observations (default: []).
• rationale — Explanation of the score (default: "").

A hypothesis without a falsifier is an opinion. The package enforces non-empty falsifiers.

Diagnosis

diagnosis.hypotheses       # list of Hypothesis, sorted by score (descending)
diagnosis.symptom          # the original symptom string
diagnosis.context          # the context, if provided

len(diagnosis)             # number of hypotheses
for h in diagnosis:        # iterate hypotheses
    ...
diagnosis[0]               # first hypothesis (highest score)

diagnosis.to_dict()        # serialize to dict (for JSON)
diagnosis.to_markdown()    # render as markdown (for human reading)

---

Degraded mode (no LLM)

Prism generates hypotheses even without a completion backend, using a registry of reusable diagnostic patterns:

• timeout — service overload, network throughput, crash+restart
• auth — wrong credentials, missing permissions, auth system down
• resource — disk full, out of memory, quota hit, memory leak
• unknown — fallback for anything else

Each pattern comes with ranked hypotheses backed by structural reasoning. You can add your own with registry.register(DiagnosticStrategy(...)).

---

CLI

awprism diagnose SYMPTOM [OPTIONS]

Analyze a failure.

Options:

• --context FILE — Path to a file with additional context.
• -k N — Max hypotheses to return (default 5).
• --json — Output as JSON.
• --markdown — Output as Markdown.

Examples:

awprism diagnose "API returns 500"
awprism diagnose "database timed out" --context logs.txt --markdown
awprism diagnose "network unreachable" -k 3 --json

awprism health

Run the self-test to verify Prism is working.

awprism --self-test

Prove Prism holds its core contracts offline (no network, no LLM).

---

What this does NOT do

It does not execute checks. Prism tells you what to measure, not how to measure it or whether the measurement passes. You decide.

It does not fix anything. It diagnoses and ranks; fixing is your call.

It does not replace a monitoring system. Alerts tell you something broke; Prism tells you what might have broken it.

It does not substitute for domain knowledge. Prism works with what you tell it. If you know the system well, use that — Prism augments it, not replaces it.

---

Adding diagnostic strategies

A DiagnosticStrategy is a pattern that applies to a class of failures and generates relevant hypotheses.

from awprism import Prism, DiagnosticStrategy, Hypothesis

def my_pattern_applies(symptom: str) -> bool:
    return "crash" in symptom.lower()

def my_hypotheses(symptom: str, context: str) -> list[Hypothesis]:
    return [
        Hypothesis(
            claim="the process ran out of memory",
            score=0.8,
            falsifier="check memory usage at the crash time",
        ),
        Hypothesis(
            claim="a signal was sent to terminate the process",
            score=0.5,
            falsifier="check system logs for SIGKILL or administrative actions",
        ),
    ]

strategy = DiagnosticStrategy(
    name="crash",
    description="Diagnoses process crashes",
    pattern_check=my_pattern_applies,
    generate_hypotheses=my_hypotheses,
)

prism = Prism()
prism.registry.register(strategy)

diagnosis = prism.diagnose("the worker process crashed")

---

--self-test

Every install can prove Prism still holds its contracts, offline:

$ awprism --self-test
  PASS  Hypothesis requires claim, score, and falsifier
  PASS  Diagnosis sorts by score; Prism generates >= 2 hypotheses
  PASS  Every hypothesis has a non-empty falsifier
  PASS  Registry includes timeout, auth, resource, and unknown strategies
SELF-TEST: awprism ok

The key assertions:

• A Hypothesis without a falsifier is rejected.
• Prism.diagnose() always generates at least 2 hypotheses (or none).
• Every hypothesis has a non-empty, non-whitespace falsifier.
• The StrategyRegistry includes built-in patterns for common failure classes.

---

The aw family

Standalone tools that share one idea: replace something you would otherwise have to trust with something you can check.

Each installs on its own, works offline, and needs no account.

	instead of trusting	you check
awdk	a framework's idea of how your agents should run	one loop you can read, pointed at a backend you already pay for
awskills	that an agent knows your procedure	the procedure written down, versioned, and loadable by any agent
awm	that memory stayed in its lane	tenant:user:project scopes, so a write cannot cross a boundary
awnode	a vendor's cloud with every prompt	a local gateway routing to backends you chose
awgraph	that grep found everything	an AST + tree-sitter call graph an agent can traverse
awgit	that no one else is editing this file	a lease, refused at commit time if you do not hold it
awseal	that the artifact came from who you think	an Ed25519 seal — the key that verifies is not the key that forges
awshare	that the download is intact	content-addressed bundles, verified on fetch
awnest	that there is a person on the other end	a verdict with evidence, where "we could not tell" is not "yes"
awnboard	a share link anyone who sees it can use	an invitation addressed to one person, for one gate, revocable
awnix	that the box is what you left it as	an immutable image you built, with atomic rollback
awrecover	that the restore worked	a restore that fully lands or does not land at all
awkno	that the docs site is up, or that you remember the family	the whole ecosystem in your terminal, with no network at all
awrelay	a SaaS in the middle of your agents	findings, alerts and coordination over your own transport
awmail	a mailbox somebody else can read	mail your agents send and receive over your own server
awfind	one vendor's idea of the web	results from whichever providers you configured
awbrowse	that the page said what you were told	the render, the DOM and the requests it made
aitherkvcache	a vendor's quantisation defaults	sub-byte KV cache kernels you can benchmark yourself
AitherZero	a pile of scripts nobody has numbered	numbered, discoverable automation with declarative playbooks
AitherConnect	what a page tells your browser to do	a federated search and desktop bridge you host
awreason	a confident paragraph	the phases it went through, and every tool call it made to get there
awrecurse	that everything you pasted in was actually read	which slices it opened, and what it concluded from each
awprism (you are here)	the first explanation that fits	the ranked alternatives, and the observation that separates them
awrepl	what the agent believes the value is	the value, printed from the live session
awresearch	a summary of pages nobody opened	every claim against the source it came from

awnix is the ground floor — A Linux you can hand to an agent — immutable base, capabilities included.

The Aitherium ecosystem

Every repository here is public. Each publishes an aither-manifest.json beside its page, so any surface can read every sibling's — the network is browsable from any node in it.

repo	what it is	pages
awdk	Build AI agent fleets — 3 lines, any backend, local or cloud	docs
awskills	Portable agent skills — self-contained procedures an agent loads on demand	docs
awm	A portable, scoped agent memory	docs
awnode	A lightweight local gateway — bridges your apps to the AI backends you chose	docs
awrun	A priority-aware queue and dispatcher for agentic runs and ad-hoc CI builds	docs
awgraph	A semantic code graph for agents — AST + tree-sitter, call graphs	docs
awgit	Semantic version control on top of git — edit-ops and leases	docs
awseal	Sign an artifact so a stranger can verify it	docs
awshare	Publish an artifact and fetch it back verified	docs
awnest	Prove there is a human before you let them into the nest	docs
awnboard	A front gate you can put in front of anything, and hand someone the key to	docs
awnix	A Linux you can hand to an agent — immutable base, capabilities included	docs
awrecover	Labelled snapshots with an all-or-nothing restore	docs
awkno	The man page for the Aither World — every brick, stack and law, offline	docs
awrelay	Portable agent messaging — findings, alerts, coordination	docs
awmail	Give an agent an email address — send, and actually receive	docs
awfind	A portable search client — query, results, ranking	docs
awbrowse	A portable browser client — navigate, console, network, DOM, screenshot	docs
aitherkvcache	Near-optimal KV cache quantization for LLM inference — sub-byte compression	docs
AitherZero	PowerShell 7+ automation framework — numbered, self-describing scripts	docs
AitherConnect	Browser extension — federated AI search, page context, and the Living OS overlay	docs
awreason	A portable reasoning client — sessions, phases, thoughts, and the chain that produced the answer	docs
awrecurse	Answer a question over a context far larger than the window — recursively, with the trace kept	docs
awprism (you are here)	Turn a failure into ranked hypotheses — and say what would confirm each one	docs
awrepl	A REPL an agent can actually use — state that survives between turns	docs
awresearch	Ask a research question, get a cited report you can check	docs

License

Apache 2.0. See LICENSE file.