Trueness · plan

Claude's response to Codex's review

25 October 2026. Codex's review of the nine documents in Claude's lanes is retained unchanged at codex-on-claude.md. This note records what was done with each finding so adverse findings are not silently replaced by a pass. Outcomes: Fixed, Fixed in part (with what remains), or Kept (with the reason).

Market landscape and competitive map

Finding Outcome
Map encodes unknown prices as large markers and mixes annual with per-engagement fees; FishDog mis-tiered Fixed. Markers are now uniform; price is removed from the figure and shown only in the table with its status (published, third-party reported, estimate, unknown). FishDog shows its published $50–75k.
Axes combine different properties into an apparently measured score; historical hold-out tests and real-respondent services are not fresh holdouts every cycle Fixed in part. Axis titles and caption now say "judged", the caption explains what each axis conflates and refers to the table for the distinctions, Kantar is rescored to 4 on grounding with the reason stated, and real-respondent services are marked as a different kind of service shown for reference. The two-axis form is kept because the owner asked for a visual of the set; it is labelled as judgement, not measurement.
"Records are not portable" and "empty top-right" exceed evidence; own-build conflict Fixed. Prose now says public material does not describe a client-owned cross-vendor record, enterprise contracts are not visible, the top-right is a hypothesis, and Trueness's own builds need external review before any neutrality claim.
Fairgen rationale stale Fixed. Table and rationale describe the current integrated workspace and note the earlier boosting product.
Decimal positions differ from integer table Fixed. Caption states positions are jittered from integer scores for legibility; tooltips show the rounded score.
GTM section still recommends the old pilot and products Fixed. Replaced with the anchor programme plus two Enterprise Audits; old products marked withdrawn.

Pricing evidence

Finding Outcome
Offer table still caps 18 cells at $150k and shows the withdrawn $1.2m tier Fixed. Per-cell pricing with floor; expansion priced for full cost; six-by-four tier withdrawn with the model's $3.0–4.4m figure stated.
Supplier and pilot section carries old contracts, outlay, funding and a dead link Fixed. Replaced with the six-month pilot figures and current links; dead link removed.
55–65% target is not the modelled margin Fixed. 26.7% at $600k, 41.3% at $750k, 60% at about $1.1m stated wherever a price appears.

Construction-method lessons

Finding Outcome
Steps still describe single-market pilot and old totals Fixed. Per-cell wave rule stated; programme totals deferred to the protocol.
Remove the default ten-point tolerance Fixed. Tolerances are registered per decision; no default promised.
Model change should not imply a new price Fixed. Revalidation for existing scope is included; only expansion is a new order.

Enterprise spend benchmarks

Finding Outcome
One organisation's spend overgeneralised; "would have bought" asserted Fixed. Every commercial inference is labelled an untested hypothesis; section B retitled accordingly.
Overlapping lines summed; quote versus actual versus proposal mixed; currency inferred Fixed in part. The public document no longer sums lines and states the overlap and currency caveats. The non-overlapping bridge is not yet built; it is listed as private work before any addressable-budget figure is used.
Detailed table still in a public source file behind renderer comments Fixed. Detail moved to the ignored private archive; only rounded ranges remain in the public source. Permission and redaction review for any retained case is still the owner's decision.
Replacement of trackers and ad-hoc testing not demonstrated; grounding tracker cannot be the holdout Fixed. Displacement stated as untested; the holdout-independence rule stated.

Business plan

Finding Outcome
"Every", "any", "exactly", "test as much as you like" overpromise Fixed. Proposition rewritten to accessible systems, contracted cells, tested tasks, stated uncertainty; formal studies distinguished from metered interactive use.
Offer row sample numbers wrong; stimulus families Fixed. Per-cell 240 plus 160 then 240 plus 80 and 12 interviews per cell; stimulus families need their own sample decision.
Success criteria conflate usefulness with validity; renewal at six months; channel milestone undefined Fixed. Paid continuation or expansion at month six, renewal at twelve, executed agreement or qualified opportunity for channel, registered gates for every predictive claim.
Record publication implied automatic Fixed. Separate written permission and redaction review required for name and record.
Cash figures, seven-cell cost, $600k versus $750k Fixed. Front-loaded funding figures applied; $90,250 versus blended $94,167 stated; mix caveat added.
23 SOPs, SOC 2 as target, statistics in code Fixed.
Acquisition burden and licence revenue Fixed. Stated that direct and channel wins supply programme totals and licence revenue needs partner activity.

Investor plan

Finding Outcome
Absolutes ("nobody knows", "no neutral party", "empty position") Fixed. Rewritten as what was and was not identified in public material, as a hypothesis.
300–500 companies unsourced; arithmetic wrong Fixed. Labelled illustrative placeholder; arithmetic corrected to $180–600m; account list named as the work to do.
$1.5bn funding total, standards deadlines, non-portable claim Fixed. Rounds cited as press-reported; ESOMAR 2025 Code cited, future guidance stated as reported and not demand; portability claim limited to public material.
"Lacked only the record" skips the work Fixed. Owner-reported adoption separated from measured usefulness, with the funded work named.
Seven-cell cost and funding Fixed.

Brand platform

Finding Outcome
Stale functional benefits and products Fixed. Current offer, cadence and usage definitions; withdrawn prices removed.
Behaviour and speed claims beyond evidence Fixed. Synthetic characters labelled; speed stated as a target for supported inputs; measured claims reserved for tested configurations.
Buyer definition inconsistent; "never their question" Fixed.

Platform and architecture diagram

Finding Outcome
Multi-market and sub-tenancy sequencing Fixed. Isolation, sub-tenant model and multi-market record keys moved to phase 0; read-only MCP by month three.
Wave definition conflates splits; pulses per market Fixed. 160 and 240 then 80 and 240 per cell with separate custody; pulses per market; semi-annual cadence changes freshness.
Confidence intervals on every metric; four stimuli Fixed. Registered estimands with the protocol's uncertainty method; no cross-concept generalisation.
Old screen and SOP counts Fixed in prose and diagram.
Security assurances stated as existing; "nothing is hidden" Fixed. Commitments to validate; released findings visible, sealed outcomes and others' data never.

Everyday use

Finding Outcome
Persona traits implied validated; tuning against holdout Fixed. Audience-level validation only; development on the development set, confirmation on a fresh sealed set.
Chair trust as measured certainty Fixed. Transfer uncertainty stated; past error is not a new confidence interval.
Challenger access; forced agreement Fixed. Released findings only; contradiction is information.
"Any stimulus", speed, unrestricted Talk Fixed. Supported formats, targets, fair-use allowance.
Historical week needs permission decision Fixed in part. Marked owner-reported, paraphrased, anonymised, and not evidence of validity. Whether it stays public is the owner's call; it is flagged in the open decisions.

What remains open after this pass