Period snapshots
When a reporting period closes, its figures are written once and never written again. This is where those closed periods live: the numbers exactly as they stood on the day they were reported, and the only figures anyone is allowed to quote outside this console.
Three things hold for every period on this page
- A closed period is never recalculated. Data that arrives after the close date — a late endline, a survey answered a week too slowly, a corrected count — does not enter a closed snapshot. It is reported separately, alongside the original, never folded into it.
- There is no edit, re-open or void control. Not on this page, not behind a confirmation, not for the system administrator. A poor result stays on the record and is still published.
- Suppression is frozen too. The minimum cohort size that applied at close is the one that stays applied. A group that was too small to publish then is still not published now, even though we now know how the story ended.
Register of reporting periods
| Period | Dates | Cohort | Instrument | Closed by | State | |
|---|---|---|---|---|---|---|
| Q2 2026 the funded three-month pilot |
1 Apr – 30 Jun 2026 endline window runs to 31 Jul |
64 | Assessment v1.2 Survey S-02 v2.0 |
— | Open · figures move nightly | Open live metrics |
| Q1 2026 PS-2026-Q1 |
1 Jan – 31 Mar 2026 | 12 | Assessment v1.1 no survey in field |
Priya Raman 3 Apr 2026, 09:12 |
Closed · locked | Open snapshot |
Q1 2026 was an internal trial run before the funded pilot began: nine staff and contractors and three volunteers, on an earlier version of the knowledge assessment. Those twelve people are counted in no Q2 figure anywhere. The trial is on the register because it was reported to the funding body as the first progress report, and anything reported has to stay readable in the state it was reported in.
PS-2026-Q1 · internal trial, 1 January – 31 March 2026
How a cohort of twelve is reported
Twelve people clears the minimum cohort size, so this period can be published at all. It does not clear enough to publish percentages.
- Figures are given as counts out of 12. A percentage on a base of twelve reads as though it generalises, and it does not.
- No breakdown by age, by residence or by anything else. Every one of those cells is below ten and each would be a step towards naming somebody.
- Where a measure has no usable data, the snapshot says so and prints the counts that do exist. It never prints an estimate and never draws an empty chart.
The same rules were applied by the system at close. There is no control on this page that reveals a suppressed figure.
Locked metric set
read-only · closed 3 Apr 2026| Measure | What it is | Figure at close | State |
|---|---|---|---|
| METRIC-01 assessment before and after |
Measured, single arm | Not enough data. 12 baseline sittings, 0 endline sittings, 0 matched pairs. The two-week endline instrument only went into field on 1 April, so no eligible endline opened inside this earlier period. | Reported qualitatively |
| METRIC-02 return at day 1, 7 and 30 |
Measured | 10 of 12 · 8 of 12 · 5 of 12 no percentage published on a base of twelve |
Published as counts |
| METRIC-03 self-report survey |
Self-reported proxy | Not collected. Survey S-02 v2.0 went into field with the pilot on 1 April 2026. Nothing was asked, so nothing is reported. | Out of scope for the period |
| METRIC-04 continued after a risk warning |
Measured | 152 of 214 warnings shown event count, from twelve people |
Published as counts |
| METRIC-05 learning, resetting, or leaving |
Measured | 38 decision points · 19 back to learning · 12 reset · 7 left without choosing | Published as counts |
| METRIC-06 support links opened |
Measured reach, not outcome | 21 opens, by 6 people breakdown by placement withheld — every placement is below ten |
Partly suppressed |
| ACT residents share of cohort |
Self-declared at sign-up | 11 of 12 postcode given by the person, not verified |
Published as counts |
| Risk stratification never bet · betting · cutting down |
Self-declared at T0 | Not collected. The T0 screen that asks this went live with the pilot on 1 April 2026. | Out of scope for the period |
Two of the eight rows carry no number. That is the honest state of an internal trial, and it is what was sent to the funding body. Filling those rows in later — now that the pilot has run and the instruments are in field — would be describing Q1 with data Q1 never had.
Analyst note, frozen with the figures
P. Raman · 3 Apr 2026“Q1 was a build trial, not a measurement of anything. Twelve people, most of them working on the product, cannot tell us whether it helps a stranger. The one figure worth keeping is that 152 of 214 risk warnings were read and then ignored, because it sets the expectation for the pilot: the warning is not going to stop most people the first time, and any design that assumes it does is designing for a person who does not exist. METRIC-01 has no result. Do not report a knowledge gain from this period; there is not one to report.”
The note is part of the snapshot and is locked with it. A later interpretation is written as a new note against a later period, never over the top of this one.
Records that arrived after the period closed
Three endline sittings arrived between 8 and 12 April 2026, from people who had sat their baseline in the Q1 trial on assessment v1.1. Q1 closed on 3 April.
- They were not added to PS-2026-Q1. That snapshot still reads 0 endline sittings and 0 matched pairs, and it will read that permanently.
- They were not pooled into Q2 either. Q2 is measured on assessment v1.2. A v1.1 baseline and a v1.1 endline make a valid pair, but not a pair that belongs in a v1.2 period — pooling them would mean comparing scores taken on two different rulers.
- They are held as an addendum of three. Three is below the minimum cohort size, so the addendum carries the count and nothing else: no scores, no average, no direction of travel.
This is the whole reason closed periods exist. If late data could flow backwards into a published figure, no figure would ever be quotable, and the incentive to wait for better-looking numbers before reporting would sit there permanently.
Q1 2026 compared with the open period
Q1 2026 is closed and locked. Q2 2026 is still open, so its column is provisional: it moves at 06:00 every night and nothing in it may be quoted outside this console until the period closes on 31 July 2026.
| Measure | Q1 2026 · locked | Q2 2026 · provisional | Comparison |
|---|---|---|---|
| Cohort | 12 · internal trial | 64 · public pilot | Different people recruited in different ways. This alone rules out reading any row below as a trend. |
| METRIC-01 assessment before and after |
no matched pairs | 41 pairs · +3.8 of 20 | Not comparable Q1 sat assessment v1.1, Q2 sits v1.2. Two different instruments produce two different scales, and Q1 has no endline to compare in the first place. No difference is calculated here, and none may be calculated elsewhere. |
| METRIC-02 still returning at day 7 |
8 of 12 | 29 of 64 | Shown, not differenced Both counts are real. The gap between them is mostly the difference between colleagues and strangers, and retention is not a success measure for this product in any case. |
| METRIC-03 self-reported behaviour |
not collected | 38 of 64 opted in | No baseline Q2 is the first period with any self-report data at all, so it is a first baseline and not a change. |
| METRIC-04 continued after a risk warning |
152 of 214 warnings | 68% of 1,842 warnings | Shown, not differenced The trial figure is not restated as a percentage — a rate resting on twelve people invites exactly the comparison it cannot support. The two counts sit side by side and are not subtracted. |
| METRIC-05 chose learning after a loss |
19 of 38 decisions | 132 of 244 decisions | Shown, not differenced Around half in each, on decision counts rather than people. Worth watching across three periods, worth claiming after none. |
| METRIC-06 support links opened |
21 opens · 6 people | 214 opens · 37 people | Shown, not differenced A count of opens measures reach. Neither period knows whether anybody was helped, and neither ever will from this console. |
| ACT residents | 11 of 12 | 51 of 64 | The slice the funding body asks for in every period. Both are self-declared postcodes and neither is verified. |
Nothing on this page produces a trend line. Two periods, one of them a twelve-person build trial on a different instrument, is not a time series. The console will not draw one, and the pilot report states plainly that the first comparable pair of periods is Q2 2026 against Q3 2026, both on assessment v1.2 — and that even then, a difference between two quarters with no control group is a description, not a finding.
What a snapshot holds, and what it never holds
Held
- Period start and period end, and the moment the snapshot was generated.
- Cohort size at close, and the version of every instrument in force during the period.
- The aggregate metric set, with each figure carrying the sample it was computed on.
- Every suppression that applied, and the reason for it.
- The analyst's note, and the name of whoever closed the period.
Assembled from the de-identified behaviour store and the aggregate tables, never from live personal records.
Never held
- Any name, account code, device identifier or postcode at unit level.
- Any group smaller than the minimum cohort size, in the console or in an export.
- Any figure that was estimated, extrapolated, back-filled or rounded to look tidier.
- Any control to edit, re-open, void or delete a closed period — including for a system administrator, and including where the result is unflattering.
- Any route back into gameplay. Nothing measured here personalises odds, rewards or notifications for anyone.
A figure discovered to be wrong is corrected by publishing a correction beside the original, so the original and the correction are both on the record.
Where these figures go next
A closed snapshot is the only source the funding body's reports are built from. Nothing in a report is typed in by hand, which is what makes a report checkable against the period it came from.