← Part 26

AQ-INDEPENDENT-REVIEW

The seven-stage independent technical review: what a reviewer receives and may challenge at IR-1 to IR-7, how reviewers are selected and which conflicts exclude them, and the seven obligations the programme owes a reviewer whose finding it rejects, including published dissent and a contested marker on the claim.

Charter artifact · gated by Part 26 · revision 0, draft for internal review

AQ-INDEPENDENT-REVIEW — Technical Reviewer Protocol

PURPOSE, AND THE DIFFERENCE FROM AQ-IOP-001

An observer under AQ-IOP-001 answers whether the written procedure is what actually happened. A reviewer under this protocol answers a different and harder question: does the evidence support the claim, and is the design sound in the first place. A run can be flawlessly observed and still support nothing, because the claim was mis-stated, the assumption set was silently doing the work, or the value that carries the argument was MODELLED all along. Review is the mechanism that catches that. Like AQ-IOP-001, it is an L9 instrument reading L6 records.

STATUS AT ISSUE

Nothing has been built. No Akasha-Q element has reached Lab validation, so every review commissioned at issue can only address material at Concept, Theory or Simulation level: specifications, decision logic, assumption sets and models. Commissioning a review does not raise a claim's evidence level. A favourable review does not either — only measurement does that.

WHAT A REVIEWER MAY CHALLENGE

Nothing in the programme's own documentation is out of reach. A reviewer may challenge:

  • the wording of the claim, and the evidence level asserted against it;
  • the DC-1 tag of any value, and specifically any value whose tag is stronger than the method that produced it;
  • the assumption set, and whether its stated invalidating conditions are observable in practice;
  • whether DC-2 holds: whether any missing input can reach a decision path as a benign default rather than as undetermined;
  • whether DC-3 abstention is a real, reachable output of the implementation or only a described one;
  • whether DC-4 monotonic degradation is enforced — that trust falls on a single observation and rises only through the full recovery sequence out of RECOVERING;
  • whether DC-5 reconstruction from the Evidence Fabric alone is achievable, tested by attempting it;
  • completeness of the failure-mode set F-1 to F-14 and the conflict set CF-1 to CF-7, and the resolution rule when two conflicts are live at once;
  • placement of the decision boundaries B-1 to B-5 and of the trust boundaries TB-1 to TB-5 in AQ-NODE-01, in particular whether TB-4 is genuinely hardware-gated or gated in software that claims to be hardware;
  • whether Q4 can ever widen what Q1 to Q3 permitted (it may not: Q1 to Q3 are hard constraints and Q4 can only reduce), and whether any code path or configuration allows it;
  • whether any L8 output can reach a decision path other than as advisory;
  • the sufficiency of an Evidence Export Package the claim relies on.

A reviewer may also challenge the question they were asked. If the review question is framed so that the interesting failure falls outside it, saying so is itself a finding and is recorded as one.

REVIEWER SELECTION CRITERIA

  • Named layers, not the programme. A reviewer is commissioned for specific layers (L0 to L9) matched to their domain. No one is commissioned to review “Akasha-Q” as a whole.
  • Willingness to read primary evidence — raw output, source, bitstream description or specification — not only a summary. A reviewer who will read only the summary is commissioned as an editorial reviewer and published as such, and their sign-off never supports an evidence-level claim.
  • No reporting line into the programme, and no expectation of one.
  • Named, or anonymous by their choice with their institution class stated and identity held under the restricted-access record. Anonymity is the reviewer's right, never the programme's convenience.
  • A fixed fee agreed before materials are issued, never contingent on conclusion, schedule or the programme's satisfaction. Fee terms are published with the review.
  • No contractual bar on filing a finding. The restricted-access record may delay publication and may redact detail; it may not suppress the fact that a finding exists.

CONFLICT-OF-INTEREST EXCLUSIONS

RelationshipWindowEffect
COI-1Employment, contract, internship or advisory role with EduRankAI, Akasha-Q or Visvambhara — current or within 24 monthsExcluded
COI-2Equity, option, revenue share or convertible interest in EduRankAI or any programme vehicle — any timeExcluded
COI-3Candidate for, or in live discussion about, a role or contract with the programmeExcluded for the duration and for 12 months after
COI-4Supplier, or employee of a supplier, of hardware, firmware or services within the review scope — current or within 24 monthsExcluded
COI-5Co-authorship or a shared funded project with a named author of the document under review — within 36 monthsExcluded unless jointly assigned with a second unconnected reviewer, and the connection published
COI-6Family or household relationship with a named author or with programme leadershipExcluded
COI-7Holds a competing published claim that the finding would directly advantage or damageDeclarable and published; permitted only with joint assignment
COI-8Reviewed an earlier version of the same documentNot a conflict. Continuity is preferred; their earlier findings are re-issued to them at IR-2
COI-9Paid a fee for a previous reviewNot a conflict, provided no fee was ever contingent on a conclusion

Declarations are made at IR-1, before materials are issued, and are published with the review. A conflict discovered after IR-2 is published as discovered; it does not retroactively erase the findings, which are marked with the conflict and re-reviewed under IR-7.

THE SEVEN STAGES

StageReviewer receivesReviewer producesWindow
IR-1Draft review question set; the claim and asserted evidence level; scope and layer assignment; fee terms; the COI declaration formSigned COI declaration; the question set, accepted or amended. A programme refusal of an amendment is recorded and published with the reviewbefore materials
IR-2The full review pack: the document under review; assumption set; DC-1 tag map; primary evidence (raw data and any EEP relied on); design artefacts in scope (specification, decision-logic source, firmware or bitstream description); prior reviews with their unresolved findings; the open-findings register. An L8 summary may never be issued in place of primary evidence.Confirmation of pack completeness, or a gap list10 working days to raise gaps
IR-3Written answers to every challenge, or an explicit “unanswered” with reasonWritten challenges, itemisedprogramme answers within 10 working days
IR-4A recorded working session with the named authors; the right to require a live walkthrough of a decision path, a re-derivation from raw data, or a demonstration that an abstention path is reachableSession record; any further evidence requestafter IR-3 closes
IR-5A factual-accuracy pass on the draft only: the programme may correct a fact, and may not soften a judgementDraft findings, each typed BLOCKING, MATERIAL, ADVISORY or EDITORIAL, each attached to a specific claim, section or evidence item5 working days for the accuracy pass
IR-6The programme's written response to each finding: accept, accept-with-modification, or reject with a typed reasonOptional right of reply15 working days
IR-7The published record as it will appearDissent, if any, published verbatim; register entries confirmedat publication

A reviewer may bring one named specialist into IR-4 under the same COI rules and the same restricted-access terms. Specialists are named in the published review.

Figure 1 - Finding lifecycle, IR-5 to the open-findings register

  reviewer finding (BLOCKING | MATERIAL | ADVISORY | EDITORIAL)
                |
                v
  IR-6 programme response, in writing, 15 working days
                |
   +------------+------------+--------------------+
   |            |            |                    |
 accept   accept-with-     reject          no response in window
   |      modification       |                    |
   v            v            v                    v
 change      change      reason typed:        recorded as
 made,       made,       fact | interp |      UNANSWERED
 finding     finding     scope | priority    (treated as reject,
 closed      closed          |               without a reason)
                             v
            reviewer dissent published verbatim,
            beside the claim, in every document citing it
                             |
                             v
     open-findings register: named L9 owner + re-review trigger
     (trigger = the next evidence level this claim attempts,
      or the next run that relies on the disputed value)

WHAT THE PROGRAMME OWES A REVIEWER WHOSE FINDING IT REJECTS

  • A written, typed reason within 15 working days, stating whether the disagreement is about fact, interpretation, scope or priority. “Noted” is not a reason and is recorded as no response.
  • Publication of the finding in full and verbatim, beside the rejection, in every document where the affected claim appears — not only in the review document.
  • A right of reply: one further written statement, published unedited, with no programme text beneath it.
  • An open-findings register entry with a named L9 owner and a named re-review trigger. Rejected findings are closed by re-review, never by elapsed time.
  • A contested marker on the claim, carried beside its evidence level until the trigger fires and the re-review completes, and travelling with the claim into every derived document.
  • No change to access, fee or future eligibility. If a reviewer who filed a rejected finding is not re-commissioned, the reason is given to them in writing and published in the register.
  • For a rejected BLOCKING finding: a claim freeze. The programme may continue designing, building and running — the finding restricts nothing about what may be built — but the affected claim may not be raised to a higher evidence level until the re-review completes.
HOW A FINDING IS NOT RESOLVED

A finding is not resolved by replacing the reviewer, by re-scoping the review question so the finding falls outside it, by letting the commission lapse, or by publishing a summary that omits it. If the programme wants a second opinion on a rejected finding, it commissions an additional reviewer under IR-1 with the original finding included in the IR-2 pack. It does not re-run the same review with a different person and publish the second result alone.

PUBLISHED DISSENT

A reviewer's dissent is published verbatim. The only permitted alteration is redaction required by the restricted-access record shared with the Visvambhara programme; each redaction is marked, counted and given a reason, and the reviewer is shown the redacted text before publication. If the programme cannot publish a dissent at all, it publishes that a dissent exists, who holds it, and the restriction under which it is withheld. A reviewer may withdraw their name from a published review. They cannot be asked to withdraw the finding.