Methodology — Glassbox

The substance behind the surface.

Every score in this product, every comparable record, every recommendation is produced under five rules. None are secret; all are demonstrable; each can be verified against the source data on demand.

Section 01

Calibrated language

Output describes what the data shows. It does not assign verdicts. It does not aggregate intent. It does not direct the reader.

Forbidden patterns

Hover any line for a description of what the pattern catches.

/\b(should have|ought to have|was supposed to)\b/i

/\b(proves?|proven|definitely|certainly|clearly shows?)\b/i

/\b(failed to|refused to|deliberately|intentionally|knowingly)\b/i

/\b(coverups?|cover[- ]ups?|schemes?|scams?)\b/i

/\b(allegedly|reportedly|sources say|many believe|it is widely known)\b/i

/\bbecause of\b[^.]{0,80}\b(grant|contract|funding|donation|lobbying)\b/i

/\b(caused|led to|resulted in)\b[^.]{0,80}\b(grant|contract|funding)\b/i

/\bin (?:exchange|return) for\b/i

/\b(this raises serious questions|warrants investigation|warrants further investigation|the public deserves answers|officials should explain)\b/i

Rejected → calibrated

✗ The government failed to deliver the program.
✓ The dataset does not show recorded outcomes for this program.
✗ Evidence of fraud in the contribution.
✓ Pattern consistent with sole-source amendment growth across the agreement chain.
✗ This grant should have stated its outcomes.
✓ Comparable filings in this program typically state expected outcomes.
✗ The donations were in exchange for grants.
✓ Lobbying registrations were filed 47 days before grant award dates.
✗ This raises serious questions about oversight.
✓ The dataset shows 31 amendments to the underlying contribution agreement.

Try it yourself

Type any sentence below. PYTH-GOV runs the calibration sweep against it in real time, the same way every output is gated before it ships.

Type any sentence — the calibration sweep runs live as you type. No submit button, no API call. Try one of the seeded examples below to see how PYTH-GOV catches uncalibrated language (causal claims, hedging absent, directives, “clearly shows”, and so on).

PYTH-GOV idle — type a sentence(Status updates with every keystroke — try an example below.)

Seeded examples — click to load

Accepted phrasings

· The dataset shows federal contributions totalling $134M between fiscal 2017 and fiscal 2024.
· Pattern consistent with sole-source amendment growth across the agreement chain.
· Records indicate the contribution agreement was terminated on July 3, 2020.
· Comparable filings typically state the program stream and the recipient's expected outcomes.
· Public records do not contain a final disposition statement for this contribution.

Section 02

Citation discipline

Every prose claim ships with at least one source pointer. No quote may exceed 15 words. No source may be quoted directly more than once.

Tier 1 — Auditor General reports. Highest authority.
Tier 2 — Departmental Results Reports, Hansard, parliamentary committee evidence.
Tier 3 — Major-outlet news with editorial review (Globe, La Presse, CBC investigative).
Tier 4 — Other news. Requires multiple corroborating sources before any claim ships.

Section 03

The Proof token

Every output of an accountability AI must itself be accountable. The token records what evaluated the input, what model produced the analysis, what gates passed, and what audit was issued.

Section 04

Structural correspondence with the federal AIA

The audit token is structurally equivalent to a Treasury Board Algorithmic Impact Assessment, applied at every output rather than once at deployment.

Tier	Audit token (Glassbox)	Federal AIA
Tier 1 (Input)	Filters applied + landmines respected (F-3, A-13, A-10)	Project Description + scope of automated decisions
Tier 2 (Context)	Embedding model + synthesis prompt version + temperature	Risk Assessment — algorithm class, training data, error modes
Tier 3 (Output)	Citation count, max-quote length, calibration verdict	Mitigation Measures — peer review, monitoring, recourse
Tier 4 (Audit)	Operator agent, token hash, previousTokenHash chain	Approval & Review — sign-off, periodic re-assessment

Section 05

Data landmines

The corpus carries documented defects. Every canonical query guards against them explicitly. The Proof token's Tier 1 records which guards applied.

F-1
ref_number collisions across distinct recipients (~41K rows)
Guard: Partition by (ref_number, COALESCE(bn, legal_name, _id::text))
F-3
agreement_value is cumulative — naive SUM triple-counts amendments
Guard: WITH agreement_current AS (DISTINCT ON … ORDER BY amendment_number DESC)
A-13
AB exact duplicates + 951 reversal pairs
Guard: Dedupe on (ministry, business_unit_name, recipient, program, amount, payment_date); pair-collapse opposite-sign matches
A-10
AB recipient-NULL roll-up rows (~$25B in FY24+25)
Guard: Filter recipient IS NOT NULL; disclose the omitted aggregate
C-7
CRA name history is mostly missing (1.4% of BNs)
Guard: Treat cra.cra_identification.legal_name as current-state, not historical

Section 06

The five agents

Each agent operates within bounded perception. The strata are not decorative — they are how the architecture is scoped.

PYTH-LEAD · Task agent (descends along provenance strands)
Orchestrator, schedule, scope cuts
PYTH-DATA · S4 source-linked
Canonical SQL, landmine guards, hybrid retrieval, embedding job
PYTH-SYN · S3 semantic
Similarity scoring, awardee patterns, calibrated recommendation text
PYTH-FE · S2 procedural
Every UI component, every page, every motion. Reads DESIGN-SYSTEM.md as authoritative.
PYTH-GOV · S3 semantic
Calibrated-language gate, Proof token completeness, route-existence + visual smoke gates. Veto authority.

The methodology.