How a figure gets into the data.

Version 2026-09-08. Public. Each change to this page is dated in the changelog at the end.

1. What the data is

Factual information about Australian superannuation investment options, super platform products and listed or unlisted investment products: fees and fee structures, asset allocations, historical returns and the arithmetic to price a product at a stated balance. It is not a rating, a ranking or a recommendation. Every figure carries its source document and date, its completeness and its provenance, so a reader can check it against the issuer's own disclosure.

2. Where figures come from

The issuer's Product Disclosure Statement or its fees guide is the legal source for a fee. We read the current document from the issuer's site, never a cached or archived copy, because funds reprice at 1 July and caches go stale. Regulator statistics (APRA) supply returns and the base catalogue of options.

3. How a fee is recorded

• Units. Percentages are decimal fractions (0.0019 means 0.19% a year). Dollar amounts are dollars a year.
• Raw text beside every number. The issuer's own wording for the fee (including caps, rebates and any statement about costs met from reserves) is stored with the parsed number.
• Tiered fees are explicit tables. A balance-tiered administration fee is a list of bands: the balance where each band starts, its rate, and any dollar cap on that band. The API prices any balance from the table; the headline rate is only the first band.
• Blank means unknown. If a figure cannot be verified from a current document it is left blank with a note saying why. It is never estimated and never treated as zero.
• What counts as the administration fee. The issuer-labelled asset-based administration fee. A separately named levy that is deducted from the member's account (a trustee levy, an operational risk reserve charge) is part of the fee: folded into the rate when uncapped, or recorded as its own band with the cap's remaining headroom when capped. A cost is excluded only when the issuer's own words say it is not deducted from the member's account, and those words are stored on the record.
• Fixed dollar fees, rebates, discounts and waivers are recorded in words and flagged; they are never netted into a percentage. Employer-negotiated or licensee-negotiated rates are not public; the published standard rate is recorded and the record says so.
• Published totals. Where an issuer or the regulator publishes a total cost figure it is served as published and labelled as_published. A cost built from components is a separate calculation that lists which components it lacks.

4. Worked examples

Pricing a tiered admin fee at $250,000. A wrap platform discloses 0.20% on the first $250,000, 0.12% on the next $250,000, 0.10% on the next $500,000 and nil above $1 million. Stored as four bands starting at $0, $250,000, $500,000 and $1,000,000. At $250,000 the fee is $500; at $750,000 it is $500 + $300 + $250 = $1,050; at $1.5 million it is $1,050 + $500 = $1,550, and the balance above $1 million costs nothing.

A capped levy as a headroom band. An account fee of 0.15% to $750,000 (nil above) sits alongside an operational risk levy of 0.023% capped at $300 a year. Recorded as 0.173% on the first $750,000 (both charges), then 0.023% above $750,000 capped at $127.50, which is the levy cap less the $172.50 that has accrued by $750,000. Both charges and the $300 cap are reproduced exactly at any balance.

A rebate quoted, not netted. A fund charges 0.29% and rebates administration fees above an indexed balance after year end. The rate stays 0.29%; the rebate, its threshold and its timing are recorded in words and flagged. A consumer can apply it; the headline never pretends it already has.

A reserve-funded exclusion. A fund discloses 0.19% plus "0.07% deducted from Fund reserves (not from your account)". The 0.07% is excluded from the rate because the issuer says it is not charged to the member, and that sentence is stored on the record.

A member-charged levy folded in. A platform discloses a 0.2583% platform fee and a 0.0250% operational risk charge, both deducted from the account. The rate is 0.2833%, matching the issuer's own headline for administration fees and costs.

A restatement. Three options were stored under the wrong product name (a pension label on accumulation options). The correction was recorded with a dated note in the corrections file, applied by script, logged, and every affected record carries the note.

5. What the API tells you about each figure

• eligible: the product is open, has a source document and has a management fee. Not a recommendation. (recommendable is the deprecated alias of the same flag.)
• fee_components and fee_data_complete: each fee group (investment, transaction, performance, admin percentage, admin fixed) is present, an explicit zero, or missing; complete means nothing is missing and a source exists.
• components_missing, known_cost_dollars and total_ongoing_dollars on cost calculations: the total is null whenever a component is missing; the known-cost subtotal always says what it includes.
• allocation_provenance: how the asset allocation was obtained (issuer disclosed, regulator derived, inferred from the product name, mixed, unknown).
• fee_verification_age_days and stale: how old the fee evidence is; stale after 365 days or past a recorded re-verification date. Every fee record is re-verified within 60 days of 1 July.
• total_cost_basis: as_published on issuer or regulator totals.
• Expected-return figures state their method, the source and period of every input series and a version, and are marked internal-use-only until their inputs are licensed for redistribution.

6. How the data is checked before it is served

1. A research agent reads the current issuer documents and writes a dated batch whose header states the conventions applied.
2. Code maps the batch onto records by product, phase and menu; every record must map exactly once.
3. An independent reviewer from a different AI provider re-reads the live documents for a deterministic sample of the batch (every fund; at least three records per fund or the square root of the fund's record count; every blank record; every changed figure) and grades the batch PASS or FAIL. A report that skips a sampled record is rejected.
4. A named data owner reviews a sheet of the highest-impact records and approves the change on its final revision.
5. Validators check units, duplicates, negative fees, tier-table shape and totals before anything is deployed; the live service is read back after deployment.

The record-level discrepancy rate found by the reviewer, with the number of records inspected, is published from the review log.

7. Corrections policy

A correction is any change to a figure after it has been served. Corrections are recorded with a dated reason and evidence, applied by script so they are reproducible, logged in the department log, and visible on the affected record's flags. Anyone can request a correction by citing the issuer document; the figure changes only when the document supports the change. The department log is the public restatement trail.

8. Limits

• Negotiated employer and licensee rates are not public; published standard rates are served and flagged.
• Returns are quarterly, published about three months after quarter end, on a net-investment-return basis before administration fees; pension-phase returns exist only where the regulator publishes them.
• Asset allocations for many regulator-sourced options are inferred from category data and flagged as such.
• Issuer verification of records and an external audit of a sample have not yet been done; both are planned and their absence is stated in the control framework.

9. Rates and calculators

The Knowledge API serves a second kind of fact: the legislated rates and thresholds Advium's own calculators run on, by financial year, and the documented method of each calculator.

• Where a rate comes from. Every figure in a rate set is entered from a primary source (the ATO, Services Australia, APRA, a state revenue office) with the citation written beside it in the source file. Since September 2026 the same citations also live in a machine-readable sources table: one entry per figure with the citation, the page URL where one is known, the date the figure was checked, its effective date, its unit and jurisdiction. A test fails the build if a verified figure has no source entry or a source entry points at a figure that does not exist.
• Verified or indicative. Each domain of a rate set (income tax, super, Age Pension, stamp duty and so on) is either verified against its source or marked indicative. Indicative domains are placeholders; the API hides their values unless a caller asks for them and labels them when it does. Modelling defaults (an assumed return, inflation or fee) have no published source and are never served.
• How currency is kept. A scheduled job fetches each cited page, finds the figure next to its label, and reports MATCH, CHANGED or UNVERIFIABLE for every sourced figure, comparing only against pages that describe the same financial year. Its report is committed under docs/domain/rate-check/ with a summary line, and a CHANGED result opens a change for a person to verify against the source; the job never edits a rate itself. A year the rate set does not hold is refused by the API, not approximated from another year.
• How a calculator is documented. Each calculator's purpose, method, inputs (with defaults and ranges) and outputs are read from its own source code and schema by a generator, so the published methodology cannot drift from what the code does. The same generator produces the methodology library PDFs and the JSON the API serves. Running a calculator through the API uses the platform's engine on the verified rate set for the requested year and returns the result with the compliance block Advium prints on its own documents.

Changelog

• 2026-09-09 v1.1: rates and calculators section; the sources table, the automated rate check and the verified or indicative rule for the Knowledge API.
• 2026-09-08 v1: first public version; eligibility and completeness flags, staleness, component groups, headroom-band treatment of capped levies, corrections policy.