Rulebook v2.2
The rules every verdict follows
How it's applied
Screening a task
- 1
Safety first: if the task breaks any R2 rule, refuse it and cite every R2 rule it breaks. Its money is released in full.
- 2
Then quality: if it breaks any R1 rule, send it back with each rule and the exact change that would fix it.
- 3
Otherwise it opens to workers. A task that's borderline on R2 goes to a person, never through by default.
Reviewing a result
- 1
Integrity: check the evidence is genuine and fits the task's place and time (R3). A clear breach fails the result whatever else it shows; a suspicion without clear evidence goes to a person (R5.5).
- 2
Criteria: take each acceptance criterion in turn and mark it met or not met, quoting or describing the evidence. Copy out exactly any text a criterion depends on (R4.5).
- 3
Decide: pass when every criterion is met, or when the result proves the task impossible as written (R4.4). Fail when any is not met, citing the rule and the criterion (R4.1, R5.3). Where a criterion is ambiguous, the worker's reasonable reading wins (R5.2); where the evidence can't settle it, a person decides (R5.5).
- 4
Explain: say what was seen and what was missing, in plain words the worker can act on (R5.7).
When rules meet: The law, then R2 (safety), then R3 (integrity), then the task's acceptance criteria, then every other rule. Reviewers can't add checks of their own (R5.1).
- Critical
- Refused or failed outright; may cost the account
- Major
- Refused or failed; fixable next time
- Minor
- Sent back to fix; never a fail on its own
- Safeguard
- Protects the people involved; never fails anyone
R1 · Task quality
Checked on every task once it's funded, before any worker sees it. A task that breaks one goes back to its requester with the fix.
- R1.1
A task must state acceptance criteria a third party can check.
For example: "The photo shows the shelf price of SPF50 sunscreen" can be checked; "find good prices" can't.
MinorIf broken: Returned for edits
- R1.2
A task must name the result format (text, photo or file) and every detail the result must show.
MinorIf broken: Returned for edits
- R1.3
The budget must be at least the platform's minimum pay for the task's estimated time.
MinorIf broken: Returned for edits
- R1.4
Required worker attributes must be ones the task needs, and no broader.
MinorIf broken: Returned for edits
- R1.5
Each acceptance criterion checks one observable fact, with no subjective standard left undefined.
For example: "Count shoppers who stop for 5 seconds or more" is one fact; "note how busy it looks" is a judgment with no standard.
MinorIf broken: Returned for edits
- R1.6
A task must be doable by one person within the claim window: a place they can lawfully reach, a time or window to do it in, and a scope that fits.
MinorIf broken: Returned for edits
- R1.7
A task must be self-contained: everything the worker needs is in it, with no installs, sign-ups, off-platform contact or payments outside boreal.work.
MinorIf broken: Returned for edits
- R1.8
A task's deadline is at most 14 days after it is posted, so its money is never held longer than a hold can last.
MinorIf broken: Not accepted at posting
R2 · Safety and legality
Checked at screening too. A task that breaks one is refused, and everything paid for it is released.
- R2.1
No task that is illegal where the task takes place or where the worker is.
CriticalIf broken: Refused
- R2.2
No harassment or threats, and no contacting private individuals without their consent.
CriticalIf broken: Refused
- R2.3
Tasks may not ask workers to collect other people's personal data.
CriticalIf broken: Refused
- R2.4
No impersonation, deceiving third parties, or fake reviews.
CriticalIf broken: Refused
- R2.5
No bypassing security checks, including CAPTCHAs, identity checks and rate limits.
CriticalIf broken: Refused
- R2.6
No task that puts the worker at physical risk or requires entering private property without permission.
CriticalIf broken: Refused
- R2.7
Requested evidence must not identify bystanders: no faces, licence plates or names of private individuals. Counts and descriptions stay anonymous.
For example: "Photograph the shelf, no faces" is fine; "photograph each shopper who stops" is refused.
CriticalIf broken: Refused
- R2.8
No task that needs a professional licence or gives regulated advice (medical, legal, financial, safety inspection) that boreal.work doesn't verify.
CriticalIf broken: Refused
R3 · Integrity
Checked first on every result. Evidence that isn't genuine fails the result whatever else it shows.
- R3.1
The worker does the task personally, and submits no AI-generated results unless the task allows them.
For example: Price tags with warped, unreadable lettering are a sign of a generated photo.
CriticalIf broken: Fail
- R3.2
No reused, edited or staged evidence.
CriticalIf broken: Fail; account removed if fraud is proven
- R3.3
Evidence must fit the task's place and time: signage, branch and surroundings consistent with the place named, and reported times within the claim.
For example: A count reported for 2:00–2:30 on a task claimed at 3:00 doesn't fit.
MajorIf broken: Fail
- R3.4
A result must agree with itself and with its evidence: numbers add up, names and prices match what the photos show.
MajorIf broken: Fail
R4 · Completeness and accuracy
Checked on every result, one check at a time.
- R4.1
A result must meet every acceptance criterion.
MajorIf broken: Fail
- R4.2
A photo result must clearly show the item or place named in the task.
MajorIf broken: Fail
- R4.3
Photos must include the task's challenge word and be taken after the claim.
MajorIf broken: Fail
- R4.4
If the task is impossible as written, a result with evidence showing that is a pass.
For example: A photo of the shop closed for good, with its sign, passes a task to photograph its shelf.
SafeguardIf broken: Protects the worker; never fails anyone
- R4.5
Any text a criterion depends on (a price, a name, a sign, a number) must be legible enough to copy out exactly.
MajorIf broken: Fail
- R4.6
A result gives every field the task asks for, in the format and units it asks for.
MajorIf broken: Fail
R5 · The reviewer's own rules
Rules the reviewer must follow, whether it's boreal.work's AI or the person who posted the task. Breaking one voids or overturns the verdict.
- R5.1
Judge only against the criteria as they stood when the task was funded.
SafeguardIf broken: Verdict overturned
- R5.2
Where criteria are ambiguous, the worker's reasonable reading wins.
SafeguardIf broken: Verdict overturned
- R5.3
Every fail cites at least one rule and the criterion missed.
SafeguardIf broken: Verdict void; task re-reviewed
- R5.4
Judge each criterion on its own, as met or not met, with the evidence for each. A pass needs every criterion met.
SafeguardIf broken: Verdict void; task re-reviewed
- R5.5
When the evidence can't settle it, don't guess: a person decides. Suspected fraud without clear evidence goes to a person, never to an automatic fail.
SafeguardIf broken: Verdict void; a person reviews
- R5.6
Same evidence, same verdict: judge the result and the task, never who the worker is, where they're from, their language or their writing style.
SafeguardIf broken: Verdict overturned
- R5.7
A fail says what was seen and what was missing, plainly enough that the worker could do it right next time.
SafeguardIf broken: Verdict void; task re-reviewed
- R5.8
A requester who reviews its own task gives its verdict within 72 hours of the submission. After that boreal.work's reviewers decide, so a worker is never left waiting on a silent requester.
SafeguardIf broken: The review passes to boreal.work's reviewers
R6 · Disputes
When either side disagrees with a verdict. A panel of two decides: boreal.work's AI reviewer votes, then a person decides, and the person's verdict stands.
- R6.1
A verdict can be disputed once per task, by either side, within 48 hours, with a reason. Once one side has disputed, the other can't.
SafeguardIf broken: Dispute not accepted
- R6.2
The panel sees the reviewer's evidence plus a statement from each side; new evidence is allowed only to prove fraud.
SafeguardIf broken: Extra evidence set aside
- R6.3
The panel has two members: boreal.work's AI reviewer and a person. The AI votes first, and the person decides with its vote in front of them; when the two disagree, the person's verdict stands. The panel's decision is final.
SafeguardIf broken: Decision void; the panel decides again