# Independent Astra 6 data acceptance review

Public edition: internal review identifiers have been removed and the evidence link points to the public source receipt. Findings, calculations and acceptance conditions are unchanged.

**Verdict: ACCEPTED for the frozen, unofficial, partial descriptive comparison.** No P1 or P2 data findings. This accepts the supplied data and export claims within their stated limits; it does not authorize publication or certify final results.

Signed: Astra 6, independent data reviewer, October 21, 2026. Review completed approximately 21:48 UTC. Machine-readable evidence: [public audit evidence](/data/mail-voting/astra-evidence.json).

## Independent work performed

- Downloaded the official 2023 statement-of-vote workbook afresh, twice. The recorded audit download at **2026-10-21T21:44:48.556169+00:00** contains **474,095 bytes**, SHA-256 **6a43655acf5f76176886e6698aedccf362693c3a1b5347536d8ef6c6cac50da3**. Extracted each sheet's `Valid Ballots Cast` column by its header and the `Special Ballot Subtotal` and `Total (All Voting Options)` rows. Independently verified the candidate sums and the two component special-ballot rows on every sheet. No prior extraction or validation receipt was treated as proof. [Official workbook](https://www.elections.ab.ca/uploads/2023-Provincial-General-Election-Statement-of-Vote.xlsx)
- Recomputed all 87 current records directly from the frozen raw feed, selecting question ID 11 and the `Special Ballot` / `VBMHQ` row. Reconciled **both answer columns independently** from non-null poll rows to division totals and then province totals. Verified unique division codes, unique poll-type/name pairs, paired nulls, and one reported special-ballot row per division.
- Freshly fetched the official results HTML: its metadata explicitly maps **questionNumber 10 to questionID 11**. All 87 current division code/name pairs match that live primary source. All 87 cached official 2023 page headings match the corresponding current code/name pairs; all cached hashes match the recorded source hashes. The audited join is by code, without an independent boundary-geometry certification. [Official 2026 results](https://results.elections.ab.ca/8400)
- Independently recomputed shares, changes, and competition ranks using exact rational fractions before rounding. Compared every audited field for every division in JSON, CSV, and the HTML's embedded data: **zero discrepancies**. The HTML data equals the JSON data exactly.
- Verified all 15 ordered names, bar values, table counts, historical ranks, and changes on each of the two PDF pages against the independently derived records. Verified all ten report table rows. Inspected both PNGs visually, including the count PNG at native 2880 x 2160 resolution; chart labels, ordering, zero baselines and caveats agree. Read the HTML display/sorting definitions; browser interaction QA was outside this independent data review.
- Hashed the delivered outputs and confirmed those hashes remained unchanged during the audit. The exported raw snapshot has the same verified hash as the working snapshot.

## Verified results and precise source pins

| Quantity | Independent result |
|---|---:|
| 2026 Q10 valid special/mail votes | 514,970 |
| 2026 Q10 valid votes, all methods | 2,022,151 |
| 2026 counted mail-in share | 25.466446373193694% |
| 2023 valid special ballots | 51,351 |
| 2023 valid votes, all methods | 1,764,915 |
| 2023 special-ballot share | 2.9095452188915614% |
| Difference | +22.556901154302132 percentage points |
| Divisions with a positive observed share difference | 87 / 87 |

The province shares are ratios of summed valid ballots, not averages of division percentages. No significance test is needed to establish this arithmetic difference in the observed counts; these are not estimates from a random sample.

- **Highest share: ED46 Edmonton-Whitemud.** Current raw Q10 special row: 9,642 Remain + 567 Referendum-process = **10,209**; Q10 division valid total **23,870**; share **42.76916631755341%**, share rank **1**, count rank **7**. Historical workbook sheet `46`: **J26 = 609**, **J27 = 21,187**, giving **2.8744041157313447%**, share rank **42**, count rank **40**. Difference **+39.89476220182207 pp**.
- **Largest count: ED47 Airdrie-Cochrane.** Current Q10 special valid votes **12,954**, count rank **1**, share approximately **31.13%**, share rank **22**. Historical workbook sheet `47`: **K24 = 1,118**, **K25 = 30,072**; historical count rank **9**.
- Additional workbook pins: ED40 `J29 = 710`, `J30 = 19,199`; ED25 `J27 = 1,273`, `J28 = 23,079`.
- No share ties occur in either year and no current count ties occur. Historical count ties are **263** (ED10, ED19), **322** (ED20, ED37), and **590** (ED41, ED62); the delivered competition ranks handle these correctly.

## Snapshot and completeness findings

- Audited current snapshot: fetched **2026-10-21T21:31:32.218783+00:00**, upstream updated **2026-10-21T21:31:07.7426913+00:00**, SHA-256 **a28e8ffcaf1185baa91b1267656e992410ebe3eb422552b6d27b89d78c83665d**, **542,787 bytes**. The displayed **3:31 PM MT** is consistent with America/Edmonton on this date.
- **5,344 / 5,575** Q10 location rows contain counts; **231** have paired nulls: **186 ORD, 9 ADV, 36 MOB**. All **87** special-ballot rows have counts. **55 / 87** divisions have no null Q10 rows. The feed also contains **63 explicit zero/zero rows**; these are correctly distinguished from nulls. Neither non-null values nor zero values establish final completion, and the delivered caveats say that reported rows can change.
- The separate province widget says **5,329 / 5,575** reporting and contains all-question valid-vote totals. It is not a Q10 denominator or Q10-completion measure. The analysis correctly excludes its all-question totals. The official page also warns that its reporting indicator does not certify every question is complete. [Official reporting definition](https://results.elections.ab.ca/8400)
- A fresh check at **2026-10-21T21:44:53.318813+00:00** returned upstream update **21:44:21.0159989+00:00** and Q10 valid total **2,022,240**, an increase of **89** since the frozen snapshot. This is expected live-feed movement, not a discrepancy in the delivered frozen analysis. The source was not replaced.

## Interpretation and publication boundary

The partial snapshot is usable as a clearly dated provisional description. The PNG/PDF include the retrieval time, explicit count-in-progress language, missing-row daggers, valid-vote denominators, and changed-rule warning. The HTML/report carry the additional completeness and interpretation limits. **No unfixable publication issue was found.** Do not remove those caveats or relabel this as a final ranking.

The official comparison release says 2023 special ballots were restricted in eligibility and were not required to arrive by mail. The 2026 request release confirms eligibility for all eligible electors and a mandatory mail return. Therefore the current numerator can be described as mail-in valid Q10 votes, while the historical numerator must remain **all special-ballot valid votes**, not a postal-only count. The supplied labels preserve that distinction. [Official comparability warning](https://www.elections.ab.ca/resources/media/news-releases/referendum-advance-voting-and-special-ballot-voter-turnout/) · [Official 2026 eligibility and return rules](https://www.elections.ab.ca/resources/media/news-releases/2026-referendum-special-ballot-requests/)

The data supports a large **observed difference in shares under different rules and events**. It does not isolate a change in propensity under constant rules, identify individual mail voters from area demographics, or establish a causal effect on the referendum outcome. The supplied report and explorer correctly limit those claims. A final-results publication requires a fresh source and a new arithmetic/ranking check; that is not required merely to show this explicitly frozen provisional snapshot.

Remaining review limits: no independent boundary-geometry harmonization, no inspection of underlying election administration or individual ballots, no browser/deployment or security audit, and no publication action. All raw data and outputs remained unchanged; only this review and its evidence JSON were written.