TakehomeGet API keys

Docs · Findings

Findings index

Who this is for: Maintainers tracing published PDOC disagreements back to named methodology records.

When you finish: You know which findings are closed, which stay open, and where to reproduce each one.

Findings are numbered write-ups linked from CONFORMANCE.md methodology sections. They reference T4127 (Payroll Deductions Formulas) readings and live PDOC behaviour. They are not excuses to drop cent deltas from the agreement rate unless the methodology section says otherwise.

Superseded filename

002-alberta-half-cent-float.md is retired. IEEE-float as a general PDOC rule is falsified, and the behaviour is not Alberta-specific. The live record is 002-pdoc-midpoint-direction.md.

M-001: Published whole-dollar K

Known: CRA publishes bracket constant K rounded to whole dollars. Annual tax can jump by up to one dollar at a threshold. PDOC shows the same discontinuity. Tests bound period withholding jumps.

Unknown: Nothing open for M-001. The jump sizes for all thirteen 2026 bracket tables live in crates/takehome-core/tests/vectors/bracket_discontinuity_2026.json.

Reproduce: Run cargo test -p takehome-core bracket_discontinuity and read CONFORMANCE § M-001.

M-002: K2 maximum in reaching period

Known: T4127 Chapter 3 asks for maximum base CPP contribution in the pay period when year-to-date first reaches the annual maximum. PDOC uses min(base_max, max(P×C×ratio, D×ratio)) without forcing base_max in that period. Inversion on the vector yields CPP credit base 3494.12 (PDOC) vs 3519.45 (literal reading). Period federal tax differs by thirteen cents under default pdoc_observed.

Unknown: Whether CRA will align PDOC with the glossary sentence. Customers can select k2_method: t4127_literal for the document reading.

Reproduce: Five-minute PDOC steps in finding 001. Engine: same JSON in crates/takehome-core/tests/vectors/ PDOC Ontario corpus.

M-003 and M-004: closed by finding 003 (2026-10-22)

Known: Three named rules close 149 of the 164 July one-cent disagreements: K2 and K2P as two separately rounded credits, Ontario V1 as two separately rounded surtax lines (default arm), and PDOC's CPP line matching binary64, inferred from its outputs (M-004, rounding_compat: pdoc). Overall rate 8838/9002 became 8905/9002 (default) and 8987/9002 (pdoc). The rules were chosen by inspecting the full corpus; 15 cases remain unexplained.

Unknown: 15 period tax lines stay one cent apart, each within a tenth of a cent of a half-cent. They count in both rates.

Reproduce: cargo run -p takehome-conformance -- --check and the tests named in finding 003.

M-003 (history): PDOC midpoint direction

Known: Finished July capturable queue: 8,818 match, 164 one-cent disagreements (all count in the rate), 780 step-2 non-advances (named, not in the rate). Splits: 82 CPP-only (engine higher), provincial and federal mixes include PDOC higher and lower. Not all 164 are exact half-cent lines in engine arithmetic. Until finding 003, rounding_compat: pdoc returned RoundingCompatPdocNotImplemented; since 2026-10-22 it is implemented (PDOC's CPP line under binary64, ADR-007) and that error is no longer returned.

Unknown: The condition that selects PDOC down versus up on a given line. Neighbour gross pairs (164/164 have a ±1¢ neighbour; 159 neighbours match) suggest an upstream tail, not a single midpoint rule.

Reproduce: Pick any id from the M-003 rows in CONFORMANCE.md disagreements table (for example NL-P26-1718.39-F0-P0). Enter the grid form in live PDOC for the July edition. Compare provincial and CPP period lines to /v1/deductions with the same as_of, province, pay_period, gross, and federal and provincial claim codes (TD1 claim code indices). See finding 002 for the NL type specimen table.

Last reviewed: 2026-10-23
Engine: 0.1.0

Source: docs/findings/README.md in the Takehome repository, rendered as written.